Alibaba Group Holding Limited (NYSE: BABA) — A Chosen Trough Meets an Unchosen Blacklist

Date: August 7, 2026 · Coverage: Update (prior article: June 7, 2026) Fiscal year referenced: FY2026 (ended March 31, 2026) · Next reported quarter: Q1 FY2027, due August 20, 2026 Reporting: U.S. GAAP in Renminbi (RMB) · ADS ratio: 1 ADS = 8 ordinary shares · Structure: Cayman holding company controlling PRC operating entities via VIE/contractual arrangements Price reference: US$126.81/ADS (close, 2026-08-06) · Market cap: ~US$304B · 52-week range: US$94.81 – US$187.62

⚡ Claude’s Take

This block is the author’s own subjective opinion and is general information only. It is not investment advice, and it is not a recommendation to buy or sell any security. Everything that follows it (Executive Summary through Source Appendix) is deliberately position-free: that analysis takes no position, sets no price target, and carries no recommendation — the opinion is confined to this block.

Verdict: HOLD / do not add here — a downgrade from the June 7 “accumulate-on-weakness, speculative BUY.” At ~US$127 this is a stand-pat, not an add. The accumulate zone moves down to roughly the low-US$100s and below; I would want US$95–105 again before putting new money to work. Directional fair-value zone trimmed to ~US$145–180/ADS (from ~US$150–190) — not because the business got worse, but because the jurisdictional discount that belongs on this claim got wider. The EPV/asset floor of roughly US$95–125/ADS is unchanged and is the reason this is a HOLD rather than an AVOID.

Since the last report the business news has been modestly good and the jurisdiction news has been materially bad, and the stock is up 4.8% — which is the wrong combination. On the good side: Beijing formalized the food-delivery ceasefire into a draft rulebook on June 17, so the profit-destroying subsidy war now has a regulatory end-date rather than a hoped-for one; Alibaba’s Qwen won the Apple Intelligence design slot in China on July 15, the single best competitive-position datapoint in years; Qwen3.8-Max shipped on August 3 at 2.4 trillion parameters; the eight-year DOJ pharmaceutical probe closed for US$600M and a non-prosecution agreement; and — the finding I care most about — management restarted buybacks on June 22 and bought harder as the price fell, scaling roughly 4× at the 52-week low and stopping when the stock recovered. That is precisely the price-sensitive behaviour the June article said was missing, worth ~US$384M and disclosed in Alibaba’s own Hong Kong filings rather than asserted on a call.

On the bad side, one day after the last article published, the tail stopped being theoretical. The Pentagon added Alibaba to the Section 1260H “Chinese Military Companies” list on June 8; Alibaba sued the Department of Defense in San Jose on June 24, calling the designation arbitrary and capricious; Anthropic wrote to the Senate Banking Committee that same day alleging the largest known adversarial-distillation campaign against its models, run through ~25,000 fraudulent accounts; and a securities class action followed in August. To be precise, because the distinction is the whole argument: 1260H is a Defense Department identification list with procurement consequences and no securities prohibition. The list that would stop U.S. persons transacting is Treasury’s NS-CMIC list, and Alibaba is not on it. Nothing binding has happened. But the conditional probability of the binding version has plainly risen, it is now driven by an adversarial process with a docket rather than by ambient risk, and it deserves a wider discount than I applied in June. The framing, and this is evidence not assertion: falling knife that bounced. The factor model says Alibaba carries a negative momentum loading (−0.44 in the sparse model, −0.50 in the full one) and a positive value loading (+0.53), that its dominant explanatory variable is the China country factor at +1.87, and that a decade of holding it has produced a 4.95% annualized return with an 80% maximum drawdown and a Sharpe of 0.07. The July rally was +27.4% and it has stalled just under the 200-day EMA at US$129. This is not momentum; it is a violent bounce inside an intact downtrend, in a name where the marginal buyer is trading China, not Alibaba.

Conviction: medium-low (down from medium). Flips bullish if the August 20 print shows China E-commerce adjusted EBITA inflecting up materially — call it a return toward a US$25B+ annualized run-rate — and management resumes repurchases at these prices rather than only at the lows. Flips bearish if Treasury adds Alibaba to the NS-CMIC list, which converts a headline risk into a structural inability for U.S. persons to transact and would justify a permanently different multiple. Tag: “The cheapness got taken back; the tail didn’t.”

📈 Stock Price Action — Five-Year Event Map

Alibaba has spent five years describing a wide, violent range and going nowhere: US$189 in August 2021, a US$59 low in October 2022, a US$188 recovery high in October 2025, and US$126.81 today — roughly a third below both the 2021 and the 2025 peaks. The last twelve months contain the whole argument in miniature: a 52-week high of US$187.62 on 2025-10-02, a 52-week low of US$94.81 on 2026-06-26, and a +33.8% bounce off that low that has carried price to within 1.7% of — but still below — its 200-day EMA of US$128.98. Price moves below are Fact (AZI daily series); the attributed drivers are Interpretation.

Cycle narrative. (1) The 2021–22 collapse was regulatory and macro, not operational, and it set the permanent discount this security has traded at ever since. (2) The 2022–24 plateau is where the business case deteriorated — PDD’s hard-discount model and Douyin’s content commerce took share while Alibaba’s conglomerate sprawl consumed capital. (3) The 2024–25 doubling was the AI re-rating plus the most aggressive buyback in the company’s history, ~US$12B a year retiring ~5% of the float annually. (4) The retracement from October 2025 tracked the arrival of the FY2026 numbers: group adjusted EBITA −56%, operating income −64%, and the first negative free-cash-flow year in Alibaba’s public life. (5) The June 2026 leg down is the one that matters for this update — it is almost entirely geopolitical, beginning the session after the Pentagon published the 1260H list on June 8 (the ADS fell 3.9% over two sessions to US$115.38 by June 10) and accelerating into the June 24 Anthropic letter, bottoming at US$94.81 on June 26. (6) Overlapping that bottom, Alibaba’s own Hong Kong Next Day Disclosure Returns show the company buying its ADS on the NYSE from June 22, roughly 1.0M ordinary shares a day, then roughly 4.2M a day across June 29 – July 2 at the lows, tapering to 72,800 on July 9 as the stock gapped up. (7) July’s +27.4% is the best month in the five-year series and was driven by a cluster of risk-removal and AI-validation events, of which the Apple Intelligence approval is the most durable. (8) The Qwen3.8-Max launch extended the move into August; the stock now sits just beneath its 200-day EMA with the June-quarter print due August 20.

Changes Since the June 7, 2026 Report

This is a follow-up engagement (the prior article is dated 2026-06-07). Nothing in Alibaba’s reported financial statements has changed — the FY2026 20-F filed May 20, 2026 remains the latest audited data, and the June quarter is not reported until August 20. What changed is the legal, regulatory, competitive and capital-allocation picture, and the price. Six items move the thesis.

1. The geopolitical tail activated (adverse, and the most important item). On June 8, 2026 the Department of Defense added Alibaba to its Section 1260H list of “Chinese Military Companies,” expanding the list to 188 entities, on the stated basis that Alibaba is “indirectly affiliated” with SASAC and that its MIIT ties make it a “military-civil fusion contributor to the Chinese defense industrial base.” Alibaba disclosed the inclusion itself in a 6-K on June 9 and on June 24 sued the DoD in federal court in San Jose, arguing the designation is arbitrary and capricious and violates due process. The prior article described this class of risk as “dormant-but-reinstatable”; it is now live. It is not, however, binding — see the risk analysis below for the legal anatomy.

2. Buybacks restarted, and restarted well (favourable). The prior article’s sharpest criticism was that repurchases collapsed from ~US$12B/year to US$1.0B in FY2026 with ~US$19B of authorization unused. Alibaba’s own Hong Kong regulatory returns, furnished on Form 6-K, show open-market NYSE repurchases resuming on June 22, 2026 — 31,333,832 ordinary shares (3.92M ADS-equivalents) repurchased and cancelled through about July 22 for roughly US$384M, executed with clear price sensitivity.

3. The subsidy ceasefire became policy (favourable). On June 17 SAMR published draft “Ten Rules for Regulating Food Delivery Platform Subsidies,” banning long-term large-scale subsidies and coercive merchant/rider cost-sharing. Meituan, JD and Taobao Flash all pledged compliance. The prior article’s bull case rested on an informal ceasefire; it now has a draft rulebook behind it.

4. Qwen won Apple, and shipped a frontier model (favourable). On July 15–16 the Cyberspace Administration of China approved Apple Intelligence for China with Qwen as the language-model partner (Baidu handles visual search), the model compressed from 54GB to under 4GB for on-device use. On August 3 Alibaba released Qwen3.8-Max at 2.4 trillion parameters.

5. Two legal overhangs — one closed, two opened (mixed). The DOJ closed an eight-year investigation on July 1 with non-prosecution agreements and a combined US$600M payment over illegal pharmaceutical and pill-press sales, with Alibaba admitting ~80,000 illegal transactions across 2016–2024. Opening in its place: Anthropic’s June 24 letter to the Senate Banking Committee alleging a large-scale adversarial-distillation campaign, and a securities class action (Wistisen v. Alibaba, class period June 26, 2025 – June 24, 2026, lead-plaintiff deadline October 5, 2026).

6. The chip constraint inverted (mixed, and analytically important). In May 2026 the U.S. approved roughly ten Chinese firms including Alibaba to buy up to 75,000 Nvidia H200s each. Not one has been delivered — Beijing has pressed platforms to favour domestic silicon. The prior article’s bear case assumed Washington was the binding constraint on Alibaba Cloud. It is now Beijing.

Scorecard against the prior article’s falsification tests. Bull test 4 (the asset floor is at least partly monetizable to foreign holders) is partially confirmed by the buyback. Bull tests 1 and 2 (cyclical EBITA recovery; cloud sustaining 30%+) are tracking favourably but unproven — the August 20 print is the test. Bull test 3 (CMR holds share) has no new evidence. Bear test 2 (chip controls cap cloud below 20%) is not supported as framed, though the mechanism changed. Bear test 3 (the asset floor is non-fungible to ADR holders) is partially falsified — offshore cash was in fact deployed to buy ADS. Bear test 4 (the VIE/delisting tail reactivates on a geopolitical shock) is partially hit, and it is the reason this report’s stance is a step down rather than a step up.

1. Executive Summary

Alibaba is unchanged as a business since June and materially changed as a claim. The operating picture improved at the margin; the legal and jurisdictional picture deteriorated; the price rose 4.8%. Net, the same arithmetic that made the stock interesting in early June holds with considerably less slack.

Nothing new in the numbers — by construction. The FY2026 20-F (year ended March 31, 2026, filed May 20, 2026) remains the latest audited data and was already fully incorporated in the prior article: revenue RMB1,023,670M (US$148.4B), +3% reported and +11% like-for-like after the Sun Art and Intime disposals; group adjusted EBITA −56% with the margin falling from 17.4% to 7.5%; operating income −64%; and free cash flow swinging to a −RMB46.6B outflow, the first negative year in Alibaba’s public life. ROIC.ai’s computation of the same year puts return on invested capital at 2.84%, down from 8.06% in FY2025, with ROE at 10.25% and operating margin at 5.83%. That is a company that, as reported, earned well below its cost of capital. The prior article’s framing — a deliberately chosen trough, bought with a RMB380B three-year AI build-out and an instant-commerce subsidy war — remains the right lens, but the number itself sets the burden of proof squarely on recovery, and recovery is not yet evidenced. The June quarter reports August 20, thirteen days after this article.

The improvement is real but partial. Three things got better. Beijing converted the food-delivery ceasefire from jawboning into a June 17 draft rulebook that bans sustained large-scale subsidies outright — the most durable fundamental improvement in the window, because a regulatory prohibition survives competitive temptation in a way a gentleman’s agreement does not. Alibaba’s Qwen won the Apple Intelligence slot in China on July 15, a regulator-blessed design win inside the country’s most valuable consumer hardware base; it carries little near-term revenue but substantial distribution and credibility value. And management restarted buybacks on June 22 after a ten-month pause, buying roughly 1.0M ordinary shares a day initially, scaling to roughly 4.2M a day at the 52-week low across June 29 – July 2, then tapering to 72,800 on July 9 as the stock gapped +11% and stopping altogether after about July 22 — 31.3M ordinary shares and ~US$384M in total, disclosed in Alibaba’s own Hong Kong filings. That is genuine price-sensitivity, and it is the direct answer to the prior article’s single sharpest capital-allocation complaint.

The deterioration is jurisdictional, and it is the reason the stance steps down. On June 8 the Pentagon designated Alibaba a “Chinese Military Company” under Section 1260H. The legal anatomy matters and is frequently misreported: 1260H is a Defense Department identification list whose direct consequences are procurement-side — no direct DoD contracts from June 2026, no indirect component procurement from 2027 — and it imposes no restriction whatsoever on U.S. persons buying, holding or selling the securities. The list that carries that prohibition is Treasury’s NS-CMIC list under E.O. 14032, and Alibaba is not on it. So nothing binding has happened. What has happened is that the conditional probability of the binding outcome rose, that the process is now adversarial and public (Alibaba sued the DoD on June 24), and that two adjacent U.S. legal fronts opened in the same window: Anthropic’s June 24 letter to the Senate Banking Committee alleging that Alibaba-affiliated operators ran 28.8 million exchanges through ~25,000 fraudulent accounts to distil its models, and a securities class action filed in early August. Separately, the DOJ closed an eight-year probe on July 1 for US$600M and a non-prosecution agreement — financially trivial at ~0.2% of market cap, but the admitted facts (roughly 80,000 illegal pharmaceutical transactions, 2016–2024, with internal compliance warnings) are now permanent public record available to any future U.S. agency.

Cloud is the franchise; the constraint on it moved. Cloud Intelligence grew 34% in FY2026 to RMB158,132M with adjusted EBITA of RMB14,265M, accelerating to +38% in the March quarter (external +40%), with AI products at ~30% of external cloud revenue and an eleventh consecutive quarter of triple-digit AI growth. Third-party consensus for the June quarter clusters near 45% cloud growth. The interesting development is that the U.S. approved Alibaba (among ~10 Chinese firms) to buy up to 75,000 Nvidia H200s each in May 2026 — and Beijing then blocked delivery in favour of domestic silicon. Alibaba now holds American permission it is not permitted to use. That inverts the prior article’s bear case: the binding constraint is Beijing’s industrial policy, not Washington’s export controls. It caps near-term access to best-in-class training silicon while making Alibaba’s in-house accelerator programme strategically protected.

Valuation: same structure, thinner cushion. At US$126.81 against 19,174,988,918 ordinary shares (2,396,873,615 ADS-equivalents), market capitalisation is ~US$304B versus ~US$290B in June. Netting the ~US$80–105B non-operating block established in the prior article — gross cash and liquid investments of RMB520.8B (US$75.5B), net cash ~US$37.8B, the 33% Ant stake and the equity portfolio, less the US$600M DOJ payment and ~US$0.4B of buybacks — the market now pays roughly US$200–225B for all operating segments, against US$185–210B in June. The no-growth EPV of ~US$150–200B plus the asset block still brackets the capitalisation, so the AI-cloud growth is still lightly priced; but at the June 26 low of US$94.81 (cap ~US$227B) the operating stub was ~US$122–147B and the margin of safety was wide. It is now thin. The AZI own-history percentile reads 23.7 composite — but it was struck on July 24 at a price of US$112.14 and is stale to a +13% rally; re-scaled to today’s price the multiples (P/E ~20.7, P/B ~2.0, P/S ~2.1) sit closer to the low-30s percentile, roughly where the prior article found them. The June sell-off produced a genuinely cheap own-history print; July took most of it back.

The positioning read is unflattering and should be stated plainly. The factor model gives Alibaba a negative momentum loading (−0.44 sparse, −0.50 full), a positive value loading (+0.53), and a dominant China country loading of +1.87 — the fundamental analysis explains a minority of this stock’s variance. Its factor-nearest comparables are China index ETFs (FXI, KWEB, MCHI all above 0.92 similarity), not e-commerce or cloud peers. And the risk-adjusted record is poor: a ten-year annualized return of 4.95% with an 80% maximum drawdown and a Sharpe of 0.068, and a five-year annualized return of −7.14%. Price sits just below its 200-day EMA after a +27.4% July.

The investment question, restated. In June the question was whether the cheapness was a mispricing or an earned discount. That question is unresolved and will stay unresolved until at least August 20. The new question this update adds is narrower and sharper: has the required jurisdictional discount widened by more than the operating improvement is worth? On the evidence assembled here — a live military-company designation and active federal litigation on one side, a formalized subsidy ceasefire, an Apple design win and a price-sensitive buyback on the other, with the price 4.8% higher — the honest answer is that it has. This analysis takes no position on what to do about that; it is rec-free and price-target-free by design.

2. Business Overview

The structural description of Alibaba established in the prior article is unchanged and is not re-derived here; it is summarised, updated for the events of the window, and then pressure-tested where the update supplies new evidence.

Alibaba Group Holding Limited is a Cayman Islands holding company that controls PRC operating entities through variable interest entities and contractual arrangements, operating China’s largest e-commerce ecosystem and its largest public cloud business alongside international commerce, logistics, local services and an artificial-intelligence franchise. It reports under U.S. GAAP in Renminbi. Each American Depositary Share represents eight ordinary shares; the primary listing is in Hong Kong (9988, with an RMB counter at 89988) under a weighted-voting-rights structure controlled by the Alibaba Partnership. Because Alibaba is a foreign private issuer, its U.S. disclosure runs through Form 20-F and Form 6-K — there is no 10-K, no 10-Q and no DEF 14A. This is not a technicality for research purposes: the most valuable primary disclosure in this update window arrived not through a U.S. earnings release but through Hong Kong Listing Rule 13.25A “Next Day Disclosure Returns” furnished as 6-K exhibits, which is where the buyback restart is documented.

The four-segment architecture (FY2026, year ended March 31, 2026).

(Source: FY2026 earnings release segment tables, as reported and carried forward from the prior article. Group totals net inter-segment elimination; group adjusted EBITA was RMB76,416M.)

The central structural fact is unchanged and worth restating because every valuation conclusion below depends on it: China E-commerce’s RMB107,509M of adjusted EBITA exceeds the entire group’s RMB76,416M, because AIDC, All Others and unallocated corporate costs collectively lose roughly RMB45B. One mature take-rate business funds the whole enterprise — the AI build-out, the quick-commerce war, the consumer Qwen push, and the international expansion. Cloud, the strategic centrepiece, contributes RMB14,265M, or roughly 13% of what Taobao/Tmall’s monetization engine produces.

What the update adds to the business description. Three changes are worth recording.

First, the open-weight model is being commercialised. Reuters reported on August 6 that Alibaba plans to require large commercial users of the next open-source Qwen release to share a portion of the revenue they generate from it. Alibaba’s open-weight strategy to date has been a pure distribution and mindshare play with no direct monetization — the models are downloaded, modified and deployed for free, and Alibaba captures value only indirectly through Model Studio hosting and cloud consumption. A revenue share on heavy commercial users is an attempt to convert that mindshare into a take rate, which is the same economic instinct that built Taobao. Whether it is enforceable on genuinely open weights is a serious open question — the entire point of an open licence is that redistribution is not gated — and this should be tracked as a business-model experiment rather than modelled as revenue.

Second, the instant-commerce business is shifting from subsidy to acquisition. With SAMR’s draft rules banning sustained large-scale subsidies, competition in instant retail is migrating to supply-chain M&A; Alibaba is reported to have bid ~US$1.5B for Pupu Supermarket, mirroring Meituan’s earlier Dingdong Maicai purchase. This is cheaper than a subsidy war but not free, and it partially re-introduces the balance-sheet-heavy, low-ROIC offline-retail exposure that Alibaba spent FY2025–FY2026 divesting through the Sun Art and Intime sales. That is a coherent strategic trade in a 30-minute-delivery market that genuinely requires owned supply, but it deserves scepticism from anyone who watched the Sun Art and Intime experiments.

Third, the Apple Intelligence slot changes what Qwen is. Until July 15, Qwen was a strong model with a large developer following and a captive cloud channel. Chinese regulation requires foreign firms to run generative AI on a domestic partner’s model, and the CAC’s approval designates Qwen as that partner for Apple in China, with Baidu handling visual search. The engineering detail is telling — the model was compressed from 54GB to under 4GB to run locally on hardware back to the iPhone 15 — because on-device deployment at that compression is a hard capability, and the position is sticky once shipped. No commercial terms were disclosed and no launch date was set, so the near-term revenue is immaterial; the asset is distribution and validation.

Verdict. The business Alibaba operates is the same one the prior article described: an excellent, narrowing-moat take-rate marketplace attached to a genuinely good and improving cloud franchise, wrapped in a conglomerate that consumes the cash both generate. Nothing in this window improved the core marketplace economics — there is no new GMV, take-rate or share data at all — and everything that improved improved at the periphery or in cloud. The most economically significant change to the business in the window is regulatory: the ban on sustained subsidies removes the mechanism that halved China-commerce EBITA. That is a real improvement, and it is also a reminder that the profitability of Alibaba’s largest profit pool is currently being set in Beijing rather than in Hangzhou.

3. Industry Dynamics

The three industries Alibaba straddles — Chinese e-commerce, Chinese public cloud/AI, and international commerce — were mapped in the prior article and their structure has not changed in two months. What has changed is the regulatory regime over the first, and the supply constraint over the second. Both are material.

Chinese e-commerce: a mature, fragmented, price-warring market that has just had price competition partially outlawed. The structural read carries forward — China is the world’s largest e-commerce market, growing at roughly mid-to-high single digits and tracking soft domestic consumption; Pinduoduo’s hard-discount, manufacturer-direct model reset price expectations for the entire market and took share from roughly 7% to roughly 19% between 2019 and 2023; and Douyin and Kuaishou converted short-video attention into an estimated ~US$650B of GMV from a standing start, a demand-generation mechanism Alibaba structurally cannot replicate. These share figures are carried from the prior article, derive from third-party estimates, and are now roughly two years stale; no fresh share data was published in this window, and this article does not pretend otherwise.

The genuinely new element is the regulatory intervention in instant retail. The sequence is worth setting out because it establishes durability. SAMR flagged the food-delivery price war as a top-ten “involutionary competition” enforcement case in late January 2026 and summoned seven platforms on February 13. On March 25 SAMR reposted an Economic Daily commentary titled “The Food Delivery War Should End,” which moved Meituan’s Hong Kong shares more than 12% intraday and Alibaba and JD more than 5% each. Then on June 17, 2026 SAMR published the “Ten Rules for Regulating Food Delivery Platform Subsidies” as a draft for public comment, banning long-term, large-scale subsidies and coercive cost-sharing with merchants and riders, and requiring seven days’ advance notice of promotions. Meituan, JD.com and Alibaba’s Taobao Flash each pledged compliance within about an hour.

The analytical point is that this is a different kind of ceasefire from the one the prior article relied on. An informal truce among three well-capitalised rivals in a land-grab market is unstable by construction — the first defector wins share, so all three defect. A regulatory prohibition changes the payoff matrix: defection now carries an enforcement penalty rather than a competitive reward. Marathon’s capital-cycle framework would read the whole episode as a textbook supply-side overshoot — high perceived returns in 30-minute delivery attracted enormous capital, which destroyed the returns that attracted it — with the unusual wrinkle that the correction is being imposed by a regulator rather than by bankruptcy. Cumulative losses across the war are estimated at roughly RMB87B (~US$12.9B) for Alibaba, RMB46.6B for JD’s new-business unit, and a swing to loss at Meituan. These are third-party estimates with no filing support — Alibaba does not disclose a discrete instant-commerce loss line — and they are not used anywhere in this article’s valuation work.

Two cautions. The rules are a draft: final text, effective date and enforcement teeth are all unknown, and Chinese regulators have called an end to this particular war before without it ending. And the competitive energy has not dissipated, it has redirected — into supply-chain M&A, which consumes capital differently but still consumes it.

Chinese cloud and AI: a consolidated, re-accelerating oligopoly whose input supply is now controlled by Beijing rather than Washington. Structurally this remains the far better industry: high fixed-cost barriers, genuine switching costs in embedded workloads and data, and a top-three concentration with Alibaba at roughly 36–40% share, ahead of Huawei and Tencent. Alibaba’s cloud EBITA margin of roughly 9% is thin for the structure, which is a function of the build-out phase rather than of competitive pressure.

The update’s contribution is the chip-supply inversion, which deserves careful statement because it is easy to get backwards. In May 2026 the U.S. Commerce Department approved roughly ten Chinese firms — Alibaba, Tencent, ByteDance and JD.com among them — to purchase up to 75,000 Nvidia H200 units each, under a January 2026 arrangement in which China-bound advanced chips route through the United States for testing and certification and the U.S. government takes 25% of the sale revenue. Zero units have been delivered. Beijing has pressed domestic platforms to slow or redirect Nvidia orders in favour of Huawei and other domestic silicon.

So the constraint on Chinese AI compute is no longer primarily a U.S. export control; it is a Chinese industrial-policy preference. For Alibaba specifically this cuts three ways. It caps near-term access to best-in-class training silicon, which is a real growth constraint on the frontier-model programme. It converts Alibaba’s in-house accelerator work from a defensive hedge into a strategically favoured asset. And it means that a future U.S.–China thaw on chips would not automatically unlock compute for Alibaba, because the second veto is domestic — which removes one of the cleaner bull catalysts the market has periodically priced.

International commerce (AIDC): unchanged. Cross-border and Southeast Asian retail through AliExpress, Trendyol, Lazada, Daraz and Alibaba.com narrowed its adjusted-EBITA loss from RMB15,137M to RMB2,051M on +9% revenue, approaching breakeven on AliExpress “Choice” unit economics. It remains subscale against Amazon, Sea’s Shopee and PDD’s Temu, and has never demonstrated franchise economics. No material development in the window.

Verdict: two structurally attractive positions inside one structurally unattractive industry, with the regulatory environment for the unattractive one improving. Chinese e-commerce is a bad industry — low barriers since the 2021 exclusivity ban, multi-homing merchants, chronic price competition, compressed profit pools — and Alibaba’s position within it is the best available but is narrowing. Chinese cloud is a good industry — high barriers, real switching costs, an oligopoly, and demand re-accelerating on AI — and Alibaba’s position within it is the strongest in the country. The June 17 subsidy rules genuinely improve the first; the H200 delivery block genuinely constrains the second. On balance the industry backdrop is modestly better than it was on June 7, and both changes were made by the Chinese state, which is itself the observation an investor should take away: the two largest swing factors in Alibaba’s forward profit pool were both set by regulators in a two-month window.

4. Competitive Position

The moat analysis from the prior article stands, and this update supplies genuinely new evidence on one side of it (cloud/AI) and none at all on the other (the marketplace). That asymmetry is itself the finding.

Naming the moats in Greenwald’s taxonomy. Alibaba does not have one moat; it has two, of different types and different trajectories.

Taobao/Tmall is a demand-side/customer-captivity advantage reinforced by economies of scale — and it is narrowing. The mechanism is habit and default behaviour on the consumer side, a two-sided liquidity advantage (the most merchants because the most buyers, and vice versa), and a genuinely captive high-value cohort in 88VIP, which passed 62 million members with double-digit growth. The financial signature of that moat is customer management revenue — the pay-for-performance advertising, commissions and software-service fees levied against GMV — which produced RMB343,867M in FY2026, +5% reported and +7% like-for-like. Greenwald’s decisive test for a moat is market-share stability, and Alibaba fails it: Taobao/Tmall’s GMV share fell from roughly 50% in 2020 to roughly 40–44% by 2024 while PDD rose to ~19% and Douyin built ~US$650B of GMV from zero. Share that moves that far, that fast, in a market with multi-homing merchants and no exclusivity (banned since 2021) is share protected by inertia rather than by a barrier. The most honest single piece of evidence on this remains the FY2026 accounting change flagged in the prior article: Alibaba reclassified merchant subsidies from sales-and-marketing expense into contra-revenue against CMR, which suppressed reported Q4 CMR growth to +1% against +8% like-for-like. A platform paying merchants to spend more on it is a platform whose pricing power has weakened. No new share, GMV or take-rate data was published in this window, so nothing has confirmed or refuted the narrowing since June 7; the August 20 print is the next observation.

Alibaba Cloud is an economies-of-scale advantage plus switching costs — and it is widening. The mechanism here is conventional and durable: enormous fixed costs in data centres and silicon amortised over the largest workload base in China, plus the customer captivity that comes from migrated data, integrated services and rewritten applications. The financial signature is 34% FY2026 revenue growth accelerating to +38% in the March quarter (external +40%), with adjusted EBITA up 35%. The AI layer strengthens the same mechanism — AI products reached ~30% of external cloud revenue with an eleventh consecutive quarter of triple-digit growth — because a customer that has built on Qwen through Model Studio faces a much higher switching cost than one that merely rents compute.

What the update adds: two hard datapoints on the cloud/AI side.

The Apple Intelligence approval on July 15–16 is, in competitive-position terms, the most valuable event of the window. Chinese regulation forces foreign firms to run generative AI on a domestic partner’s model; the CAC’s approval installs Qwen as that partner for Apple in China. The engineering requirement — compressing the model from 54GB to under 4GB to run on-device on hardware as old as the iPhone 15 — is a real capability barrier, and the resulting position is sticky: replacing the on-device language model in a shipped operating system across an installed base is not a decision Apple revisits casually. What it is not is a near-term revenue event; no commercial terms were disclosed and no launch date was set. Read it as distribution and validation, not as a P&L line.

Qwen3.8-Max, released August 3 at 2.4 trillion parameters and benchmarked against Moonshot AI’s 2.8-trillion-parameter Kimi K3, establishes that Alibaba remains at the Chinese frontier and close to the global one. The market reaction — ADS +4.1%, Hong Kong shares +6–7% — reflects that investors now treat model releases as cloud-franchise news, which is the correct read given the Model Studio monetization channel.

And one hard datapoint against it. On June 24 Anthropic wrote to the U.S. Senate Committee on Banking, Housing and Urban Affairs alleging that operators affiliated with Alibaba and its Qwen lab conducted the largest known “adversarial distillation” campaign against its Claude models — more than 28.8 million exchanges through approximately 25,000 fraudulent accounts between April 22 and June 5, 2026. This is an allegation, not a finding: there is no litigation, no regulatory action and no admission as of this report’s date, and no Alibaba public rebuttal was located. Its relevance to competitive position is nonetheless direct. If substantiated, it would partially undercut the premise investors are paying for — that Alibaba possesses independent, full-stack frontier-model capability — by suggesting some of that capability was harvested from a competitor rather than built. Adversarial distillation, if it works at the scale alleged, is a cost advantage of a very particular kind: it lowers the price of reaching the frontier but does not confer the ability to define it, and it is precisely the sort of advantage that disappears the moment the target hardens its defences or a court intervenes. An investor underwriting Alibaba’s AI franchise on the basis of durable research capability should treat this allegation as directly on point, and should note that it landed in the Senate at the exact moment Alibaba was contesting a Pentagon blacklisting.

Direct comparison against the competitive set. In e-commerce, PDD competes on price with a structurally lower cost-to-serve and has proven it can take share; Douyin competes on demand generation, which Alibaba cannot replicate because Alibaba is a search-and-fulfilment destination and Douyin is impulse discovery; JD competes on owned logistics and 1P reliability. In instant retail, Meituan holds the incumbent density advantage and Alibaba has been buying share with subsidies that are now being outlawed. In cloud, Huawei is the strategically favoured domestic alternative — an important point given Beijing’s silicon preference — and Tencent is the distant third. The factor model offers a blunt cross-check on how the market actually sees the comp set: Alibaba’s factor-nearest names are China index ETFs (FXI 0.929, KWEB 0.926, MCHI 0.923), with only JD (0.901) and Baidu (0.843) registering as single-stock comparables. The market is not benchmarking Alibaba against Amazon or Microsoft; it is benchmarking it against China.

Verdict: a durable but narrowing marketplace advantage and a widening cloud advantage, with the balance of new evidence favouring cloud. The prior article’s conclusion — that the moat is real, is not impregnable, and is deteriorating on the commerce side while strengthening on the cloud side — survives this window intact and, if anything, is reinforced. Every piece of new competitive evidence in two months concerned cloud and AI; not one concerned the marketplace. That is consistent with a company whose strategic centre of gravity is moving, and it is also a warning: roughly 140% of group adjusted EBITA still comes from the franchise about which nothing new was learned.

5. Growth History and Forward Opportunities

The growth picture is unchanged in its reported history and modestly clarified in its forward setup.

Reported history (FY2026). Group revenue of RMB1,023,670M grew 3% as reported and 11% like-for-like, the gap being the deliberate disposal of the Sun Art and Intime offline-retail businesses. Underneath: China E-commerce +9% (CMR +5% reported / +7% like-for-like, quick commerce +47% to RMB78,520M, direct sales and logistics RMB105,518M, 1688 wholesale RMB26,312M); Cloud +34%; AIDC +9%; All Others −25% on the disposals. This is a company growing low-double-digits organically with the mix shifting toward its two structurally better businesses (cloud) and its worst-economics one (subsidised quick commerce) simultaneously.

The quality question. The prior article’s verdict — that this is mixed-quality growth, high-quality in cloud and low-quality in quick commerce — is worth restating with the update’s evidence. Quick commerce grew 47% by paying for it: the revenue is reported net of contra-revenue subsidies, and the segment is the principal cause of the 44% collapse in China E-commerce adjusted EBITA. Growth bought at negative contribution margin is not growth in any sense a fundamental investor should credit; it is market-share purchase, and its value depends entirely on whether the acquired customers retain at economic prices once the subsidies stop. The June 17 SAMR rules will force exactly that experiment, which makes the next several quarters unusually informative: for the first time, Alibaba’s instant-commerce cohort will be tested at something approaching real prices, by regulatory fiat rather than by choice. That is a genuine natural experiment, and its result is the most important single piece of forward evidence available on the segment.

Cloud growth, by contrast, is high quality on every test that matters: it is organic, it accelerated through the year (34% full-year, 38% in Q4, external +40%), it carries positive if thin EBITA (~9% margin) that grew 35%, and its fastest-growing component — AI products at ~30% of external revenue with eleven straight quarters of triple-digit growth — is the piece with the highest switching costs. Third-party consensus for the June quarter clusters near 45% cloud growth, which would be a record if delivered. That is market expectation, not company guidance — Alibaba does not guide — and it is reported here as the bar the print will be measured against, not as a forecast.

Forward opportunities, updated. Four are live, and this window changed the standing of three of them.

Cloud/AI monetization is the largest and is strengthening. The Apple Intelligence slot supplies distribution; Qwen3.8-Max supplies frontier credibility; the reported plan to charge large commercial users of the next open-weight release supplies a possible new revenue mechanism. The constraint is compute, and the constraint is now Beijing’s rather than Washington’s.

Instant-retail normalization is the highest-variance opportunity. If the SAMR rules hold, Alibaba retains a materially larger quick-commerce business than it had two years ago at dramatically lower burn. The migration to supply-chain M&A — the reported ~US$1.5B Pupu Supermarket bid — is how that consolidation is being executed, and it is a strategically coherent but capital-hungry path that partially rebuilds the asset-heavy retail exposure the company just spent two years shedding.

AIDC scaling to profitability is on track and modest: the loss narrowed from RMB15,137M to RMB2,051M. Nothing new in the window.

Ant Group monetization remains the largest unexercised option and remains entirely dormant. Alibaba’s 33% stake is carried within the RMB206,800M equity-method line and contributed RMB375M of profit in the March quarter (recognised one quarter in arrears). No re-IPO catalyst, no valuation event and no restructuring milestone emerged in this window. Investors have now been waiting nearly six years since the cancelled 2020 IPO; the correct treatment is to hold this as unpriced optionality with no timetable, not as a catalyst.

Verdict: low-to-mid quality growth in aggregate, improving at the margin, with the mix moving in the right direction for the wrong reason. Group growth is unimpressive and the reported figure is depressed by deliberate disposals; cloud growth is genuinely excellent and accelerating; quick-commerce growth is bought and about to be tested; the marketplace core is growing mid-single-digits at best. The improvement in this window is real but is regulatory rather than commercial — the growth mix improves because Beijing is forcing Alibaba to stop buying the low-quality kind. That is a better outcome than the alternative, and it is not the same thing as the company having found a better growth engine.

6. Financial Quality

No new financial statements were published in this window, so this section restates the FY2026 position with the third-party ratio work as a cross-check, and adds the two balance-sheet-relevant events that did occur.

The FY2026 collapse, quantified three ways. Group adjusted EBITA fell 56% to RMB76,416M with the margin falling from 17.4% to 7.5%. Operating income fell 64%, with the margin falling from roughly 14% to roughly 5%. Free cash flow swung from +RMB73.9B to −RMB46.6B, the first negative year in Alibaba’s public life. ROIC.ai’s independent computation of the same year, reconciled directionally to the filing, gives the following multi-year trend:

(Source: ROIC.ai MCP, get_profitability_ratios, annual, six periods. Third-party aggregated data, reconciled directionally to the FY2026 20-F; the filing is the authority.)

Three observations from that table. First, FY2026 ROIC of 2.84% is below any defensible cost of capital for a China-domiciled, VIE-structured equity — this company, as reported, destroyed economic value last year. That is what the words “chosen trough” actually mean in numbers, and stating it plainly is more useful than softening it. Second, gross margin barely moved (39.95% → 39.81%), which is the single most important quality signal in the table: the collapse is entirely below the gross line, in subsidies and operating spend, not in the pricing or cost structure of the core marketplace. A moat that was breaking would show up in gross margin; it has not. Third, ROE of 10.25% overstates the operating deterioration and understates the core’s quality simultaneously — it is held up by RMB87.5B of non-operating equity mark-to-market gains while being held down by a balance sheet bloated with cash, low-yield associates and goodwill. Neither ROE nor ROIC as computed on the consolidated entity is a fair measure of the Taobao/Tmall engine, which earns very high incremental returns on very little incremental capital.

Earnings quality: GAAP net income remains the wrong number. FY2026 GAAP net income of RMB102,127M was down 19% and is flattered by RMB87.5B of non-operating equity mark-to-market gains. Non-GAAP net income fell 62%. Any analysis anchored on the GAAP figure — including, mechanically, the P/E-based own-history percentile discussed in the valuation section — overstates the earnings power actually attributable to operations. The cleaner reads are non-GAAP net income, segment adjusted EBITA, and free cash flow, and all three say the same thing: FY2026 was far worse than GAAP EPS suggests.

Capital intensity: the structural change. Alibaba spent most of its life as an asset-light marketplace. Capex rose 47% to RMB126B in FY2026 within a RMB380B (~US$53B) three-year AI and cloud programme, and property and equipment rose 39% to RMB282.7B. This is the swing factor that turned free cash flow negative and it is not reversible in the near term — the programme has roughly two years to run. The right way to hold this is that Alibaba has voluntarily converted itself from an asset-light compounder into a capital-intensive infrastructure business in one of its two segments, and the return on that RMB380B is the largest single unresolved question in the equity.

Balance sheet: a fortress, with a caveat about whose fortress it is. Gross cash and liquid investments of RMB520.8B (US$75.5B) and net cash of roughly US$37.8B, plus the 33% Ant stake and an equity-securities portfolio carried at RMB449.9B, give Alibaba one of the strongest balance sheets in global equities. The caveat established in the prior article is unchanged and important: a substantial portion of the cash is onshore RMB subject to capital controls, and the Ant stake is illiquid and politically constrained, so the accounting balance sheet overstates what is fungible to a foreign ADR holder. This update supplies the first direct evidence on that question in either direction, and it is favourable — see the capital-allocation section below.

Balance-sheet events in the window. Two, both small. The US$600M DOJ settlement is roughly 0.2% of market capitalisation and 0.8% of gross liquid assets — immaterial as cash. And ~US$384M of share repurchases were funded and executed offshore. Together they consume under US$1B against a US$75.5B liquid stack.

Dilution and the convertibles. Share-based compensation remains modest at roughly 1.4% of revenue, and the diluted share count fell slightly in FY2026 — genuinely good discipline by large-cap technology standards. The July monthly return discloses two convertible issues outstanding: US$4,999,798,000 of May 2024 notes at a US$12.5222 conversion price (up to 519,056,000 ordinary shares) and US$3,168,000,000 of September 2025 notes at US$23.9244 (up to 173,801,552 ordinary shares). At the current ordinary-share equivalent of ~US$15.85 (US$126.81 ÷ 8), the May 2024 notes are deep in the money — roughly 519M ordinary shares, or ~65M ADS-equivalents and ~2.7% of the ADS count, of dilution sold cheaply in 2024. The prior article flagged the internal inconsistency of issuing equity-linked debt while pausing buybacks; that inconsistency has now become a realised cost, because the company is repurchasing stock at prices well above where it sold the option.

Verdict: economics do not currently improve with scale, and the reason is deliberate. On reported FY2026 figures Alibaba is a low-return business — ROIC 2.84%, negative free cash flow, operating margin cut in two-thirds. The evidence that this is investment rather than impairment is genuine and rests on three legs: gross margin held flat, the collapse is traceable to two identifiable discretionary programmes, and the disposals that suppressed reported revenue were value-accretive exits from low-ROIC retail. But “we chose this” is a hypothesis until the recovery is observed, and it has not been. The financial-quality verdict is therefore unchanged from June and deliberately unresolved: the core marketplace’s unit economics remain excellent and are untouched by the collapse; the consolidated entity’s economics are currently poor; and which of those is the true steady state is precisely what August 20 begins to answer.

7. Capital Allocation

This is the section the update changes most, and the change is favourable.

What the prior article concluded. Alibaba repurchased US$12.5B of stock in FY2024 and US$11.9B in FY2025, each retiring roughly 5.1% of the float, executed into a low valuation — textbook capital allocation. Then FY2026 buybacks collapsed to US$1.0B (73 million ADS, and only between April and August 2025), leaving roughly US$19B of authorization unused, while management redirected cash into the RMB380B AI programme and the subsidy war and simultaneously issued equity-linked debt. The prior article called this “the single most important capital-allocation signal” and read it as a pivot from returning capital to consuming it, with the added inconsistency of selling cheap optionality on the shares while declining to buy the shares.

What actually happened next. Alibaba’s own Hong Kong Listing Rule 13.25A returns, furnished as exhibits to Forms 6-K on June 26, July 6, July 7 and July 10, disclose daily NYSE open-market repurchases resuming on June 22, 2026 — the first disclosed since August 2025. The daily record, in ordinary shares at USD prices per ordinary share (multiply by eight for the ADS-equivalent price):

(Source: SEC Forms 6-K dated 2026-06-26, 2026-07-06 and 2026-07-10, Exhibits 99.1–99.4 — Hong Kong Next Day Disclosure Returns. The ADS-equivalent prices reconcile to the AZI daily closes: the June 22 figure of $13.12 × 8 = $104.96 against an actual close of $104.97, which confirms the unit convention.)

The June total of 13,387,512 ordinary shares reconciles exactly to the June monthly return. The July monthly return, filed August 6, discloses cancellations of 29,791,432 ordinary shares on July 9 at a VWAP of US$12.17 and 1,542,400 on July 22 at US$13.61 — 31,333,832 ordinary shares (3,916,729 ADS-equivalents) repurchased and cancelled in total, for approximately US$384M. Issued shares fell from 19,206,078,670 at June 30 to 19,174,988,918 at July 31.

Why the pattern matters more than the amount. US$384M against a US$304B capitalisation is 0.13% — trivially small in isolation, and materially below the ~US$1B/month pace of FY2024–25. Taken alone it would be a footnote. What makes it a genuine thesis input is the shape: purchases began as the stock broke down on June 22, ran at roughly 1.0M ordinary shares a day through the initial decline, scaled roughly four-fold to ~4.2M a day across June 29 – July 2 with the ADS at its 52-week low, collapsed to 1.47M on July 8 as the stock gapped +11%, fell to 72,800 on July 9, and stopped entirely after about July 22 at ~US$109/ADS. That is not a 10b5-1-style programmatic drip; it is discretionary, price-sensitive, buy-more-when-cheaper behaviour, and it is exactly what a shareholder should want from a management team sitting on US$75.5B of liquid assets.

It also resolves — partially, and favourably — the prior article’s deepest structural worry. Bear test 3 held that the asset floor might be largely non-fungible to ADR holders, trapped onshore or locked in illiquid Ant. The buyback is direct evidence to the contrary: offshore-accessible cash was in fact deployed to buy ADS on the NYSE at the moment of maximum stress. That does not make the whole stack fungible — onshore RMB capital controls and Ant’s illiquidity are untouched — but it demonstrates the mechanism works and that management will use it.

The honest counter-arguments, stated. Three, and they are not trivial. The scale is small relative both to the ~US$19B authorization and to what US$75.5B of liquidity could support; buying US$384M at the bottom is better than buying nothing, but it is not the ~US$12B/year that shrank the float 5% annually. The buying stopped when the stock recovered, which is defensible discipline but also means the company has been absent for the +27% move and has bought back roughly 0.16% of shares outstanding across the whole episode. And it does not repair the convertible inconsistency: Alibaba sold conversion options at US$12.5222 per ordinary share in May 2024 and is now repurchasing ordinary shares at US$11.84–13.69 — buying back stock at roughly the strike at which it previously sold the upside is not value-destroying, but it is not the sequence a disciplined issuer would have chosen.

The rest of the capital-allocation picture is unchanged. The recurring dividend of US$1.05/ADS (~US$2.5B) continues, with a 33.5% payout ratio on FY2026’s depressed earnings. The RMB380B AI/cloud programme has roughly two years to run and is the dominant use of capital. The M&A story has inverted from acquisition to disposal (Sun Art, Intime, Trendyol GO), which was value-accretive, though the reported ~US$1.5B Pupu Supermarket bid signals a partial return to balance-sheet-heavy retail. Insider economic alignment remains low — all directors and officers hold roughly 1.9% — with control resting on the Alibaba Partnership’s weighted-voting-rights structure rather than on economics, and the annual bonus pool is tied to adjusted pre-tax operating profits.

The insider tape is uninformative, and one reading needs correcting. Thirteen Form 4s were filed since May 29, 2026, and every transaction is equity-plan mechanics: code A annual RSU grants on May 29 (Wu Yongming 448,000 ordinary shares; Jiang Fan 248,000; Yu Siying 120,000), code M vesting and settlement on the June 25 tranche, and code S sell-to-cover. In particular, a June 29 Form 4 reporting 720,000 ordinary shares sold at “94.92–95.59” is not a discretionary insider sale and is not denominated in USD per ADS: the filing’s own footnotes state the shares “were withheld and sold in the open market in Hong Kong … to satisfy tax withholding obligations related to … vesting of restricted share units,” at prices in Hong Kong dollars of HK$94.00–95.00 per ordinary share — roughly US$12.1 per ordinary share, or ~US$97 per ADS-equivalent, on 90,000 ADS-equivalents worth about US$8.7M. Routine. There were zero discretionary open-market purchases (code P) and zero discretionary open-market sales by any insider in the window. The meaningful insider signal this quarter came from the issuer’s own bid, not from the officers’.

Verdict: materially improved, from poor to acceptable, and still short of good. The prior article’s judgement — that management had pivoted from returning capital to consuming it at exactly the wrong valuation — was correct on the FY2026 evidence and has been partially answered by the FY2027 conduct. Management demonstrated it will buy its own stock aggressively at distressed prices using offshore cash, which is the single most valuable thing it could have demonstrated. What it has not yet demonstrated is willingness to repurchase at ordinary prices, or to deploy the authorization at a scale commensurate with the balance sheet. The upgraded verdict is: capital allocation is opportunistic and price-disciplined but under-scaled, inside a capital plan still dominated by an unproven RMB380B bet. Whether repurchases resume above US$120 is the next observation, and it is a cleaner test of management’s view of intrinsic value than anything they will say on the August 20 call.

8. Changes and Headwinds — Last Two Years

The two-year picture from the prior article is intact — the segment restructuring into four groups, the exit from offline retail, the RMB380B AI commitment, the instant-commerce war, and the FY2026 margin and cash-flow collapse. This section covers what the update window added, in chronological order, because sequence is part of the argument.

May 2026 — H200 licences granted, then neutralised. The U.S. Commerce Department approved roughly ten Chinese firms, Alibaba among them, to purchase up to 75,000 Nvidia H200 units each under the January 2026 routing-and-revenue-share arrangement. Beijing subsequently pressed platforms to slow or redirect those orders in favour of domestic silicon, and no units have been delivered. A headline positive that became an operating non-event.

June 8 — the Section 1260H designation. The Department of Defense published its updated Chinese Military Companies list after the U.S. close, adding 65 entities to reach 188 and including Alibaba, Baidu and BYD. DoD’s basis: indirect affiliation with SASAC, and MIIT ties making Alibaba a “military-civil fusion contributor to the Chinese defense industrial base.” Alibaba disclosed it in a 6-K on June 9. The ADS fell 3.9% over two sessions to US$115.38 on June 10, and continued lower.

June 17 — SAMR’s subsidy rules. The draft “Ten Rules for Regulating Food Delivery Platform Subsidies,” banning long-term large-scale subsidies and coercive merchant/rider cost-sharing and requiring seven days’ notice of promotions. Meituan, JD and Taobao Flash pledged compliance within about an hour. In a normal tape this would have been the dominant news of the month for Alibaba’s largest profit pool; it was swamped by the geopolitical flow, which is itself a useful observation about what sets this stock’s price.

June 22 — the buyback restarts. Documented in the capital-allocation section above.

June 24 — two events on one day. Anthropic sent its letter to the Senate Banking Committee alleging the adversarial-distillation campaign (28.8 million exchanges, ~25,000 fraudulent accounts, April 22 – June 5). And Alibaba filed suit against the Department of Defense in federal court in San Jose, arguing the 1260H designation is arbitrary and capricious, has no basis in fact or law, and violates due-process and First Amendment rights; Alibaba’s position is that it is regulated by SASAC and MIIT, not affiliated with them, and has no choice about compliance. Bloomberg reported that Alibaba had spent months attempting to resolve the matter directly with DoD, submitting evidence and a formal written reply, without response.

June 26 — the 52-week low at US$94.81.

July 1 — the DOJ settlement. Non-prosecution agreements under which Alibaba Group and AUS Merchant Services pay a combined US$600 million to resolve allegations of failing to prevent illegal sales of pharmaceuticals, controlled substances, listed chemicals and pill presses into the United States. Alibaba admitted its platforms processed roughly 80,000 illegal pharmaceutical transactions across 2016–2024 with combined merchandise value over US$200M; DOJ conducted more than 40 undercover purchases; prosecutors stated that Alibaba employees had internally raised concerns that compliance controls were insufficient. The cash cost is immaterial and the market read it as risk removal, which it was. The residue is not immaterial: an eight-year record of admitted facts and internal compliance warnings is now permanently available to any U.S. agency building an adverse case, at a moment when U.S. agencies are demonstrably interested.

July 8 — the +11% session. 38.3 million shares traded, roughly 3.3× the 90-day average of 11.5 million. The catalyst cluster reported at the time: the DOJ resolution, a reported temporary court stay permitting Alibaba to continue U.S. lobbying, press reports that instant-commerce losses had narrowed in the June quarter, and the H200 licensing news. The lobbying-stay report rests on a single trade-press source with no docket citation and is treated here as low-confidence.

July 15–16 — Apple Intelligence approved in China with Qwen. Covered in the competitive-position section above.

July 22 — the last disclosed repurchase, at US$13.61 per ordinary share (~US$109/ADS).

August 3 — Qwen3.8-Max, 2.4 trillion parameters. ADS +4.1%.

August 3–7 — the securities class actions. At least ten plaintiffs’ firms announced actions or investigations. The lead case is captioned Wistisen v. Alibaba; the class period runs June 26, 2025 through June 24, 2026; the lead-plaintiff deadline is October 5, 2026. The core allegation is that Alibaba failed to disclose that it was considered a Chinese military company and was at risk from the U.S. crackdown. Assessment: the volume of filings reflects the size of the drawdown, not the merit of the claim. The theory is weak on its face — Alibaba’s 20-F has disclosed U.S.-listing, sanctions and designation risk for years, and the claim requires proving the company knew of a DoD determination it says it was never informed of. Likelihood of a material adverse judgment: low. Nuisance value, headline flow through October, and discovery risk: real.

August 6 — Qwen open-weight monetization. Reuters reported the plan to charge large commercial users of the next open-source release a share of their revenue.

August 7 — earnings date set for August 20, 2026, before the U.S. open. (One third-party source circulated an August 28 date; the company’s own announcement says August 20.)

Verdict: the changes in this window strengthen the operating thesis and weaken the ownership claim, and the second effect is larger. Three of the developments — the subsidy rules, the Apple design win, the buyback restart — are genuine improvements to the fundamentals or to shareholder alignment, and together they address two of the four bull tests the prior article set. But the 1260H designation, the federal litigation against the U.S. government, the Senate-directed IP allegation, and the class action all attack the same thing: the security’s standing as an ownable U.S.-listed claim. A business can be improving while the wrapper around it deteriorates, and that is precisely what happened between June 7 and August 7.

9. Risk Analysis

The risk matrix from the prior article is re-scored here, with the changed rows marked. Likelihood and impact are over a 12–24 month horizon.

Risk 1 deserves its own paragraph, because it is the one that changed the shape of the distribution. The legal anatomy: Section 1260H of the FY2021 NDAA directs the Department of Defense to identify Chinese military companies operating in the United States. Its direct consequences are procurement-side — designated firms are barred from direct DoD contracts from June 2026, with indirect procurement restrictions from 2027 — and, critically, it imposes no restriction on U.S. persons purchasing, holding or selling the securities. The regime that does impose that restriction is Treasury’s Non-SDN Chinese Military-Industrial Complex Companies list under Executive Order 14032, under which U.S. persons are generally prohibited from purchasing or selling the covered securities or derivatives providing exposure to them, and under which pre-designation holdings may be held but not readily sold absent authorization. Alibaba is on 1260H and is not on NS-CMIC. Trade counsel commentary is consistent that 1260H designation raises the risk of subsequent NS-CMIC designation without triggering it.

So the correct characterisation is neither “nothing happened” nor “the risk materialised.” What happened is that a low-probability, severe-impact tail moved from ambient and unobservable to active and observable, with a docket attached — Alibaba’s own suit against DoD in San Jose. It also means the risk is now, unusually, trackable: an investor can monitor the litigation and Treasury’s actions rather than guessing at geopolitical mood. That is worth something. What it does not do is make the tail thinner.

Catastrophic-loss assessment. The probability of a permanent total loss remains very low: the business is profitable, unlevered on a net basis, and generates enormous gross cash. The realistic catastrophic scenario is not bankruptcy but impairment of the claim — an NS-CMIC designation, a forced ADR delisting, or a VIE reinterpretation — in which the underlying business continues to exist and prosper while U.S. holders find their instrument illiquid, untradeable or structurally severed from the economics. That is the specific risk this security carries and no other large-cap technology company does, and it is the reason position sizing rather than valuation is the operative discipline here.

10. Valuation Discussion — Embedded Expectations

No price target and no recommendation appear in this section or anywhere in this analysis. The analysis below establishes what the current price requires one to believe.

The capitalisation, built from primary sources. Ordinary shares outstanding were 19,174,988,918 at July 31, 2026 per the Hong Kong monthly return furnished on Form 6-K, equal to 2,396,873,615 ADS-equivalents at the 8:1 ratio. At the August 6 close of US$126.81 that gives a market capitalisation of approximately US$304B, against roughly US$290B at the prior article’s US$121.06 reference — a 4.8% increase driven entirely by price, partially offset by the 31.3 million ordinary shares cancelled.

A data-quality note that materially affects the arithmetic. ROIC.ai’s headline market_cap and enterprise_value for BABA read approximately 2.33 trillion and 2.34 trillion — these are Renminbi figures mislabelled as USD, the known ADR unit-garble for this security. Its diluted_mkt_cap (US$301.6B) and ev_to_sh_out (125.70) are USD-consistent. Separately, its enterprise-value calculation nets only “cash and near-cash” of US$19.1B and ignores short-term investments, producing an enterprise value above market capitalisation for a company with roughly US$37.8B of net cash. All aggregator EV and capitalisation figures for this ADR were discarded. Capitalisation here is built from the filed share count and the exchange close; the cash and investment stack is taken from the FY2026 20-F.

The sum-of-the-parts arithmetic, re-based. Carrying forward the balance-sheet work established in the prior article from the FY2026 20-F — gross cash and liquid investments of RMB520.8B (US$75.5B), net cash of roughly US$37.8B, the 33% Ant Group stake, and an equity-securities portfolio carried at RMB449.9B — and adjusting for the US$600M DOJ payment and approximately US$384M of repurchases, the non-operating asset block remains roughly US$80–105B. The implied market value of all operating segments is therefore:

The prior article’s central observation survives: a no-growth earnings-power value for the operating businesses of roughly US$150–200B, plus the US$80–105B asset block, still brackets the market capitalisation, so the cloud and AI growth remains lightly priced. But the cushion has compressed sharply. At the June 26 low, the operating stub of US$122–147B sat below the low end of the EPV range — the market was pricing the operating businesses at less than their no-growth worth, which is the condition Greenwald’s framework treats as the strongest form of value. At US$126.81 the stub of US$200–225B sits above the EPV range’s midpoint. The mispricing that existed six weeks ago has been substantially arbitraged away by the July rally, without a single new financial statement being published.

What the current price requires you to believe. Decompose the US$200–225B operating stub against the FY2026 segment facts. China E-commerce produced RMB107,509M (~US$15.6B) of adjusted EBITA in a trough year, down 44%, against a prior run-rate above US$25B. Cloud produced RMB14,265M (~US$2.1B) on RMB158,132M of revenue. AIDC and All Others lost a combined ~RMB37.8B. Corporate costs consume the rest.

• If China E-commerce merely holds its FY2026 trough EBITA of ~US$15.6B in perpetuity with no growth, capitalised at a China-discounted 10–12× on a mid-teens cost of equity, it is worth roughly US$155–190B on its own — that is, essentially the entire operating stub at the low, and roughly 75–90% of it today. On that arithmetic, the buyer at US$126.81 is paying approximately US$10–70B for everything else: the whole cloud franchise, AIDC, the Qwen model estate, and the option on EBITA recovery.
• If instead China E-commerce recovers to a US$25B run-rate — the prior article’s bull test 1 — the same multiple gives US$250–300B for the core alone, which exceeds the entire current stub. On that path, the current price implies negative value for cloud.
• Conversely, if the trough is the new normal and still falling, and China-commerce EBITA settles at US$12B while cloud takes another three years to reach meaningful margin, US$200–225B for the operating businesses is roughly fair rather than cheap.

So the embedded expectation at US$126.81 is approximately: China-commerce EBITA stabilises near or modestly above its trough, cloud is worth something but not much, and the jurisdictional discount stays where it is. That is not a demanding set of assumptions. It is also not the deeply pessimistic set the market was underwriting at US$94.81 six weeks ago.

Own-history multiples, reported with their staleness. The AZI own-history percentile framework — which ranks the current multiple against roughly a decade of the stock’s own range — reads a composite percentile of 23.7 (P/E 21.4, P/B 25.5, P/S 24.2), with a P/E of 18.35, P/B of 1.76 and P/S of 1.85 on TTM EPS of US$6.11, book value per share of US$63.72 and TTM sales per share of US$60.51. That reading is dated July 24, 2026 and was struck at a price of US$112.14 — it is stale to a subsequent 13.1% rally and must not be quoted as the current figure. Mechanically re-scaling to US$126.81 gives P/E ~20.7, P/B ~1.99 and P/S ~2.10, which on the same distribution places the composite in roughly the low-30s percentile — approximately where the prior article found it on June 7. The honest summary: the June sell-off produced a genuinely cheap own-history print in the low-20s percentile, and the July rally has taken most of it back.

Two caveats on that percentile. The P/E component is the least reliable of the three, because GAAP TTM EPS of US$6.11 is flattered by the RMB87.5B of non-operating equity mark-to-market gains identified in the financial-quality section while operating earnings sit at a trough — a distorted numerator over a distorted denominator. P/S at ~2.10 and P/B at ~1.99 are the cleaner own-history reads, and both tell the same story. And the percentile is an own-history measure only; it says nothing about whether Alibaba deserves to trade where it has historically traded, which is precisely the question the 1260H designation reopens.

Cross-sectional comparison, and why it is nearly useless here. The natural comparables — Amazon, Microsoft, Google on the cloud side; MercadoLibre, Sea, Coupang, PDD, JD on the commerce side — trade at multiples that make Alibaba look extraordinarily cheap on every metric. That comparison has been available and unhelpful for five years, because the discount is not an analytical error; it is the market’s price for VIE structure, jurisdictional risk and non-fungible assets. The factor model supplies the empirical confirmation: Alibaba’s factor-nearest securities are China index ETFs (FXI 0.929, KWEB 0.926, MCHI 0.923), with only JD (0.901) and Baidu (0.843) as single-stock comparables. The marginal buyer of BABA is expressing a view on China, not on e-commerce or cloud. Any valuation argument that ignores this is arguing with the wrong counterparty.

The scenario frame. Three coherent paths, with the assumptions stated rather than the outputs emphasised:

Bear. China-commerce EBITA stabilises at the trough or drifts lower as PDD and Douyin continue taking share; the RMB380B AI programme earns below its cost of capital; Beijing’s silicon preference caps cloud growth in the twenties; the jurisdictional discount widens because the 1260H litigation goes badly. The operating stub is worth less than today’s US$200–225B, and the asset floor does the work.

Base. The SAMR rules hold, instant-commerce losses narrow materially through FY2027, and China-commerce EBITA recovers part-way toward the prior run-rate — call it US$18–22B — while cloud compounds in the 30s with the margin grinding from ~9% toward the mid-teens. Cloud becomes visibly worth something. The stub grows into and modestly beyond its current value.

Bull. The ceasefire holds completely, China-commerce EBITA returns toward US$25B+, cloud sustains 35–45% growth with margin expansion validating the capex, the Apple slot and Qwen3.8-Max convert into enterprise share, the DoD litigation is won or the designation quietly lapses, and buybacks return to scale. On that path the operating businesses alone justify substantially more than the current capitalisation.

The distribution of those outcomes is wide, the leftmost tail is jurisdictional rather than operational, and the single largest near-term update to it arrives on August 20.

Verdict: the arithmetic is unchanged and the margin of safety is not. Alibaba remains a net-cash, asset-rich, China-discounted value situation in which the no-growth earnings power plus the asset block brackets the market capitalisation, and in which the cloud franchise is lightly priced. That was compellingly true at US$94.81 and is only modestly true at US$126.81. What the update changes is not the structure of the valuation but the width of the discount one should apply to it: a live military-company designation and active federal litigation against the U.S. government are, on any honest reckoning, worth a wider jurisdictional haircut than was appropriate on June 7 — and the price moved the other way.

11. Variant Perception

The consensus view. Sell-side and financial-media consensus on Alibaba is constructive and has become more so through July. The prevailing frame is: a cheap stock with an under-appreciated AI-cloud franchise, a trough in earnings caused by identifiable and ending discretionary spending, and a China risk that is well known and already discounted. Third-party consensus for the August 20 print clusters around RMB245–253B of revenue (+1% to +4% year over year), cloud growth near 45%, and EPS up roughly 29%. Published sell-side price targets in the window ranged from roughly US$120 on constrained assumptions to US$190 on unconstrained ones. These are third-party estimates and targets, reported as a description of consensus only; they are not this article’s view and no target of any kind is adopted here.

The strongest bull case. Alibaba is being valued as a Chinese e-commerce company in structural decline when it is in fact the operator of Asia’s leading AI-cloud franchise, which the market gets for nearly nothing. The FY2026 earnings collapse was purchased, not suffered — a RMB380B infrastructure programme and a subsidy war that a regulator has now ended by decree. The core marketplace’s gross margin never moved, which is the tell that the moat is intact and the damage was in discretionary spend. Cloud is compounding in the high thirties with AI revenue growing triple-digits for eleven consecutive quarters, and it just won the Apple Intelligence slot in China, which is a distribution position that cannot be bought. Management proved in June that it will buy its own stock aggressively at distressed prices with offshore cash, which both resolves the fungibility question and signals its own view of value. And the balance sheet — US$75.5B of liquid assets, net cash, a 33% Ant stake — means the downside is bounded by arithmetic rather than by hope.

The strongest bear case. The cheapness is earned, and the July rally was a bounce that removed most of what compensation existed for owning it. Taobao/Tmall has lost roughly a fifth of its GMV share in four years to structurally cheaper and structurally better-at-demand-generation rivals, and had to start paying merchants through contra-revenue to keep them — that is a moat eroding, not merely narrowing. FY2026 ROIC was 2.84%, below any plausible cost of capital, and the RMB380B bet that caused it is two years from resolution with a cloud EBITA margin still at 9%. The instant-commerce “ceasefire” is a draft rule that regulators have announced before, and the competitive energy has already redirected into balance-sheet-heavy M&A that rebuilds the low-ROIC retail exposure the company just spent two years shedding. Beijing, not Washington, now gates access to frontier silicon, which removes the cleanest bull catalyst. And overriding all of it: on June 8 the U.S. government formally designated this company a Chinese military company, Alibaba is now suing the Department of Defense, Anthropic has taken an IP allegation to the Senate Banking Committee, and a securities class action is under way — while the escalation path to Treasury’s NS-CMIC list, which would stop U.S. persons transacting, remains open.

The 3–5 assumptions that actually matter.

1. Is the China-commerce EBITA collapse cyclical or structural? Everything in the valuation turns on whether ~US$15.6B is a trough or a new base. Falsified either way by four to six quarters of segment data beginning August 20.
2. Does cloud margin expand as revenue scales? Growth without margin validates nothing; a 9% EBITA margin on a business consuming RMB126B of annual capex is not yet a franchise, it is a build. Falsified by cloud EBITA margin over the next four quarters.
3. Does the jurisdictional discount widen, hold, or narrow? This is now the dominant variable and it is not analysable by fundamental means. Falsified by observable events: the San Jose docket, any Treasury action, any list movement.
4. Is the asset floor fungible to a foreign holder? Partially answered favourably by the June–July buyback. The remainder — onshore RMB and Ant — is unanswered.
5. Is Alibaba’s frontier-model capability its own? The Anthropic allegation, if substantiated, changes what an investor is paying for in the cloud franchise. Currently an allegation only.

Where I think consensus is offside — and it is not where the bulls think. The variant perception here is not that Alibaba is cheaper than the market believes; the sum-of-the-parts arithmetic is public, widely circulated, and was more compelling six weeks ago than it is now. The variant perception is about who is setting the price and what they are pricing.

The factor evidence is unusually clean on this point. Alibaba carries a negative momentum loading (−0.44 in the sparse model, −0.50 in the full one) and a positive value loading (+0.53) — it is empirically an out-of-favour value name, not a momentum name, despite a +27.4% month. Its dominant explanatory factor is the China country loading at +1.87, roughly four times the magnitude of any style factor, with the full model explaining 65% of variance while a purely U.S.-style model explains 17%. Its factor-nearest securities are China index ETFs. And the realised record of holding it is dismal: a ten-year annualized return of 4.95% with an 80% maximum drawdown and a Sharpe ratio of 0.068, and a five-year annualized return of −7.14%. Idiosyncratic volatility runs at 30% annualized against total volatility above 45%, meaning roughly two-thirds of the risk is not company-specific.

The implication is uncomfortable for fundamental work and should be stated anyway: an investor can be entirely right about Alibaba’s business and still not get paid, because the marginal buyer is trading a country exposure and the country exposure is what moves the price. That is the real variant perception. The bulls’ error is not their arithmetic; it is the implicit assumption that correct arithmetic will be rewarded through a mechanism — multiple re-rating on fundamental delivery — that has demonstrably failed to operate in this security for five years. The bears’ error is the mirror image: treating an 80% drawdown history as proof of a broken business, when gross margin, cloud growth and the balance sheet all say otherwise.

Which leaves the position that the tape and the fundamentals jointly support: this is a cheap, high-quality-core business inside a wrapper whose value is set by a political process, in a security that has not compensated its holders for its risk over a decade, that has just rallied 27% into its 200-day moving average, and whose principal known risk became measurably more active two months ago. Consensus is offside not on the value but on the compensation — the market has re-priced the operating stub upward by roughly 8% since June while the tail risk on the claim went from dormant to live.

12. Fact vs. Interpretation

13. Open Questions

1. Did China E-commerce adjusted EBITA inflect in the June quarter, and by how much? The single most important unknown. Reports on August 20.
2. Did cloud EBITA margin expand, or did revenue growth come at flat-to-lower margin? Growth without margin does not validate the RMB380B programme.
3. What are the commercial terms of the Apple Intelligence arrangement, and when does it launch? Neither has been disclosed. Without them the design win cannot be valued, only described.
4. Will repurchases resume above US$120? The June–July pattern proves price-sensitivity but leaves open whether management regards current prices as attractive. This is a cleaner read on their view of intrinsic value than any commentary.
5. What is the current remaining buyback authorization, and is it being extended? The prior article established ~US$19B unused; ~US$0.4B has since been consumed, but no updated authorization figure has been published.
6. Will Treasury act on NS-CMIC? Unknowable, but observable. This is the dominant tail.
7. What is the litigation timetable in the San Jose DoD case, and is there any preliminary relief? No public schedule located.
8. Is there an Alibaba response to the Anthropic allegation? None located as of 2026-08-07.
9. Will the SAMR Ten Rules be finalised, when, and with what enforcement mechanism? The draft is the whole basis of the ceasefire’s durability.
10. Does the Pupu Supermarket bid proceed, at what price, and how is it funded? It signals a strategic return to asset-heavy retail.
11. Can an open-weight revenue share actually be enforced? The reported plan to charge large commercial Qwen users is economically sensible and technically dubious.
12. What is Taobao/Tmall’s current GMV share? The figures in circulation are roughly two years old. Any moat conclusion drawn from them carries that staleness.
13. Is there any Ant Group re-IPO or restructuring timetable? Nearly six years after the cancelled IPO, none is public.
14. How much of the US$75.5B liquid stack is genuinely offshore and deployable? The buyback demonstrates the mechanism exists; it does not size it.

14. What Must Be True (Bull and Bear, each with a falsification test)

For the BULL case to be right — the cheapness is a mispricing of a chosen trough and the jurisdictional noise is noise:

1. China E-commerce adjusted EBITA inflects upward from its ~US$15.6B trough as the SAMR rules end the subsidy war, tracking back toward the prior US$25B+ run-rate over four to six quarters.
2. Cloud sustains 30%+ growth with EBITA margin expansion from ~9% toward the mid-teens, converting the RMB380B capex from a cost into a franchise.
3. The CMR core holds share and pricing — no further deterioration in like-for-like growth, and the contra-revenue merchant-subsidy programme does not expand.
4. Buybacks resume at scale and at ordinary prices, not only at distressed ones, demonstrating that the offshore asset floor is both fungible and being used.
5. The 1260H designation is reversed, narrowed, or simply stays non-binding — no Treasury NS-CMIC escalation.

Falsification test: If, four to six quarters after the SAMR rules take effect, China-commerce EBITA is still below roughly US$20B normalized, or cloud growth falls below 20%, or cloud EBITA margin fails to exceed 12% while capex remains at FY2026 levels, then the “chosen trough” thesis is wrong and the cheapness is earned. A single further test dominates all of these: if Treasury adds Alibaba to the NS-CMIC list, the bull case is void regardless of operating results, because the claim itself becomes impaired.

For the BEAR case to be right — the cheapness is a correct discount for a business ceding share inside a wrapper of deteriorating standing:

1. The core profit pool is in secular decline — defending GMV share against PDD and Douyin permanently costs cash, and normalized EBITA never returns to prior levels even with subsidies outlawed.
2. Cloud growth is capped by Beijing’s silicon preference, not liberated by Washington’s licences, and margin stays in single digits through the capex cycle.
3. The jurisdictional discount widens structurally — the 1260H litigation fails, escalation risk persists or materialises, and the market applies a permanently higher required return to a claim it may not be able to exit.
4. The asset floor stays largely inaccessible — the ~US$384M buyback proves the mechanism exists at trivial scale but management never deploys it meaningfully, Ant stays dormant, and onshore cash stays trapped.

Falsification test: If China-commerce EBITA recovers toward US$25B+, and cloud holds 30%+ growth with EBITA margin above the mid-teens through FY2027, and repurchases resume at a multi-billion-dollar annual pace at prices above US$120, and the DoD designation is vacated or allowed to lapse without Treasury escalation, then the structural-impairment thesis is wrong and the discount was a gift. The nearest single observation that would begin to falsify the bear case is the August 20 China E-commerce segment EBITA line.

15. Source Appendix

A full source appendix — primary filings (the FY2026 20-F filed 2026-05-20, the FY2026 earnings 6-K, the June–August 2026 6-K series including the CMC List announcement and the Hong Kong share-movement returns, and the Form 4 corpus), U.S. government and legal sources (the DoD Section 1260H list, the DOJ press release, law-firm client alerts on 1260H versus NS-CMIC), Chinese regulatory sources (SAMR’s draft Ten Rules), quantitative feeds (the price series and own-history valuation index, third-party ratio and enterprise-value data, and a public factor model), and trade and financial press — is reproduced in full as Appendix B below.

Primary sources are the authority for every financial figure. Where a third-party aggregator conflicted with a filing or produced an internally inconsistent result — specifically ROIC.ai’s RMB-mislabelled capitalisation and enterprise value for this ADR — the aggregator figure was discarded and the discrepancy noted in the body. Third-party estimates (GMV share, cloud share, cumulative subsidy-war losses, Ant valuation, consensus estimates) are labelled as such throughout and are never used as the basis for a verdict. Statistical estimates from the factor model are reported as facts about the model’s output, with the interpretive claims drawn from them labelled as interpretation.

Disclosure gaps, stated for the record: no earnings-call transcript exists for the quarter under discussion, because it has not been reported; and the GMV-share data carried forward from the prior article is roughly two years stale, with no update published in this window.

APPENDIX A — Standard Diligence Questionnaire

Companion to the analysis of Alibaba Group Holding Limited (NYSE: BABA · HKEX: 9988) above, dated August 7, 2026. This is an update; the baseline is the June 7, 2026 article. Fact / Interpretation / Assumption labels are applied where they matter. Latest audited financials are FY2026 (ended March 31, 2026); the June quarter reports August 20, 2026. Figures in RMB unless noted; 1 ADS = 8 ordinary shares. Price reference US$126.81 (2026-08-06).

General

What thoughtful questions have other investors asked about this company? The June report catalogued six: whether the FY2026 margin and FCF collapse is cyclical or structural; whether cloud can scale under chip constraints; how much of the US$75.5B cash and the Ant stake is fungible to a foreign ADR holder; whether the VIE/delisting tail is genuinely dormant; whether Taobao/Tmall is permanently ceding GMV share; and why management slashed buybacks ~90% while the stock was cheap and simultaneously issued equity-linked debt.

Two months on, the list has changed in an instructive way. Question six has been partially answered — management restarted buybacks on June 22 and bought hardest at the low. Question four has been overtaken by events: nobody is asking whether the tail is dormant anymore; they are asking whether the Section 1260H designation escalates to Treasury’s NS-CMIC list. And a new question has joined the list that did not exist in June: is Alibaba’s frontier-model capability genuinely its own, following Anthropic’s June 24 allegation of large-scale adversarial distillation. Those are the right questions, and the analysis addresses each.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Low — a deliberately chosen trough, still unconfirmed as a trough. (Fact) FY2026 adjusted EBITA −56%, operating income −64%, free cash flow −RMB46.6B, ROIC 2.84% versus 8.06% the prior year. (Interpretation) The cause is self-inflicted — the instant-commerce subsidy war and the RMB380B AI capex — not a demand collapse; gross margin barely moved (39.95% → 39.81%), which locates the damage below the gross line. (Open Question) Whether it is a trough or a new base is unresolved until segment data arrives from August 20 onward.

Driven by the external environment or internal actions? Predominantly internal. The subsidy war and the capex programme are management choices. The external backdrop (soft Chinese consumption, deflationary pressure) amplified rather than caused it. The de-escalation, however, is now external and regulatory: SAMR’s June 17 draft rules ban sustained large-scale subsidies outright, which ends the war by decree rather than by mutual restraint.

How stable are revenues? Group revenue is stable-to-slow-growing: +3% reported, +11% like-for-like after the Sun Art and Intime disposals. The high-margin CMR core (+5% reported, +7% like-for-like) is reasonably stable and quasi-recurring. Quick commerce (+47% to RMB78,520M) is transactional and subsidy-dependent. Cloud (+34%, accelerating to +38% in the March quarter) is the fast-growing component.

Outlook for products/services? Bifurcated and now more so. E-commerce: low-to-mid-single-digit GMV growth in a mature, fragmenting market, with the instant-retail segment about to be tested at something approaching real prices for the first time. Cloud/AI: third-party consensus clusters near 45% June-quarter growth, with the Apple Intelligence design win and the August 3 Qwen3.8-Max release supporting the franchise. International (AIDC): scaling toward breakeven, loss narrowed from RMB15,137M to RMB2,051M.

How big will this market be — growing, shrinking, domestic or international? China e-commerce is the world’s largest but decelerating and structurally more competitive; the profit pool has been actively destroyed by the subsidy war and is now being administratively rebuilt. China public cloud is re-accelerating on AI and is the structurally better market. Revenue is predominantly domestic; AIDC adds an international leg that has never demonstrated franchise economics.

Business Quality & Competitive Moat

Is the industry getting more or less competitive? E-commerce: more, but with the most destructive form of competition now restricted. Barriers fell after the 2021 exclusivity ban; merchants multi-home; PDD’s hard-discount model and Douyin’s content commerce have taken share. What changed in this window is that SAMR has moved to prohibit the specific competitive weapon — sustained large-scale subsidy — that was destroying the profit pool. Cloud: consolidated and oligopolistic, with Alibaba at roughly 36–40% share ahead of Huawei and Tencent, though Beijing’s preference for domestic silicon strengthens Huawei’s structural position.

How profitable is the business (ROIC, ROE)? (Fact) FY2026 ROIC 2.84%, ROE 10.25%, operating margin 5.83% — a collapse from FY2025’s 8.06% / 13.00% / 14.76%. (Interpretation) ROIC below 3% is beneath any defensible cost of capital for a China-domiciled VIE equity, so as reported, FY2026 destroyed economic value. That said, the consolidated figures materially understate the core: the balance sheet carries ~RMB1.4T of assets including US$75.5B of liquid investments, a low-yield Ant stake and goodwill, while the Taobao/Tmall CMR engine earns very high incremental returns on very little incremental capital.

How profitable is the industry — how many competitors, what barriers to entry? E-commerce: low barriers, multi-homing merchants, chronic price competition, compressed profit pools; four material competitors (PDD, Douyin, JD, Meituan in instant retail). Cloud: high fixed-cost barriers, real switching costs from migrated workloads and data, three-player concentration — structurally more profitable, though Alibaba’s cloud EBITA margin is still only ~9% because it is mid-build.

Can the business be easily understood? Mostly. A take-rate marketplace plus an IaaS/PaaS/MaaS cloud plus international retail plus a long tail of loss-makers. The genuine complications are the VIE structure, the equity-investment portfolio that injects large mark-to-market swings into GAAP earnings (RMB87.5B in FY2026), the contra-revenue reclassification of merchant subsidies, and the RMB-versus-USD unit conventions that garble aggregator data for this ADR.

Can it be undermined by foreign low-cost labor? Not directly — this is a domestic platform business. The relevant analogue is domestic low-cost disruption, and it has already happened: PDD’s manufacturer-direct model undermined Taobao/Tmall’s share position.

Do brands matter? Yes. Taobao, Tmall and the 88VIP membership (>62 million, growing double digits) are powerful consumer brands with a genuinely captive high-value cohort. Alibaba Cloud and Qwen are now strong enterprise and AI brands, and the Apple Intelligence selection is a substantial brand validation. But brand has not prevented roughly a fifth of GMV share migrating to cheaper and more discovery-driven rivals.

What is the nature of competition? Price and selection in e-commerce; subsidy and delivery density in instant retail (now regulated, and shifting to supply-chain M&A — the reported ~US$1.5B Pupu Supermarket bid); scale, performance and model differentiation in cloud; price and logistics internationally.

Customers’ switching costs? Moderate and falling on the merchant side — multi-homing is standard, and the FY2026 shift of merchant subsidies into contra-revenue is direct evidence that Alibaba is paying to retain spend rather than commanding it. Moderate-to-high on the cloud side (embedded workloads, migrated data, applications built on Model Studio/Qwen). High for the 88VIP consumer cohort. The Apple Intelligence position is the highest-switching-cost asset the company has added in years, because replacing an on-device model in a shipped operating system across an installed base is not casually revisited.

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? Yes. The 33% Ant Group stake, carried within the RMB206.8B equity-method line and contributing only RMB375M of profit in the March quarter (recognised a quarter in arrears), plausibly carries embedded value well above its carrying basis — though with no re-IPO or restructuring catalyst nearly six years after the cancelled 2020 IPO, that value is unpriced optionality with no timetable. The equity-securities portfolio is carried at RMB449.9B. Brand, the data asset and the Qwen model estate are not capitalised.

Off-balance-sheet liabilities? The principal exposure is structural rather than financial: the VIE contractual arrangements through which the Cayman holding company controls the PRC operating entities, plus related guarantees and commitments. Add the remaining ~two years of the RMB380B capex commitment, and disclosed operating leases and purchase commitments. New in this window: the contingent legal exposures — the securities class action (Wistisen v. Alibaba), the Anthropic allegation, and any downstream consequences of the DOJ admitted-facts record — none of which is currently quantifiable and all of which are assessed in the article as low-probability for material financial liability.

How conservative is the accounting? Mixed, and unchanged. Conservative where it costs: goodwill impairments are taken rather than avoided, and share-based compensation is modest at ~1.4% of revenue with the diluted share count actually falling. Less helpful elsewhere: GAAP net income of RMB102,127M is flattered by RMB87.5B of non-operating equity mark-to-market gains, so GAAP EPS materially overstates operating reality — non-GAAP net income fell 62% against GAAP’s 19%. The contra-revenue reclassification of merchant subsidies optically suppresses reported CMR growth (Q4 +1% reported versus +8% like-for-like), which is transparent but requires the reader to do work.

How CapEx-hungry is the business? Historically asset-light; currently and deliberately capital-intensive. Capex rose 47% to RMB126B within the RMB380B three-year AI and cloud programme; property and equipment rose 39% to RMB282.7B. This is the swing factor that turned free cash flow negative, it has roughly two years to run, and the return on it is the single largest unresolved question in the equity.

Capital Allocation & Management

How much FCF does the business generate, how does management use it, what is the philosophy? Historically a strong generator (+RMB73.9B in FY2025); in FY2026 a −RMB46.6B outflow as capital was redirected into capex and the subsidy war. The philosophy shifted from returning capital to consuming it. This window supplies the first evidence of a partial shift back: with the war ending by regulation and the stock at a 52-week low, management deployed offshore cash into repurchases.

Significant acquisitions recently? No large completed acquisitions; the multi-year story has been disposals — Sun Art and Intime (offline retail, FY2025) and Trendyol GO (FY2026) — a deliberate and value-accretive retreat from balance-sheet-heavy, low-ROIC retail. New and worth watching: the reported ~US$1.5B bid for Pupu Supermarket, which would partially reverse that retreat as instant-retail competition migrates from subsidy to supply-chain consolidation.

Buying back shares? Yes — restarted, and this is the most important capital-allocation development since the prior article. History: US$12.5B (FY2024) and US$11.9B (FY2025), each retiring ~5.1% of the float; then a collapse to US$1.0B in FY2026 with ~US$19B of authorization unused. Alibaba’s own Hong Kong Rule 13.25A returns, furnished on Form 6-K, show open-market NYSE repurchases resuming June 22, 2026 and totalling 31,333,832 ordinary shares (3,916,729 ADS-equivalents, ~US$384M) repurchased and cancelled through about July 22.

The pattern matters more than the amount: ~1.0M ordinary shares/day at the outset, scaling roughly 4× to ~4.2M/day across June 29 – July 2 at the 52-week low, collapsing to 1.47M on July 8 as the stock gapped +11%, 72,800 on July 9, and stopping after ~July 22 at ~US$109/ADS. (Interpretation, high confidence) That is discretionary, price-sensitive behaviour rather than a programmatic drip — precisely what a shareholder should want. (Interpretation) It also partially falsifies the “asset floor is non-fungible to ADR holders” bear case, because offshore cash was demonstrably deployed to buy ADS at the moment of maximum stress. Honest counterweight: US$384M is 0.13% of market capitalisation and ~0.16% of shares outstanding, far below the FY2024–25 pace, and the buying stopped as soon as the price recovered.

Issuing large amounts of new shares to insiders? No. SBC is ~1.4% of revenue and the diluted count fell in FY2026 — good discipline for a large-cap technology company. The dilution that matters is contractual, not compensatory: the July monthly return discloses US$4,999,798,000 of May 2024 convertible notes at a US$12.5222 conversion price (up to 519,056,000 ordinary shares, ~2.7% of the ADS count) and US$3,168,000,000 of September 2025 notes at US$23.9244. At the current ~US$15.85 ordinary-share equivalent the 2024 notes are deep in the money. (Interpretation) Selling conversion options at US$12.52 in 2024 and repurchasing ordinary shares at US$11.84–13.69 in 2026 is not value-destructive, but it is not the sequence a disciplined issuer would have chosen, and it converts the prior article’s flagged inconsistency into a realised cost.

Compensation policy of directors/management? Modest cash plus equity; the annual bonus pool is tied to adjusted pre-tax operating profits, which is a reasonable profit-linked metric. Insider economic alignment remains low — all directors and officers hold roughly 1.9% — with control resting on the Alibaba Partnership’s weighted-voting-rights structure rather than on economics. Shareholders cannot force a different capital plan.

Motivations of management? (Interpretation) The Wu/Tsai team appears focused on long-term AI and cloud leadership and on defending the ecosystem, having genuinely abandoned the prior conglomerate ambition — the disposal record supports this. The June–July buyback adds evidence of shareholder-value orientation that the FY2026 record lacked. The residual concerns are unchanged: whether the RMB380B AI bet earns its cost of capital, and that shareholders have no mechanism to compel a different path under the WVR structure.

Insider transactions? (Fact) Thirteen Form 4s since May 29, 2026, all equity-plan mechanics — annual RSU grants on May 29 (Wu Yongming 448,000 ordinary shares; Jiang Fan 248,000; Yu Siying 120,000), vesting and settlement on the June 25 tranche, and sell-to-cover. Zero discretionary open-market purchases (code P) and zero discretionary open-market sales. A June 29 filing reporting 720,000 ordinary shares sold at “94.92–95.59” is not a discretionary sale and is denominated in Hong Kong dollars per ordinary share, not USD per ADS: the footnotes state the shares were withheld and sold in Hong Kong to satisfy RSU tax withholding — roughly US$8.7M on 90,000 ADS-equivalents. Routine. The meaningful “insider” signal this quarter was the issuer’s own bid.

Valuation & Market Data

Is the stock an ADR, MLP, or K-1 issuer? It is an ADR — 1 ADS = 8 ordinary shares, NYSE-listed, with the primary listing in Hong Kong (9988; RMB counter 89988). Not an MLP; no K-1. Alibaba is a foreign private issuer, so U.S. disclosure runs through Form 20-F and Form 6-K — there is no 10-K, 10-Q or DEF 14A. Practical consequence for research: the most valuable primary disclosure in this window arrived through Hong Kong Listing Rule 13.25A/13.25B share-movement returns filed as 6-K exhibits, not through any U.S.-style filing. Practical consequence for data: aggregator feeds routinely garble RMB and USD units for this security — ROIC.ai reports a market capitalisation and enterprise value of ~2.33–2.34 trillion (RMB mislabelled as USD) and an enterprise value exceeding market capitalisation for a company with ~US$37.8B of net cash. All such figures were discarded and the capitalisation rebuilt from the filed share count.

Dividend policy? A recurring annual dividend of US$1.05/ADS (~US$2.5B), maintained through the FY2026 trough, with a payout ratio of 33.5% on depressed earnings. Trailing yield roughly 0.8%.

How profitable is the business? Covered above. Consolidated FY2026 profitability is poor and deliberately so (ROIC 2.84%); the core marketplace’s incremental economics remain excellent; the gap between the two is the entire investment debate.

Is net income diverging from cash from operations? Yes, materially and in the unfavourable direction. FY2026 GAAP net income was RMB102,127M while free cash flow was −RMB46.6B. The divergence has two sources: RMB87.5B of non-cash equity mark-to-market gains inflating net income, and RMB126B of capex consuming cash. This is precisely the configuration that should make an analyst distrust the GAAP figure — and it flows through to the valuation, because the AZI own-history P/E percentile is computed on a GAAP TTM EPS of US$6.11 that is flattered by those gains. P/S (~2.10) and P/B (~1.99) at the current price are the cleaner own-history reads.

Where does valuation sit against its own history? The AZI composite own-history percentile reads 23.7 (P/E 21.4, P/B 25.5, P/S 24.2) — but that reading is dated July 24, 2026 at a price of US$112.14 and is stale to a subsequent 13.1% rally. Re-scaled to US$126.81, the multiples place the composite in roughly the low-30s percentile, about where the prior article found it on June 7. The June sell-off produced a genuinely cheap own-history print; July took most of it back.

Positioning and factor data. (Fact, third-party statistical estimates) Momentum loading −0.44 (sparse model) / −0.50 (full model); Value +0.53; China country factor +1.87 — roughly four times any style loading, with the full model explaining 65% of variance versus 17% for a purely U.S.-style model. Beta 0.909, alpha −0.069, idiosyncratic volatility 30.1% annualized. Relative strength: 6-month −21.8, 12-month +9.34, versus peak −57.54. Risk-adjusted record: 10-year annualized return +4.95% with an 80.1% maximum drawdown and a Sharpe of 0.068; 5-year annualized −7.14%; 3-year +12.33%. Price at US$126.81 sits below the 200-day EMA of US$128.98 after a +27.4% July. Factor-nearest securities are China index ETFs (FXI 0.929, KWEB 0.926, MCHI 0.923), with only JD (0.901) and Baidu (0.843) as single-stock comparables. (Interpretation) The market prices this as a China country exposure, not as an idiosyncratic e-commerce/cloud franchise — which is why fundamental edge here has historically been poorly compensated.

Risks & Downside

What factors would cause the stock to decline? In rough order of importance as this article assesses them: (1) escalation from the DoD Section 1260H list to Treasury’s NS-CMIC list, which would prohibit U.S. persons from transacting in the securities; (2) China-commerce EBITA failing to inflect in the August 20 print and subsequent quarters, proving the trough is a base; (3) cloud growth decelerating or margin failing to expand, invalidating the RMB380B programme; (4) further GMV-share loss to PDD and Douyin; (5) the SAMR draft rules being diluted, delayed or unenforced, restarting the subsidy war; (6) escalation of the Anthropic distillation allegation into litigation or export-control action; (7) continued weakness in Chinese consumption; (8) broad China-risk repricing, to which this security has a +1.87 country loading.

Risk of a catastrophic loss? Yes, but of an unusual kind. The realistic catastrophic scenario is not insolvency — the company is profitable, net-cash, and sits on US$75.5B of liquid assets. It is impairment of the claim: an NS-CMIC designation, a forced ADR delisting, or a VIE reinterpretation, in any of which the underlying business continues to prosper while the U.S.-listed instrument becomes illiquid, untradeable or structurally severed from the economics. That risk is specific to this security and to a small number of Chinese ADRs, and it is the reason position sizing rather than valuation is the operative discipline. The realised drawdown history corroborates the magnitude: −80.1% maximum drawdown over ten years, −68.5% over five.

Chance of a total loss? Very low as a business matter; low but non-zero as a claim matter. A complete write-off would require both a severing of the VIE economics and an inability to realise value through the Hong Kong listing — a conjunction that has not occurred for any major Chinese ADR to date, including through the 2021–22 regulatory crackdown and the HFCAA episode. Note that the Hong Kong primary listing provides a partial structural mitigant that did not exist before Alibaba’s 2024 conversion to primary listing status, since holders can in principle convert ADSs into Hong Kong-listed ordinary shares.

Recent News & Events

Has the business environment changed recently? Yes, in two opposite directions, and this is the core of the update.

Improved: SAMR published draft “Ten Rules for Regulating Food Delivery Platform Subsidies” on June 17, 2026, banning long-term large-scale subsidies and coercive merchant/rider cost-sharing — converting an informal ceasefire into policy over Alibaba’s largest profit pool. The Cyberspace Administration approved Apple Intelligence in China on July 15–16 with Qwen as the language-model partner. Qwen3.8-Max, at 2.4 trillion parameters, shipped August 3. The DOJ closed an eight-year probe on July 1. And management restarted buybacks on June 22.

Deteriorated: The Department of Defense added Alibaba to its Section 1260H “Chinese Military Companies” list on June 8, 2026, expanding the list to 188 entities; Alibaba disclosed it in a 6-K on June 9 and sued the DoD in federal court in San Jose on/about June 24. Anthropic wrote to the Senate Banking Committee on June 24 alleging a 28.8-million-exchange adversarial-distillation campaign run through ~25,000 fraudulent accounts between April 22 and June 5, 2026. Securities class actions followed in early August (Wistisen v. Alibaba; class period June 26, 2025 – June 24, 2026; lead-plaintiff deadline October 5, 2026). And in May 2026 the U.S. approved Alibaba and ~9 other Chinese firms to buy up to 75,000 Nvidia H200s each — with zero delivered, because Beijing steered demand to domestic silicon.

A note on the legal anatomy, because it is widely misreported: Section 1260H is a Department of Defense identification list. Its direct consequences are procurement-side — no direct DoD contracts from June 2026, no indirect component procurement from 2027 — and it imposes no restriction on U.S. persons buying, holding or selling the securities. The list that carries that prohibition is Treasury’s NS-CMIC list under Executive Order 14032. Alibaba is on 1260H and is not on NS-CMIC. Trade counsel is consistent that 1260H designation raises the probability of subsequent NS-CMIC designation without triggering it.

Significant acquisitions? None completed. The reported ~US$1.5B bid for Pupu Supermarket is the item to watch.

Change in accounting policies? No new change in this window. The material change remains the FY2026 reclassification of merchant business-development subsidies from sales-and-marketing expense into contra-revenue against CMR, which suppressed reported Q4 CMR growth to +1% versus +8% like-for-like. Readers of the August 20 print should expect the same treatment and should read like-for-like.

Recent changes — new markets, facilities, management? No management changes disclosed in the window. The annual general meeting notice and proxy materials were furnished on August 6, 2026. Facilities: the RMB380B AI/cloud build-out continues, with property and equipment up 39% to RMB282.7B in FY2026. New market position: the Apple Intelligence slot in China is the most significant new distribution channel added.

What is the next scheduled catalyst? June-quarter (Q1 FY2027) results on August 20, 2026, before the U.S. market opens, with a call at 07:30 ET — announced by the company on August 7. (One third-party source circulated an August 28 date; the company’s own announcement is authoritative.) Third-party consensus clusters around RMB245–253B of revenue and ~45% cloud growth; Alibaba does not guide, and those figures are reported as the market’s expectation rather than as any forecast of this article.

APPENDIX B — Source Appendix

Companion to the analysis of Alibaba Group Holding Limited (NYSE: BABA · HKEX: 9988) above, dated August 7, 2026. All URLs accessed 2026-08-07 unless otherwise stated. This is an update; sources marked [BASELINE] were established in the June 7, 2026 article and are carried forward, while sources marked [NEW] were gathered in this window. All sources listed are public.

A. Primary — Company filings (SEC EDGAR, CIK 0001577552)

Alibaba is a foreign private issuer: its U.S. disclosure runs through Form 20-F and Form 6-K. There is no 10-K, no 10-Q and no DEF 14A. Insider disclosure is via Forms 3/4/5.

FY2026 earnings release / 6-K segment tables — the source of all FY2026 segment revenue and adjusted-EBITA figures used in the business-overview, growth, financial-quality and valuation sections. [BASELINE]

Filing index retrieved via scripts/edgar.sh since BABA 2026-05-01; documents fetched directly from https://www.sec.gov/Archives/edgar/data/1577552/.

B. Primary — U.S. government and legal sources

Law-firm client alerts (used for the 1260H-versus-NS-CMIC legal distinction in the risk analysis and Appendix A):

• WilmerHale, “Pentagon Adds 65 New Entities to the 1260H List of Chinese Military Companies,” 2026-06-11 — https://www.wilmerhale.com/en/insights/client-alerts/20260611-pentagon-adds-65-new-entities-to-the-1260h-list-of-chinese-military-companies [NEW]
• Morrison Foerster, “DoD Expands the Section 1260H ‘Chinese Military Companies’ List: Key Changes, New Designations, and What They Mean for Your Business,” 2026-06-26 — https://www.mofo.com/resources/insights/260626-dod-expands-the-section-1260h-chinese-military-companies-list [NEW]
• Cleary Foreign Investment and International Trade Watch, “Department of Defense Adds Major Technology, Automotive, and Biotechnology Firms to 1260H Chinese Military Companies List,” June 2026 — https://www.clearytradewatch.com/2026/06/department-of-defense-adds-major-technology-automotive-and-biotechnology-firms-to-1260h-chinese-military-companies-list/ [NEW]
• Holland & Knight, “Department of War Updates Section 1260H Chinese Military Companies List,” July 2026 [NEW]

C. Primary — Chinese regulatory sources

Reported via: SCMP, “China moves to end ‘irrational’ food-delivery subsidies and sector’s price wars” — https://www.scmp.com/tech/article/3357458/ ; Inside Retail Asia, 2026-06-18 — https://insideretail.asia/2026/06/18/irrational-competition-chinese-government-addresses-food-delivery-wars/

D. Quantitative data feeds

E. Financial and trade press (validation layer)

News triage was run through the ROIC.ai MCP get_company_news tool (identifier BABA, 50 items, from 2026-05-25). Every material item was validated against the underlying article or filing and is cited to the original below, not to the feed.

The Section 1260H designation and Alibaba’s litigation

• CNBC, “Pentagon expands list of China military-linked firms to include Alibaba, Baidu in fresh blow to diplomatic thaw,” 2026-06-09 — https://www.cnbc.com/2026/06/09/alibaba-baidu-byd-named-on-pentagons-china-military-list-.html
• SCMP, “Alibaba sues Pentagon over China military blacklist” — https://www.scmp.com/news/us/article/3358113/alibaba-sues-pentagon-over-china-military-blacklist
• Reuters via Yahoo Finance, “Alibaba sues Pentagon to remove Chinese military company label,” 2026-06-24/25 — https://finance.yahoo.com/technology/articles/alibaba-sues-pentagon-remove-chinese-134204733.html
• Quartz, “Alibaba sues Pentagon to remove Chinese military blacklist,” 2026-06-24 — https://qz.com/alibaba-sues-pentagon-chinese-military-blacklist-062426
• Bloomsbury Intelligence and Security Institute, “Alibaba Challenges Pentagon Over Chinese Military Companies List Designation” — https://bisi.org.uk/reports/alibaba-challenges-pentagon-over-chinese-military-companies-list-designation

The Anthropic distillation allegation

• CNBC, “Anthropic accuses Alibaba of campaign to ‘brazenly’ and ‘illicitly’ extract AI capabilities,” 2026-06-24 — https://www.cnbc.com/2026/06/24/anthropic-alibaba-distillation-campaign.html
• Bloomberg, “Anthropic Accuses Alibaba of ‘Illicitly’ Accessing AI Models,” 2026-06-24 — https://www.bloomberg.com/news/articles/2026-06-24/anthropic-accuses-alibaba-of-illicitly-accessing-its-ai-models

The DOJ settlement

• Al Jazeera, “Alibaba to pay $600m to settle illegal drug sales allegations in US probe,” 2026-07-02 — https://www.aljazeera.com/news/2026/7/2/alibaba-to-pay-600mn-to-settle-illegal-drug-sales-allegations-in-us-probe
• Fox Business, “DOJ says Alibaba failed to stop illegal pharmaceuticals and banned goods from reaching US buyers”

Cloud, AI and the Apple design win

• TechCrunch, “Apple Intelligence approved for launch in China with Alibaba’s Qwen AI,” 2026-07-16 — https://techcrunch.com/2026/07/16/apple-intelligence-approved-for-launch-in-china-with-alibabas-qwen-ai/
• Engadget, “Apple Intelligence finally gets regulatory approval in China” — https://www.engadget.com/2215606/apple-intelligence-finally-gets-regulatory-approval-in-china/
• Reuters, “Alibaba unveils its most capable AI model to date, not far behind Moonshot’s in size,” 2026-08-02/03 — https://www.reuters.com/business/retail-consumer/alibaba-unveils-its-most-capable-ai-model-date-not-far-behind-moonshots-size-2026-08-03/
• CNBC, “Alibaba shares rally after unveiling its ‘most powerful’ AI model as U.S.-China competition heats up,” 2026-08-03 — https://www.cnbc.com/2026/08/03/alibaba-ai-model-qwen-rival-anthropic.html
• Forbes, “Alibaba Unveils Qwen3.8-Max Model—China’s Latest AI Challenger To OpenAI And Anthropic,” 2026-08-03
• Reuters, “Alibaba plans to charge big users of its next open-source AI model, sources say,” 2026-08-06 — https://www.reuters.com/business/retail-consumer/alibaba-plans-charge-big-users-its-next-open-source-ai-model-sources-say-2026-08-07/

Chip access

• Benzinga / Yahoo Finance, “Trump Clears Nvidia H200 Sales To Alibaba, Tencent And 8 Others, But Beijing Halts Deliveries,” 2026-05-14
• Cybernews, “US approves Nvidia H200 sales to Chinese firms but no chips delivered, Beijing unhappy with conditions”
• TechRepublic, “US Approves Nvidia H200 Sales to China, But Shipments Remain Stalled”

Instant commerce / subsidy war

• BigGo Finance, “China’s Food Delivery Subsidy Crackdown Ends Billion-Dollar Price War, Industry Pivots to M&A and Compliance” — https://finance.biggo.com/news/b5a03f62-9591-4c14-9023-858fede6755f (source of the ~RMB87B Alibaba / RMB46.6B JD loss estimates and the ~US$1.5B Pupu Supermarket bid — third-party estimates, no filing support, not used in any valuation work)
• Momentum Works, “Chinese regulators calling an end to the food delivery war again. It might not end.” — https://thelowdown.momentum.asia/chinese-regulators-calling-an-end-to-the-food-delivery-war-again-it-might-not-end/ (the counter-argument on durability)

Securities class actions

• PRNewswire (Robbins Geller Rudman & Dowd), 2026-08-04 and 2026-08-06 — class period 2025-06-26 to 2026-06-24; lead-plaintiff deadline 2026-10-05; case captioned Wistisen v. Alibaba
• GlobeNewswire (Rosen Law Firm; Bragar Eagel & Squire), Businesswire (Kirby McInerney; Glancy Prongay; Frank R. Cruz; DJS Law Group; SBS Law), Newsfile (Kaplan Fox), 2026-07-31 to 2026-08-07

Earnings date and consensus

• Businesswire, “Alibaba Group Will Announce June Quarter 2026 Results on August 20, 2026,” 2026-08-07 — https://www.businesswire.com/news/home/20260807747217/en/ (authoritative; supersedes a third-party “August 28” citation)
• itiger.com, “Alibaba Q1 Earnings Preview: AI Boosts Cloud Growth, Flash Sales Spending Pressures Profits” — https://www.itiger.com/hant/news/1164600473 (third-party consensus aggregation: RMB245–253B revenue, ~45% cloud growth, ~29% EPS growth — market expectation, not guidance; Alibaba does not guide)
• tradingkey.com, “Alibaba (BABA) Stock Forecast July 2026: Four Catalysts Drove 12% in Two Days” (source of the July 8 catalyst cluster including the reported U.S. district court stay on a Pentagon lobbying restriction — single-source, no docket citation, treated as LOW-CONFIDENCE and excluded from all verdicts)
• Motley Fool, “Why Shares of Alibaba Soared 27.4% In July,” 2026-08-06 — https://www.fool.com/investing/2026/08/06/why-shares-of-alibaba-soared-274-in-july/

F. Prior coverage

• Prior article on Alibaba, dated 2026-06-07. The baseline thesis for this update. Source of: the FY2026 20-F balance-sheet build (gross cash and liquid investments RMB520.8B / US$75.5B; net cash ~US$37.8B; equity-method line RMB206.8B including the 33% Ant stake; equity securities RMB449.9B), the US$80–105B non-operating asset block, the US$150–200B no-growth EPV range, the FY2024–FY2026 buyback history and the ~US$19B unused authorization, the GMV-share and cloud-share figures, and the four bull and four bear falsification tests re-scored here. Every figure in that build traces to the FY2026 20-F filed 2026-05-20 (SEC accession 0001193125-26-231755), which is the underlying public source.

G. Disclosure gaps and data-quality flags

Stated for the record, and repeated in the body of this article where they bear on a conclusion:

1. Industry framing carries forward from the prior article plus fresh public regulatory sources; no new third-party industry primer was consulted for this update.
2. No earnings-call transcript for the period under discussion. The June quarter has not been reported; the last call (FY2026 Q4) was already incorporated in the baseline report.
3. No new reported financial statements in this window. The FY2026 20-F filed 2026-05-20 remains the latest audited data. Every forward figure in this report is expectation, not data.
4. GMV and market-share data is roughly two years stale. Taobao/Tmall ~40–44%, PDD ~19%, Douyin ~US$650B GMV are third-party estimates carried from the prior article; no update was published in this window, and any moat conclusion drawn from them inherits that staleness.
5. ROIC.ai capitalisation and enterprise value for this ADR are unit-garbled and were discarded; capitalisation was rebuilt from the filed Hong Kong share count and the exchange close.
6. The AZI own-history percentile is stale to the rally (struck 2026-07-24 at US$112.14) and is reported with its as-of date and a re-scaled current estimate rather than quoted raw.
7. The reported U.S. district court stay on a Pentagon lobbying restriction is unverified against a court docket and is excluded from every verdict.
8. Cumulative subsidy-war loss figures are third-party estimates with no filing support and are not used in any valuation work.
9. The Anthropic allegation is an allegation. No litigation, no agency action, no admission, and no located Alibaba rebuttal as of 2026-08-07.