Roblox Corporation (NYSE: RBLX) — The Float Runs Out and the Guidance Goes Dark

Report date: 2026-08-22. As-of price: $38.40 (close 2026-08-21). A follow-up to my 14 June 2026 write-up on the same company.

⚡ Claude’s Take

This block is the author’s own subjective opinion. It is general information, not investment advice, and it is the only part of this article that takes a position. The analysis that follows is deliberately position-free and carries no price target.

Verdict: DOWNGRADE to AVOID-here / hold-if-owned — not a short. This is a real franchise that just failed its own bull test, and $38 does not yet pay you for a business with no guidance, ~9% cash operating margins and $1.1B a year of stock compensation. Constructive only below roughly $28–32 (EV ~$15–18B, ~2.3–2.7x forward bookings). Conviction: medium-high on “don’t buy yet,” low on direction.

Ten weeks ago I wrote that Roblox’s 2026 air pocket was self-inflicted, transitional, and probably worth accumulating in the high-$30s–low-$40s. I set an explicit falsification test: “If reported DAUs fail to grow sequentially in Q3 2026, or management cuts the bookings guide again, the ‘transitional’ thesis is broken.” On July 30 management did something worse than cut the guide — they withdrew the full-year outlook entirely and guided Q3 bookings to a 14–18% year-over-year decline. The stock fell 26.9% the next session, the worst single day in its listed history. I was wrong, and I was wrong about the mechanism, which matters more than being wrong about the price. My June thesis rested on the claim that “engagement, monetization and retention all held — the damage is entirely top-of-funnel.” Engagement and retention did hold. Monetization did not. Bookings per hour rolled over in the U.S. under-13 cohort because Roblox deliberately re-weighted its recommendation algorithm toward 28-day retention and away from the 2025-vintage viral hits that were carrying the spend. That is a self-inflicted wound too — but it is a wound to the part of the model I said was intact.

What the market may still be mispricing is the balance sheet and the base: $5.08B of net cash (18.5% of the market capitalisation), 123M DAUs, 29B quarterly hours, Japan and India DAUs compounding at 67% and 64%, and a U.S. over-18 cohort growing 32% and monetising 50% better than the kids. The stock sits at the 0.26th percentile of its own ten-year price/sales range — literally the cheapest it has ever been on sales. What the market is probably pricing correctly is the cash economics. Strip the deferred-revenue float — which supplied ~83% of trailing operating cash flow and is deflating exactly on schedule — and Roblox generates roughly $560M a year of cash operating profit before $1.1B of stock comp, against a $22.4B enterprise value. That is ~40x pre-SBC and negative after it. The “3x EV/bookings” headline is only cheap if you assume cash margins roughly double from here; that assumption is the whole investment. The framing is a falling knife with a fortress balance sheet, not a compounder-at-a-price — and the factor evidence agrees: 12-month relative strength −67.6%, a −74.9% one-year drawdown, alpha −0.17, 62% idiosyncratic volatility, and a factor-similar peer list (ARKK, PLTR, C3.ai, Lemonade, ARKW) that contains not one game company. The tape is not treating this as an interactive-entertainment name; it is treating it as a busted speculative-growth position.

Conviction and triggers. Medium-high that the right action is to wait, low conviction on direction from here. The single piece of evidence that flips me bullish: two consecutive quarters of sequential bookings growth with bookings-per-hour rising in U.S. under-18 — i.e. management’s “the retention curves cross” claim shows up in the printed number rather than in unshared internal A/B tests. The single piece that flips me outright bearish: Q4 2026 bookings below ~$1.7B (against a $2,223M comp) or a state-AG settlement that imposes structural limits on under-13 monetisation rather than cash payments. Tag: “They took away the dashboard, then asked you to trust the driver.”

📈 Stock Price Action — Five-Year Event Map

Roblox has made a full round trip and then some. From a $23.19 low in May 2022, the stock compounded more than six-fold to an all-time high of $141.56 on 2025-09-29, then collapsed 72.9% to $38.40 on 2026-08-21. The 52-week range is $35.55 (2026-08-12) to $141.56. Note a factual correction to my June piece: the all-time high is the September 2025 print, not the November 2021 peak of ~$134.72 — the 2025 melt-up exceeded the meme-era high before giving all of it back and more.

Cycle narrative. (1) The 2021–22 collapse was a multiple event, not a business event — bookings never declined. (2) The 2022–23 recovery tracked a genuine re-acceleration in bookings and the first evidence of operating leverage. (3) The May 2024 session is the template for what happened again in 2026: a single guide-down repricing the whole growth curve. (4) The 2024–25 melt-up was the real thing — bookings +55% to $6.79B and DAUs to 144M — but it was disproportionately carried by a small number of extraordinarily high-monetising viral titles aimed at younger users, a dependence the company has since dismantled on purpose. (5)–(6) The de-rating began before the numbers broke, as investors handicapped the January 2026 age-check mandate. (7) The May 1 session priced the first guide-down. (8) The June rally was, in hindsight, a sentiment bounce on capital-return news that changed nothing operationally. (9) July 31 is the event that matters: the shareholder letter guided Q3 bookings to $1,576–1,653M against a $1,922M comparable and declined to guide the year, and the market repriced the entire forward curve in a single session on the withdrawal of information rather than on the magnitude of the miss. (10) August has been a low-volume stabilisation with no new disclosure.

Price moves are FACT (AZI adjusted daily closes, accessed 2026-08-22). Attributed drivers are INTERPRETATION, cross-referenced to earnings 8-Ks and the filing index. No recommendation, price target, or technical level appears in this section.

🔄 Changes Since the 14 June 2026 Write-Up

The June piece was written between the Q1 print and the Q2 print, at $43.31. Everything below is new.

Grading the June write-up’s falsification tests:

Net: the bull case failed its own test on the specific trigger I nominated, and the bear case was confirmed on the specific mechanism the bear nominated. That is about as clean a verdict as an update ever produces, and it obliges a downgrade.

1. Executive Summary

Roblox operates the world’s largest user-generated-content gaming platform: a vertically integrated stack (engine, cloud, creator tools, virtual economy) on which roughly 20 million creators build 3D experiences that 123 million daily users play, monetised through a single virtual currency, Robux. The franchise is real. Q2 2026 delivered 123M DAUs (+10%), 29 billion hours (+5%), 27 million monthly unique payers (+15%), and international momentum that most consumer-internet companies would envy: Japan DAUs +67%, India +64%, hours +61% and +59% respectively.

The investment situation has nonetheless deteriorated sharply since the earlier write-up. On July 30 Roblox reported bookings of $1,557M (+8%, the low end of guidance), guided Q3 bookings to a 14–18% year-over-year decline, and withdrew full-year 2026 guidance, framing the change as an acceleration of a previously announced move to quarterly-only guidance. The stock fell 26.9% the following session — its worst day on record. The cause was not the top-of-funnel problem that dominated the April guide-down. It was monetization: bookings per hour fell in the U.S./Canada under-13 cohort as engagement rotated out of 2025-vintage viral hits and as Roblox’s own re-weighted “Recommended For You” algorithm — now optimising over a 28-day retention window instead of seven days — deliberately surfaced stickier, lower-monetising content. Management asserts, on the basis of internal A/B tests it declines to share, that the retention benefit crosses over the monetization cost. That assertion is currently the load-bearing element of the bull case and it is unverifiable from outside.

The quality-of-earnings problem the earlier write-up flagged has arrived on schedule. Roblox recognises revenue over a 27-month estimated paying-user life, so bookings lead revenue and the gap accumulates in deferred revenue. Trailing operating cash flow of $2,100M contains roughly $1,743M of deferred-revenue float build — about 83% of the total. That build is collapsing: the change in deferred revenue was $99M in Q2 2026 against $365M in Q2 2025, and Q3 2026 guided free cash flow is $(60)M to $5M against approximately $443M a year earlier. Roughly 96% of that year-over-year operating-cash-flow decline is the float, not operations. Simultaneously, GAAP revenue is still printing +36% because it is amortising the 2025 bookings boom — which means the reported revenue line is a rear-view mirror that will roll over hard through 2027, and any EV/sales multiple computed today flatters the company.

Regulatory and litigation exposure has changed category, not just magnitude. Ten state attorneys general plus Los Angeles County have filed suit over youth safety and consumer protection; five states have settled; Roblox accrued $91M in the first half of 2026 for settlements and settlement negotiations, added back in Adjusted EBITDA as non-routine. Non-cash settlement commitments — public-service campaigns, law-enforcement liaison headcount, platform changes, liquidated damages — are disclosed but unquantified. A federal securities class action covers purchasers from 2024-10-31 to 2026-04-30. On 2026-08-19 Australia’s eSafety Commissioner secured a court-enforceable undertaking after testing found adult strangers could still contact children despite the new controls — a direct empirical challenge to the “safety infrastructure is becoming a moat” argument.

Valuation, rebuilt by hand: 714.3M shares at $38.40 gives a $27.4B market capitalisation; $6,084M of cash and investments less $1,009M of notes gives $5,075M of net cash; enterprise value $22.35B. That is 3.0x trailing bookings, 3.9x trailing revenue, 3.5x annualised Q3-guided bookings, and 13.6x trailing free cash flow — the last of which is meaningless because trailing FCF is a float artefact. On the cash economics that will persist, Roblox earns roughly $560M a year of operating profit before stock compensation, i.e. ~40x EV, and is loss-making after $1.14B of annual SBC. The stock trades at the 0.26th percentile of its own ten-year price/sales history, its cheapest reading ever. Both facts are true at once, and the tension between them is the investment case.

2. Business Overview

What it is. Roblox is a platform, not a studio. Four integrated layers: Roblox Studio (free creation toolset), the Roblox Client (mobile, PC, console, VR), Roblox Cloud (hosting and services), and a virtual economy denominated in Robux. Users buy Robux for cash, spend them inside experiences, and creators earn a share which qualified developers convert to fiat through the Developer Exchange (DevEx). That closed loop turns a content-hosting site into a professional creator marketplace, and it is the single hardest thing about the business to replicate.

The accounting that governs everything. Because most Robux buy durable virtual items, Roblox does not recognise the cash on receipt. It records bookings (a non-GAAP measure approximating cash collected), parks the unrecognised portion in deferred revenue, and amortises it into GAAP revenue over the estimated average lifetime of a paying user — 27 months in both Q2 2026 and Q2 2025. In Q2 2026, revenue was $1,469M while bookings were $1,557M; the reconciliation is a $99M increase in deferred revenue less $11M of “other.” At 2026-06-30 the deferred-revenue balance stood at $6,904M ($4,596M current, $2,308M long-term) against $9,745M of total assets. Read bookings for demand and revenue for the smoothed, lagged P&L — and never confuse the two, because right now they are pointing in opposite directions: GAAP revenue +36% year-over-year in Q2 while bookings guide to −16% in Q3.

The cost stack, expressed the way it should be — against bookings, not revenue. The shareholder letter provides both cuts, and the difference is instructive. Against revenue, three of four cost lines improved year-over-year. Against bookings — the cash-basis denominator — every one of them got worse:

The revenue-denominated column is flattered by the same amortisation lag that flatters the revenue line itself. On the bookings basis, infrastructure and trust & safety alone consumed 450 additional basis points of every booked dollar in a single year — that is the AI build and the safety apparatus, and it is the cost side of the strategy management is asking investors to fund.

Who uses it and where. 123M average DAUs across 180+ countries in Q2 2026. U.S./Canada DAUs grew only 6% and hours 1%; Japan +67% DAUs / +61% hours; India +64% / +59%. Russia was unblocked late in the quarter, contributing modestly to sequential trends while still representing an approximate three-point year-over-year headwind. On age-checked data, 35% of users are under 13, 38% are 13–17, and 27% are over 18 globally; in the U.S. roughly one-third of age-checked DAUs are over 18, with U.S. over-18 DAUs +32%, hours +27%, and the 18–34 cohort +42% / +37%. Over-18 users in the U.S. monetise more than 50% better than under-18 users. Content has genuinely broadened: top-10 experiences are now about 20% of hours, down from about 30% three years ago, and experiences outside the top 10 grew hours 25% and Robux spend more than 20%.

Verdict. A genuinely differentiated, vertically integrated platform with a recurring-spend model and a real creator marketplace, whose reported financials are systematically hard to read because of a 27-month revenue-recognition lag and $1.1B of annual stock compensation. The business is healthier than the P&L looks and the cash flow is weaker than the cash-flow statement looks — the two errors do not cancel, they compound the confusion.

3. Industry Dynamics

Structure. Roblox competes in interactive entertainment but is structurally distinct from publishers (Take-Two, Electronic Arts, NetEase) and engine licensors (Unity, Epic). Publishers fund and own discrete titles with finite lifecycles and hit economics; Roblox owns the platform and the economy and outsources content supply to a global creator base. Management sizes the addressable market as ~$200B of annual global gaming spend, of which the over-18 segment is roughly 80%, and states that nearly 4% of global gaming revenue already runs through Roblox and roughly 20% of U.S. under-18 gaming revenue. Both figures are management estimates and should be treated as such, but the direction is not in dispute: Roblox has saturated the cohort it started with and must now win a cohort it has never won.

Three structural forces dominate.

First, attention in youth entertainment is zero-sum and faddish. The Q2 result is the cleanest evidence yet. Roblox did not lose users — DAUs grew 10% and retention held — but the specific games carrying spend rotated, and aggregate monetization fell with them. A business whose revenue depends on which titles happen to be viral this quarter has more in common with a hit-driven publisher than the platform framing implies. Management’s own explanation is exactly this: “if you just look at the percentage of hours consumed on the platform last year that was coming from these very large and high monetizing viral hits, it was very concentrated.” That concentration was a 2025 tailwind and is a 2026 headwind, and it will recur.

Second, the storefront tax is structural. Apple and Google extract roughly 30% at the point of sale on mobile, which lands in cost of revenue. Q2 cost of revenue grew 24% to $292M and improved as a percentage of bookings because “the share of bookings from lower-cost platforms grew” — Roblox’s own web/direct-payment push working as intended. This is one of the few genuinely favourable structural trends in the model, and it is worth roughly 190 basis points of revenue margin year-over-year.

Third, and now dominant: child-safety regulation. This has moved from background risk to operating reality. Ten U.S. states plus Los Angeles County are litigating; five states have settled; Australia’s regulator has extracted a court-enforceable undertaking; the EU’s DSA and the UK’s Online Safety Act impose age-assurance duties. Roblox is not merely exposed to this regime — it has chosen to lead it, mandating age-checks for chat globally in January 2026 and launching age-banded Kids and Select accounts in June. The strategic bet is that compliance infrastructure becomes a barrier to entry. The Australian finding on 2026-08-19 — that adult strangers could still contact children despite the measures — is the first hard external evidence that the bet may not pay: if best-in-class safety spending does not actually produce best-in-class safety outcomes, it is a cost, not a moat.

The Marathon capital-cycle lens. Roblox is not capital-intensive in the classic sense, but a version of the cycle applies and has inverted since the earlier write-up. In June the read was favourable: metaverse capital had retreated, competitors had failed, and Roblox compounded as the scaled survivor. Today Roblox is itself the entity attracting and deploying capital into a hot theme — an $845M three-year hosting commitment signed in June 2026, Q3 capex guided to $170M against $24M in Q2, the Morpheus AI acquisition, and an explicit statement from the CFO that AI infrastructure costs are “more of a fixed cost… largely being driven by model training.” Marathon’s warning is precisely about this: high returns attract capital, and capital deployed into a fashionable input at the top of the enthusiasm cycle tends to earn poorly. Roblox is now spending into the most crowded input market in technology.

Verdict — structurally mixed and worse than the prior assessment. The end-market grows and the marketplace model remains more durable than single-title publishing. But (a) the platform’s revenue is more hit-dependent than the “diversified marketplace” framing admits, as Q2 demonstrated; (b) the regulatory environment has moved from tightening to actively adversarial, with an accruing cash cost and an external regulator publicly disputing the efficacy of the company’s safety spend; and © Roblox has joined the AI capital cycle as a price-taker on GPU and hosting capacity. Downgrade the industry verdict from “good structural position for the scaled incumbent, encumbered by regulation” to “good position, materially higher cost of occupancy.”

4. Competitive Position

Name the mechanism. Roblox’s advantage is a two-sided network effect wrapped around a creator-economy flywheel, reinforced by economies of scale in technology. In Greenwald’s taxonomy this is the strongest available configuration — demand-side customer captivity combined with economies of scale — and Roblox genuinely possesses both components:

1. The creator flywheel. More users → more monetizable demand → higher creator earnings → more and better creators → more users. DevEx fees ran $363M in Q2 alone (+15% year-over-year), and the June 2026 rate increase lifted the U.S. over-18 payout materially (management referenced a “DevEx premium of 50% for games consumed over 18”). A new entrant must subsidise both sides for years to conjure this loop. This remains intact.
2. Behavioural switching costs. Avatar identity, owned virtual items, friend graph, and habit. Not contractual, but real — and Q2 proves it: through a global chat-restriction mandate and a discovery overhaul, retention of existing users “remained stable relative to historical periods.” That is the strongest single piece of moat evidence in the quarter.
3. Technology scale. One build runs across phone, tablet, PC and console; Roblox operates its own edge infrastructure; 29 billion hours a quarter of interaction data feeds proprietary models. The Q2 disclosures on AI tooling are concrete rather than aspirational: creator AI adoption among the top 1,000 and top 10,000 creators by Robux spend up ~15 percentage points quarter-over-quarter; a 20% sequential increase in creators using the agentic Studio Assistant; over 60,000 3D assets generated per day; nearly 1,400 games using Cube-generated content daily; and 95% of surveyed AI-using creators reporting faster time-to-launch, with nearly half reporting more than 50% acceleration.
4. Safety infrastructure — the claim that has weakened. The earlier write-up treated this as an emerging fourth moat: build the compliance apparatus regulators will eventually mandate and sub-scale rivals cannot afford. Age-check penetration reached 57% globally (U.S. and Australia above 70%, U18 60% globally and 75% in the U.S.) against a 90% long-term goal, which is real execution. But the Australian regulator’s 2026-08-19 finding cuts directly at the mechanism. A moat built on “our safety works and theirs cannot” requires the first half to be true.

Pressure-testing. The bear’s strongest point after Q2 is no longer “the network effect is soft” — retention data refuted that — it is that the network effect does not confer pricing power. Roblox has 123M daily users, stable retention, and rising payers (27M MUPs, +15%), and it still could not hold monetization flat when the mix of popular content changed. A genuine demand-side moat should let the platform capture value regardless of which experience is in favour. Instead, spend per hour is set by whichever games happen to be trending, and Roblox’s own algorithm change moved it several points in a quarter. That is a marketplace with weak take-rate control, which is a materially different asset from a marketplace with strong customer captivity.

Greenwald’s market-share-stability test. Share of its own users’ time is stable-to-rising and content has genuinely diversified (top-10 concentration 30% → 20% over three years; long-tail hours +25%). Share of the broad gaming market is small (management’s ~4%) and rising. There is still no evidence of share loss to a named competitor — the 2026 damage is entirely self-inflicted. On the share test the moat passes; on the returns test it fails, because a company with a durable competitive advantage should earn excess returns on capital and Roblox earns none: TTM operating loss of $1,178M, cumulative accumulated deficit of $5,490M, and total stockholders’ equity of $129M after twenty-two years.

Verdict — a real moat that has not yet been shown to produce a financial outcome. The standard I hold moats to is explicit: if a moat claim cannot be tied to a financial outcome that would deteriorate without it, it is not a moat. Roblox’s moat is visible in retention and in creator lock-in; it is invisible in margins, in returns on capital, and — as Q2 showed — in the ability to defend monetization. Downgrade from “durable but not impregnable” to “durable on the user side, unproven on the economics side.”

5. Growth History and Forward Opportunities

The record, restated with the quarterly detail now available. Bookings by quarter, with Q3 2025 solved exactly from the guidance percentages ($1,576M ÷ 0.82 = $1,653M ÷ 0.86 = $1,922M) and Q4 2025 as the residual against the FY25 total of $6,790M:

* Derived from disclosed year-over-year growth rates; labelled ASSUMPTION.

Three readings of that table, all defensible, and this piece owes the reader all three:

The brutal reading. Bookings peaked in Q4 2025 at $2,223M and are running roughly 27% below that peak. Two consecutive quarters of sequential decline followed by a guided quarter that barely recovers. Annualising the Q3 midpoint gives $6,460M against FY25’s $6,790M — a declining business.

The comp-adjusted reading. Q3 2026 guidance implies +3.7% sequential growth, and the entire −14/−18% headline is lapping the extraordinary Q3 2025 viral quarter, which itself jumped 33.7% sequentially. Against Q3 2024 (roughly $1.13B, estimated from that quarter’s $919M of revenue), the Q3 2026 midpoint represents a two-year compound growth rate near 20% — precisely the “20%+ long-term” figure management keeps repeating. On a two-year stack, nothing is broken; 2025 was partly a bubble that is deflating.

The full-year reading. Adding 1H26 actuals of $3,288M to a Q3 midpoint of $1,615M and a Q4 that must lap $2,223M, FY2026 bookings land in the neighbourhood of $6.8B — approximately flat against FY2025. That is the number nobody prints. It should be set against the guidance history: 22–26% growth guided in February, cut to 8–12% on April 30, withdrawn on July 30 with the year tracking to roughly zero. This is the arithmetic that justifies the market’s reaction to the withdrawal rather than to the miss.

The forward opportunity set, in descending order of credibility:

• International. Highest credibility, already working, and does not depend on any new product. Japan +67% DAUs, India +64%, with hours growing in line. The offset is that international monetises far below the U.S., so international mix growth is dilutive to bookings-per-hour — part of the very metric that disappointed.
• The over-18 cohort. Credible and structurally attractive: 80% of the $200B market, U.S. over-18 DAUs +32%, 18–34 +42%, and monetization more than 50% above under-18. Roblox is attacking it with a DevEx rate premium, a 26-team incubator cohort, signed partnerships with established studios (launches later in 2026 and rolling through 2027), and now a platform expansion into 2D, single-player, offline and turn-based games. The strategic logic is sound and the execution risk is high — Roblox is proposing to compete for adult gamers on content quality against the entire industry, using a creator base optimised for a different audience.
• AI-assisted creation (Build). Genuine optionality with genuine cost. Build is a mobile-first, prompt-to-playable-game tab, in public alpha in New Zealand for ages 9+, with a free base tier and paid tiers. Management’s most striking datapoint: “we see already in New Zealand is more people per day on build than on studio.” If that generalises, content supply expands by orders of magnitude and discovery becomes the binding constraint — which is why Moments (a homepage short-form video tab) matters strategically rather than as a feature.
• Roblox Reality / Morpheus. A “foundational, multi-year bet” on photorealistic real-time multiplayer, to be monetised initially by subscription. This is the least credible line item and the most expensive. It has no revenue, no timeline, and management explicitly says roughly half of the guided Q3 margin compression is AI investment of this kind.
• Advertising and subscription. Repeatedly flagged, still nascent. Moments is positioned as a future “premium advertising surface.” Not in the numbers.

Quality of growth. Historically excellent — organic, engagement-led, internationally diversified, decreasingly top-10 dependent. Currently poor: the growth is negative on a cash basis, positive only on a lagged accounting basis, and the offsets that are working (international, over-18) are structurally mix-dilutive to the metric that broke. Verdict — a high-quality franchise in a genuinely low-quality growth year, with a credible but unproven 2027+ opportunity set that requires Roblox to win an audience it has never won, using content it does not yet have, funded by spending it cannot yet quantify.

6. Financial Quality

The float is the whole story, and it is now measurable. The earlier write-up’s central warning was that reported free cash flow was inflated by a deferred-revenue float that grows only while bookings outrun revenue. Here is the arithmetic, twelve months to 2026-06-30:

The deferred-revenue build alone is ~83% of trailing operating cash flow. And it is deflating in real time: the change in deferred revenue was $365M in Q2 2025, $99M in Q2 2026, and is guided at $173M for Q3 2026 against roughly $562M in Q3 2025 (Q3’25 bookings $1,922M less revenue $1,360M). Now trace the consequence:

Approximately 96% of the year-over-year decline in operating cash flow is the deferred-revenue delta. This is the cleanest possible confirmation of the earlier write-up’s quality-of-earnings thesis, and it means every bull citing “13.6x EV/free cash flow” is citing a number that is about to stop existing. Note that the float has not gone into reverse — bookings still exceed revenue by roughly $163M a quarter on the Q3 guide, so deferred revenue still grows, just far more slowly. The reversal proper would require bookings to fall below the amortisation of the 2025 cohort, which on current guidance is a 2027 question, not a 2026 one.

What Roblox actually earns in cash. Take Q2 2026, the last clean quarter. Total costs and expenses were $1,698M, of which $282M was stock compensation, leaving roughly $1,416M of cash-basis operating cost — and $34M of that was the state-settlement accrual. Against bookings of $1,557M, that is roughly $140M of cash operating profit before stock compensation, a ~9% margin, or roughly $560M annualised. Against a $22.35B enterprise value that is ~40x. After $1.14B of annual stock compensation — a real transfer of ownership, not an accounting fiction — the business loses money. This, not the 3x bookings multiple, is the honest valuation anchor.

Stock-based compensation. $1,016M (2024) → $1,129M (2025) → $557M in 1H26, with Q3 guided at $310M — a rising run-rate of roughly $1.2B, or ~19% of bookings and ~24% of revenue. The composition is telling: $195M of the $282M in Q2 sat in research and development. This is a company paying its engineers in equity at a rate that consumes its entire pre-SBC cash operating profit twice over.

Margins and the deleveraging ahead. Q2 loss from operations was $(229)M on $1,469M of revenue (−15.6%), an improvement on Q2 2025’s −29.8%. Adjusted EBITDA was $152M against $18M. Then the guide breaks the trend: Q3 consolidated net loss of $(348)M to $(307)M — nearly double Q2’s $(185)M — and Adjusted EBITDA of $0 to $41M. Management attributes roughly half of the year-over-year margin compression to fixed-cost deleveraging on lower bookings and roughly half to AI infrastructure. Critically, the guided net loss excludes loss-contingency accruals, so the actual reported figure will likely be worse by whatever the states settle for.

Capital intensity has changed character. Capex was $180M (FY24) and $441M (FY25, of which roughly $298M landed in Q4). Q3 2026 alone is guided at $170M against $24M in Q2. In June the company committed to $845M of hosting services over three annual periods beginning 2026-07-01, with annual commitments of $248–296M. The CFO’s framing is the important part: historically infrastructure scaled with hours and users — variable, self-funding. The new spend is “largely being driven by model training,” i.e. fixed cost incurred ahead of, and independent of, revenue. That converts an asset-light platform into something meaningfully more capital-hungry precisely as bookings decline.

Returns on capital. Not meaningful in the conventional sense and worth stating plainly rather than waving away: TTM operating loss $(1,178)M, TTM EBITDA $(934)M, total stockholders’ equity $129M (down from $375M at year-end on buybacks and losses), accumulated deficit $(5,490)M. Roblox has never earned a GAAP operating profit in any year of its public life. The equity base is now so thin that the price-to-book ratio (180x) is arithmetic noise rather than information.

Balance sheet — still the strongest part of the case. Cash $991M, short-term investments $2,023M, long-term investments $3,070M = $6,084M, against $1,009M of 3.875% senior notes due 2030 and no other funded debt (all leases are operating leases). Net cash $5,075M, or 18.5% of the market capitalisation and $7.10 per share. The current ratio is below 1.0 but that is an artefact of the $4,596M current deferred-revenue liability, which is an obligation to deliver virtual goods, not cash. Roblox can fund several years of this investment cycle and the buyback without touching the capital markets.

Verdict — the economics do improve with scale at the gross level and have not been shown to improve anywhere else. Gross margin is 80%, the mix shift to lower-cost payment platforms is a genuine tailwind, and cash collection is real. But cash operating margin is ~9% before a stock-compensation charge that is roughly double it; the free-cash-flow line that anchored the bull case is 83% float and deflating; capital intensity is stepping up into a fixed-cost AI build; and after twenty-two years the company has $129M of book equity. Downgrade from “prove-it P&L attached to a genuinely good business” to “prove-it P&L attached to a good business that has now had its one clean financial metric taken away.”

7. Capital Allocation

The buyback — right instinct, poor first execution. The board authorised $3.0B in May 2026 with stated intent to repurchase roughly $1B over twelve months, explicitly “designed to partially offset dilution from employee equity grants.” In Q2, Roblox repurchased and retired 8.2 million shares for approximately $380 million — an average of roughly $46.34 per share. The stock closed at $38.40 on 2026-08-21. The company’s first-ever capital return was therefore executed at a 21% premium to the price ten weeks later, and it was concentrated in the weeks immediately before management guided bookings down 16% and pulled the year. The CFO’s defence on the call is structurally sound — “the way we have structured our buyback… does result in us effectively buying more stock when the stock price goes down” — and $2.6B of authorisation remains, so the average will improve. But the fact stands: management bought $380M of its own stock into a quarter it knew was tracking to the low end of guidance. That is not an accusation of bad faith; it is an observation about what the buyback did and did not signal.

Dilution is not being controlled. Shares issued and outstanding went from 708,359K at 2025-12-31 to 714,292K at 2026-06-30 — up 5.9M despite retiring 8.2M, implying roughly 14.1M shares of gross issuance in six months, or ~4% annualised gross dilution. The fully diluted count is 752M, +2% year-over-year. A $1B annual buyback against $1.2B of annual stock compensation is mitigation, not reversal. Worse, the forward overhang expanded: shares reserved for future issuance rose from 175,582K to 205,630K in six months, now 28.8% of shares outstanding, with the 2020 Equity Incentive Plan reserve alone going 112,747K → 135,179K. Investors are buying an asset whose share count is contractually committed to grow.

Compensation design — a genuinely mixed record, and the 10-Q tells both halves. The 2021 CEO Long-Term Performance Award gave Baszucki the opportunity to earn up to 11.5 million shares on five-year stock-price targets. Those targets were not achieved and no shares vested — a real, working pay-for-performance outcome, and to management’s credit. But the award was then cancelled on 2024-03-01 and replaced with the 2024 CEO Award: a target of 446,534 shares against cumulative bookings (80% weight) and Adjusted EBITDA (20% weight) over a two-year period ending 2025-12-31, of which 67% vested in Q1 2026 with the balance vesting quarterly. The accounting note is the tell: because the modification produced no incremental fair value, “actual performance against the pre-established thresholds under the 2024 CEO PSU Award has no impact on the subsequent stock-based compensation expense recognized.” So a five-year, stock-price-linked award that was going to pay nothing was replaced with a two-year, bookings-and-EBITDA-linked award measured over 2024–2025 — precisely the two-year window in which bookings grew 25% then 55%. The hurdle was re-struck from a share price the company did not reach to an operating metric it was about to beat. It is the same failure mode seen elsewhere in the sector: an award re-cut to the achievable after the original proved unachievable.

Where the capital actually goes. In rough order: DevEx payouts to creators ($363M in Q2 alone, the highest-return outlay in the business because it directly fuels the flywheel); infrastructure and trust & safety ($363M in Q2, growing 54% year-over-year); R&D ($420M in Q2, of which $195M is stock compensation); and now capital expenditure and hosting commitments. M&A remains small and disciplined — the Morpheus AI acquisition added just $20M of goodwill (143 → 163) with $6M of cash paid net of cash acquired for the half. The absence of large, value-destroying acquisitions remains a genuine and underrated positive, especially measured against the 2021-era metaverse cohort. The CFO’s stated priority order on the call — organic investment first, “a lot of dry powder,” then buyback, then possibly “bigger things” in M&A — is sensible but leaves the door open to exactly the risk the record has so far avoided.

Governance — unchanged and still the weak point. David Baszucki holds super-voting Class B shares (20 votes each) giving him and affiliates voting control irrespective of economic ownership; 44,383K Class B shares remain outstanding against 669,909K Class A. The May 2025 reincorporation from Delaware to Nevada stands. Insider behaviour since the earlier write-up is uniformly one-directional: every open-market disposition between 2026-06-01 and 2026-08-21 was under a Rule 10b5-1 plan (director Greg Baszucki under a plan adopted 2025-11-28, selling 8,333 shares monthly through the crash; Chief Legal Officer Mark Reinstra under a plan adopted 2026-02-19), and not a single code-P open-market purchase by any insider across a 74% drawdown. Six Form 144s were filed on 2026-08-20, including one for CFO Naveen Chopra and one for the Greg & Christina Baszucki Living Trust — planned selling resumed within three weeks of the worst day in the company’s history. The plans are properly adopted and the amounts are individually small; the absence of buying is the durable signal, not the presence of selling.

Verdict — competent operator, clean M&A record, sensible capital-return instinct, poor first execution, expanding dilution, a re-struck CEO hurdle, and no insider willing to buy the low. Downgrade from “good operator, weak governance” to “good operator, weak governance, and a capital-return programme that has so far transferred value from the company to the sellers.”

8. Changes and Headwinds — Last Two Years

The strategic arc, compressed. 2024: recovery and operating leverage, the CEO award re-cut, bookings +25%. 2025: the boom — bookings +55% to $6.79B, DAUs to 144M, hours 124B, the stock to an all-time high of $141.56 in September, DevEx rates raised on 2025-09-05. Late 2025: the first cracks, with a −15.5% session on the Q3 print (2025-10-30) as age-check plans surfaced. January 2026: the pivot — Roblox becomes the first large platform to mandate proactive age-checks for access to communications globally. April 30, 2026: FY26 bookings guidance cut from +22–26% to +8–12%; the stock falls 18.3% the next session. May 2026: first-ever buyback, $3B authorised. June 2026: Kids and Select age-based accounts launch; the U.S. over-18 DevEx rate increase ships; the $845M hosting commitment is signed; the Morpheus AI acquisition closes. July 30, 2026: the break — Q2 bookings at the low end, Q3 guided to −14/−18%, FY26 guidance withdrawn, and a −26.9% session.

Headwind 1 — the monetization problem is new and is not the old problem. The April guide-down was a top-of-funnel story: age-gating chat suppressed word-of-mouth and organic installs. That problem is arguably fixing itself — sign-up activity “improved throughout the quarter,” Global Chat scaled to 30 million daily messages in June, Quick Words is generating over five million messages per day, In-Experience Friends Chat ships in Q3, and Russia came back online. The July guide-down is a monetization story with two named causes: engagement rotating out of high-monetising 2025-vintage viral games, and Roblox’s own RFY algorithm change (a 28-day retention window, up from seven days, with new play-through-rate and first-play-bounce-rate signals) intentionally surfacing stickier, lower-monetising content. Management calls it a deliberate trade and says internal A/B tests show the curves cross. They declined, twice, to say when.

Headwind 2 — litigation and regulation have become an operating line item. As of 2026-07-30, ten states plus Los Angeles County have filed suit (Arkansas, Florida, Indiana, Iowa, Kentucky, Louisiana, Nebraska, Oklahoma, Tennessee, Texas). Five states have settled (Alabama, Mississippi, Nevada, South Dakota, West Virginia), with more in negotiation. Roblox accrued $34M in Q2 and $91M in 1H26 in G&A for these, and adds them back in Adjusted EBITDA as non-routine — a characterisation that becomes harder to sustain each quarter they recur. Two motions to dismiss have been partly denied: Texas (Deceptive Trade Practices Act claim proceeds, 2026-03-06) and Iowa (Consumer Fraud Act claim proceeds, 2026-05-13). Separately, the consolidated Colvin/Gentry virtual-casino class action continues in the Northern District of California with discovery ongoing. And a federal securities class action now covers purchasers from 2024-10-31 to 2026-04-30, alleging the age-check impact was misrepresented.

Headwind 3 — the safety-moat claim took its first external hit. On 2026-08-19, Australia’s eSafety Commissioner secured a court-enforceable undertaking from Roblox after testing found adult strangers could still contact children despite the measures introduced. This is materially different from a lawsuit: it is a regulator with jurisdiction publicly concluding that the company’s flagship safety programme did not achieve its stated purpose. Note that Australia is the market where age-check penetration is highest (approaching 80%).

Headwind 4 — the cost base is rising into the downturn. Infrastructure and trust & safety grew 54% year-over-year to $236M (on the letter’s definition) and consumed 450 additional basis points of bookings. Q3 capex is guided at $170M. The $845M hosting commitment is contractual. Roblox is increasing fixed costs while bookings decline — the definition of negative operating leverage, and management guides to it explicitly.

The genuinely positive changes. Content diversification is real and improving (top-10 hours 30% → 20% over three years; long-tail hours +25% and spend +20%). Age-check penetration hit 57% globally with the U.S. and Australia above 70%. Kids and Select accounts now offer roughly 30,000 games, up nearly 50% since launch. Creator AI adoption is scaling fast. Signed partnerships with established studios bring franchise content later in 2026 and through 2027. The mix shift to lower-cost payment platforms is structurally improving gross margin. And the balance sheet is stronger than a year ago — total cash and investments of $6.1B, up $1.4B year-over-year, even after $375M of buybacks.

Verdict — the last two years contain both the best and the worst evidence about this company, and the sequence matters. The 2025 boom validated the platform; the 2026 reset revealed how much of that boom was carried by a handful of viral titles aimed at children, monetised through a discovery algorithm the company has now decided was optimising for the wrong thing. On balance these changes weaken the thesis — not because the platform is deteriorating, but because they show the economics are more fragile, the regulatory cost is permanent and rising, and management is willing to remove the information investors use to underwrite it.

9. Risk Analysis

The risk profile has shifted decisively from tail to base case. In June the top risks were “does the funnel repair” and “does the float deflate.” Both have now been answered — the funnel is repairing, the float is deflating — and been replaced by harder questions: does monetization recover, and what does the regulatory settlement wave actually cost in operating constraints rather than dollars. Note what is not on this list: solvency. Roblox cannot be forced into a bad outcome by its balance sheet, which is why the honest verdict is “avoid here,” not “short.”

10. Valuation Discussion

Rebuilt from the filing, not from a vendor. ROIC.ai publishes an enterprise value of $37,642M for Roblox. It is unusable on two counts: it is priced at the 2026-06-30 close of $54.38, and it nets only the $991M cash line while ignoring $5,093M of short- and long-term investments, while counting $1,836M of “debt” that includes operating leases. Rebuilt:

Why EV/revenue flatters and EV/bookings does not. GAAP revenue amortises bookings over 27 months. Today’s revenue line is therefore recognising the 2025 boom — which is why revenue printed +36% in a quarter when cash bookings grew 8%, and why revenue is guided +4/+10% in a quarter when bookings decline 14–18%. The revenue line is a rear-view mirror with a two-year lag, and it must roll over. Anyone comparing Roblox’s 3.9x EV/sales to Take-Two’s 6.8x or Unity’s 10.0x is comparing a peaking denominator to normal ones.

Take-Two and Unity EVs rebuilt from ROIC share counts (backed out of ev_to_sh_out) repriced at the 2026-08-21 close, with disclosed cash and debt. AppLovin approximated from public filings and market data at the same date.

The AZI own-history percentile, read correctly. Roblox’s price/sales percentile is 0.264 — the cheapest reading in its ten-year history. The P/E percentile is null (no positive GAAP EPS ever). The price/book ratio is 180.5x at the 66.7th percentile, which is arithmetic noise because book equity has collapsed to $129M; the composite reading of 33.5 is contaminated by that leg and should not be quoted. Report the split: on the only meaningful leg, this is the cheapest Roblox has ever been, on a sales denominator that is about to fall. Those two clauses must be read together.

Embedded expectations — what has to be true at $22.35B. Work backwards. For the enterprise value to compound at roughly 10% a year for five years, it must reach approximately $36B by 2030 (ignoring net-cash accretion, which is conservative in Roblox’s favour). At a terminal 15x on cash operating profit before stock compensation — a generous exit multiple for a business with this regulatory profile — that requires roughly $2.4B of pre-SBC cash operating profit in 2030, against roughly $560M today. The combinations that get there:

The market at $38.40 is underwriting something like “bookings compound at low-double-digits AND the cash operating margin roughly doubles from 9% to 18–20%.” That is not an absurd expectation — it is close to what management describes (20%+ long-term top-line growth, plus margin expansion from fixed-cost leverage, payment-platform mix and DevEx discipline). It is simply an expectation that requires both legs to work, in a year where the top line is flat, the margin is compressing, and the company has withdrawn the guidance that would let you track either.

Scenario analysis (enterprise value, 2030, with explicit assumptions — not price targets):

Adding roughly $5.1B of current net cash (and ignoring both future cash generation and future dilution, which broadly offset), the base case lands close to today’s market capitalisation. The base case is dead money for four years. The distribution is wide and roughly symmetric in log terms — which is precisely what “no guidance, negative near-term trend, fortress balance sheet, real optionality” should produce.

No price target and no recommendation appears in this section or anywhere in the analysis above.

11. Variant Perception

What consensus believes. After July 31, sell-side opinion split rather than converged. The dominant view is that 2026 is a self-inflicted, management-chosen reset of a structurally sound platform, that the algorithm change trades near-term monetization for long-term retention and lifetime value, and that a business with 123M DAUs at 3x bookings and 18.5% of its market cap in net cash is too cheap to short. The published pushback is narrower and sharper: that Roblox “is beginning to look like a business that will never be profitable,” and that withdrawing annual guidance at the exact moment visibility matters most is itself the signal.

The strongest bull case. Roblox is doing on purpose what most companies are forced to do by crisis: dismantling a monetization engine that was optimising for short-term extraction from children, in a regulatory environment that was going to force the issue anyway, while the installed base holds. The evidence for the “holds” part is strong — retention stable, DAUs +10%, MUPs +15%, content diversification improving, international compounding at 60%+, U.S. over-18 at +32% and monetising 50% better. The over-18 opportunity is genuinely 80% of a $200B market that Roblox has barely touched, and the platform is being extended (2D, single-player, offline, turn-based, plus signed studio partnerships) to address it. Build could expand content supply by orders of magnitude at near-zero marginal cost to Roblox. And the sequential arithmetic is far better than the headline: Q3 guides to +3.7% sequentially, and against Q3 2024 the two-year growth rate is roughly 20%. On that reading, 2025 was a viral bubble, 2026 is the deflation, and the underlying platform is compounding at management’s stated rate the whole time.

The strongest bear case. Every dollar of trailing free cash flow is a float artefact, and the float is deflating on a disclosed schedule. Strip it and Roblox generates roughly $560M of pre-SBC cash operating profit against $1.2B of stock compensation — it has never earned a profit and the path to one requires the cash operating margin to double while bookings are flat. The company is simultaneously stepping capex up 7x, signing $845M of hosting commitments, and building a photorealistic-gaming moonshot with no revenue, no timeline and no cost curve. Regulatory exposure has moved from tail to run-rate: $91M accrued in six months, ten states litigating, five settled, unquantified non-cash commitments, and an Australian regulator publicly finding the safety programme ineffective in the market where it is most penetrated. Dilution is 4% gross with a 28.8% reserved-share overhang. The CEO’s price-linked award was cancelled when it failed and replaced with a metric-linked award over a window the company was about to beat. No insider bought a share through a 74% drawdown. And the buyback’s first $380M was spent 21% above where the stock trades today.

The 3–5 assumptions that actually matter:

1. Does bookings-per-hour recover, and when? Everything else is downstream. Management asserts the retention curves cross but will not date it and will not share the tests.
2. Is the cash operating margin capable of reaching 18–20%? The valuation requires roughly a doubling from 9%. Fixed-cost AI investment is currently moving it the other way.
3. Do the state settlements cost dollars or degrees of freedom? Cash is affordable. Structural limits on under-13 monetisation would not be, given 35% of age-checked users are under 13.
4. Does the over-18 push produce content people actually play? Roblox has never won this audience and is asking a creator base tuned for children to build for adults, plus imported studio partners.
5. Does the float reversal arrive in 2027? Bookings still exceed revenue by roughly $163M a quarter, so deferred revenue still grows. If bookings stay near $1.6B while the 2025 cohort amortises off, reported revenue declines outright and the float goes into reverse — a second, larger optical shock.

The factor-positioning read. This is the section where the tape earns its place. The FactorsToday model returns an R² of only 14.6% on the base model and 21.1% on the fullest specification, with idiosyncratic volatility of 62.4% — Roblox’s returns are overwhelmingly its own story, not a sector trade. The L1-sparse fit zeroed both Momentum and Value entirely; the only meaningful loadings are Market +1.03 and LowVolatility −0.586 to −0.873, i.e. a high-beta, high-dispersion name. The leaderboard is unambiguous: one-year return −67.4% with a −74.9% maximum drawdown and a Sharpe of −1.03; six-month relative strength −39.1%, twelve-month −67.6%; alpha −0.17. The five-year record is a −14.4% annualised return with an −82.8% drawdown. And /related-stocks returns Samsara, ARKK, Palantir, the VanEck Social Sentiment ETF, C3.ai, ARKW, Asana, DocuSign, Lemonade, Zoom, Robinhood and Fiverr — not one game publisher, engine or interactive-entertainment company in the top twenty.

Two conclusions follow, and they cut in opposite directions. First, this is not a crowded trade — the momentum cohort left long ago, and there is no positive Momentum loading to unwind. That removes one common source of downside. Second, the market is not valuing Roblox as an interactive-entertainment franchise at all; it is valuing it as a busted speculative-growth position alongside ARKK constituents. The peer price diff confirms the dislocation is company-specific: over the twelve months to 2026-08-21, RBLX returned −66.6% while Take-Two returned +5.2%, Electronic Arts +22.7% (series ends 2026-08-05) and NetEase −0.4%. Since the earlier write-up on 2026-06-12, RBLX is −11.3% against Unity +72.3%, Take-Two +13.2%, Spotify +10.7% and NetEase +1.9% — a 72-point twelve-month gap to the nearest large publisher against a flat-to-up peer group. Nothing about that is sector beta.

Where consensus is most likely offsides. Consensus is probably right that the platform is intact and wrong about the timeframe over which that matters. The bull case requires a monetization inflection the company cannot date, a margin doubling it is currently spending against, and a regulatory resolution it cannot quantify — and it has removed the annual guidance that would let anyone track the first of those. The variant perception is not “Roblox is broken.” It is: the market is pricing a recovery whose timing has been made deliberately unobservable, and it is charging you nothing for that uncertainty.

12. Fact vs. Interpretation

13. Open Questions

1. When do the retention and monetization curves actually cross? Management says they do, has the A/B data, and declined twice on the call to give a timeline or a magnitude. This is the single most important undisclosed number in the company.
2. What does Q4 2026 look like against a $2,223M comparable? No guidance exists. A Q4 in line with Q3’s year-over-year rate implies roughly $1.85B; a Q4 that merely holds Q3’s sequential level implies roughly $1.7B. The difference between those and a seasonal recovery is the whole 2027 setup.
3. What is the total cash and non-cash cost of the state-AG settlement wave? $91M accrued in six months for five settled states. Ten more are litigating. The non-cash commitments — public-service campaigns, law-enforcement liaison headcount, platform changes, liquidated damages — are disclosed as unquantifiable.
4. Do any settlements constrain under-13 monetisation structurally? 35% of age-checked users are under 13 and that cohort is where the monetization decline is concentrated. A structural remedy would be a different order of problem from a cash payment.
5. What is the actual AI cost curve? The CFO says roughly half of Q3’s margin compression is AI investment and that it is now fixed rather than variable. The $845M hosting commitment covers three years. Nobody has quantified the steady-state run-rate or the revenue that is supposed to offset it.
6. Does Roblox Reality / Morpheus ever generate revenue? It is to be monetised “initially by subscription.” There is no date, no price, and no user estimate.
7. When does the deferred-revenue float go into reverse, not merely decelerate? On current guidance, bookings still exceed revenue. If bookings hold near $1.6B while the 2025 cohort amortises off, reported revenue declines outright — a second optical shock in 2027.
8. What is the steady-state dilution rate? Gross issuance ran ~4% annualised in 1H26 with 28.8% of the share count reserved. Management commits only to “strike a balance.”
9. Why has no insider bought? Across a 74% drawdown, with a founder-CEO who controls the company and a board that authorised $3B of buyback, there is not one code-P purchase on the tape.

14. What Must Be True

For the bull case to be right:

• Bookings inflect sequentially and bookings-per-hour stops falling. Specifically: Q4 2026 and Q1 2027 both grow sequentially, and management discloses (or the arithmetic reveals) that per-hour monetization has stabilised in the U.S. under-18 cohort.
• The over-18 and international engines scale fast enough to carry aggregate growth back toward the stated 20%+ before the 2025 cohort finishes amortising.
• The cash operating margin expands from ~9% toward the high teens as fixed-cost leverage returns and AI infrastructure spending plateaus rather than compounds.
• The state-AG wave resolves as a manageable cash cost rather than as structural limits on under-13 monetisation.
• Falsification test: If Q4 2026 bookings come in below roughly $1.7B, or if Q1 2027 fails to grow sequentially against Q4, the “deliberate reset” framing is wrong and this is a business in decline that happens to have a strong balance sheet. A second falsifier: if the cash operating margin before SBC does not exceed 12% in any quarter of 2027, the margin-doubling leg of the valuation is dead.

For the bear case to be right:

• Monetization damage proves structural rather than algorithmic — Roblox cannot both operate safely and monetise children at 2025 rates, and the over-18 substitution is too slow.
• Reported free cash flow keeps deteriorating as the float deflates and capex steps up, exposing the gap between headline cash flow and owner earnings.
• Regulatory cost escalates from settlement payments toward operating constraints, and the Australian finding is repeated by other regulators.
• Dilution and the 28.8% reserved-share overhang keep the per-share outcome worse than the enterprise outcome.
• Falsification test: If Roblox posts two consecutive quarters of sequential bookings growth with the cash operating margin before SBC above 12%, while the state settlements conclude without structural remedies, the “structurally impaired” thesis is wrong and today’s 0.26th-percentile sales multiple will look like the low.

15. Source Appendix

See Appendix B below.

Appendix A — Diligence Questionnaire

A supplemental diligence questionnaire covering the standard checklist an investor would run on this business.

General

What thoughtful questions have other investors asked about this company?

The sell-side Q&A on the 2026-07-30 call converged on four questions, and the answers management gave (and refused to give) are themselves the most useful diligence output of the quarter.

• Matthew Cost (Morgan Stanley): “What are you watching, and what should investors be watching, to determine when the engagement headwinds have peaked?” The CFO’s correction was the important part: “you talked about near-term engagement friction — that is really not what we are seeing… engagement looks good. The weakness… was really around monetization.” That reframing is the entire delta versus the April guide-down.
• Kenneth Gawrelski (Wells Fargo): when do the retention benefits show up, and how does the cost structure change? Management would not date the crossover: “We are not gonna put a specific timeline on that right now.”
• Omar Dessouky (Bank of America): “Is 2026 still a growth year? Would you be comfortable telling investors that 2026 bookings will be higher than 2025 bookings in total?” Management declined to answer. That refusal, on a direct yes/no question about whether the year grows at all, is the most information-dense moment on the call.
• Clark Lampen (BTIG): quantify the Morpheus/AI headwind. Answer: roughly half of the year-over-year Q3 margin compression is AI investment; the other half is fixed-cost deleveraging.
• Chris Scholl (UBS): does the share price change your willingness to lean into the buyback? Answer: the programme is sized to offset dilution, so it mechanically buys more as the price falls, with discretion to accelerate.

A fifth question no analyst asked, and one that matters: why was $380M of buyback executed in Q2 at ~$46.34 in a quarter management knew was tracking to the low end of its own guidance range?

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Neither, in the conventional sense — Roblox has never had positive GAAP earnings. But bookings are clearly coming off a cyclical high: Q4 2025 at $2,223M was the peak, and the current run-rate is roughly 27% below it. Reported revenue, by contrast, is at an all-time high and still growing 36%, because it amortises bookings over 27 months. The company is simultaneously at a bookings trough-in-progress and a revenue peak. Anyone reading the income statement alone is reading a two-year-old business.

Driven by the external environment or internal actions? Overwhelmingly internal, and deliberately so. Three self-inflicted decisions drive the 2026 reset: (1) mandatory age-checks for chat from January 2026, which suppressed the sign-up funnel; (2) the “Recommended For You” algorithm re-weighting from a 7-day to a 28-day retention window, which shifted impressions toward lower-monetising evergreen content; and (3) disabling cross-experience game passes. External factors are secondary: the Russia block (removed in June 2026, still a ~3-point year-over-year DAU headwind in Q2), FX (bookings +8% reported, +7% constant currency), and the regulatory environment that motivated (1) in the first place.

How stable are revenues? GAAP revenue is extremely stable by construction — it is a 27-month moving amortisation. Bookings are not: they swing on which experiences happen to be viral. Q3 2025 bookings jumped 33.7% sequentially on a run of high-monetising hits; Q1 2026 fell 22.1% sequentially. Management itself attributes the Q2 shortfall to “a greater than expected shift of engagement from high monetizing, 2025-vintage viral games.” A platform that describes its own results this way is more hit-exposed than the marketplace framing implies.

Outlook for products/services? Bifurcated. The core under-18 business is in managed decline on monetization while holding on engagement. The growth vectors are international (Japan DAUs +67%, India +64%), the over-18 cohort (U.S. O18 DAUs +32%, 18–34 +42%, monetising >50% better than U18), and three unproven product bets: Build (prompt-to-game AI creation, public alpha in New Zealand), Moments (short-form video discovery tab), and Roblox Reality/Morpheus (photorealistic real-time multiplayer, subscription-monetised, no timeline).

How big will this market be — growing, shrinking, domestic or international? Management sizes global gaming at ~$200B annually, of which the over-18 segment is roughly 80%, and states that nearly 4% of global gaming revenue and roughly 20% of U.S. under-18 gaming revenue already runs through Roblox. Both are management estimates. The market grows with global smartphone penetration and demographics; Roblox’s growth from here is a share-gain story in a cohort (adults) where its share is negligible, and a monetisation-development story internationally where DAU growth is strong but revenue per user is far below the U.S.

Business Quality & Competitive Moat

Is the industry getting more or less competitive? Less competitive on capital supply, more competitive on regulatory cost. The 2021-era metaverse capital wave retreated and most competitors failed. But the industry is now competing for the same GPU and hosting capacity as every AI company, and Roblox signed an $845M three-year hosting commitment in June 2026 to secure it. Competition for user attention remains intense and faddish.

How profitable is the business (ROIC, ROE)? Not meaningfully measurable, and that is the answer. TTM operating loss $(1,178)M, TTM EBITDA $(934)M, TTM net loss attributable $(1,001)M. Total stockholders’ equity is $129M against an accumulated deficit of $(5,490)M, so ROE is a meaningless ratio of a negative numerator to a near-zero denominator. The sector-appropriate analogues to use instead: gross margin 80%, cash operating margin before stock compensation ~9% of bookings, and free cash flow conversion, the last of which is currently an artefact of the deferred-revenue float (~83% of trailing operating cash flow).

How profitable is the industry — how many competitors, what barriers to entry? The user-generated-content gaming platform category has effectively one scaled participant (Roblox) plus adjacent giants (Microsoft’s Minecraft, Epic’s Fortnite/Unreal). Barriers to entry are genuine: a two-sided creator economy requires subsidising both sides for years, and Roblox has paid out well over $1.5B a year to creators ($363M in Q2 2026 alone). Traditional publishers earn real profits (Take-Two TTM EBITDA $1,170M); Roblox does not. The category is not proven profitable at the platform layer.

Can the business be easily understood? The business yes; the financial statements no. The 27-month revenue-recognition lag, the bookings/revenue wedge, the deferred-revenue float in operating cash flow, and $1.2B of annual stock compensation added back mean that revenue, EBITDA and free cash flow all mislead in different directions simultaneously. This is a genuine analytical hazard, not a complaint about disclosure — Roblox discloses all of it clearly.

Can it be undermined by foreign low-cost labour? Not directly — the content is supplied free by a global creator base that Roblox pays out of the revenue it generates, which is the opposite of a labour-cost exposure. Indirectly, yes: AI-assisted creation (Build) is explicitly designed to collapse the cost of producing content, which over time reduces the scarcity value of professional creators and could compress the DevEx rate Roblox must pay. That would be margin-accretive to Roblox and value-destructive to the incumbent creator elite.

Do brands matter? Yes, in two ways. The Roblox brand is a default destination for a large under-18 cohort — a genuine intangible. And third-party brands matter increasingly: the company signed partnerships in Q2 with “renowned game developers to bring popular game franchises… to Roblox,” with launches later in 2026 and rolling through 2027. That is an admission that user-generated content alone will not win the over-18 audience.

What is the nature of competition? Competition for time, not for a contract. Roblox competes with every other form of screen entertainment, and the Q2 result shows that even within Roblox, the competition between experiences moves the company’s revenue by material amounts quarter to quarter.

Customers’ switching costs? Real but behavioural: avatar identity, owned virtual items, friend graph, habit. The best evidence is that through a global chat restriction and a discovery overhaul, retention of existing users “remained stable relative to historical periods.” The countervailing evidence is that switching costs did not translate into pricing power — spend per hour fell anyway.

Financial Condition & Balance Sheet

Assets not fully recognised on the balance sheet? Substantial. The creator ecosystem (~20 million creators, the DevEx relationship, top studios earning tens of millions), the brand, the proprietary safety and 3D-generative models, the accumulated interaction data (29 billion hours a quarter), and the age-verification infrastructure are all expensed as incurred and carry no balance-sheet value. Goodwill is only $163M and intangibles $20M against $9,745M of total assets. This is the strongest argument that the $129M of book equity understates the business.

Off-balance-sheet liabilities? Three worth naming. (1) Purchase obligations: $845M of hosting services committed in June 2026 over three annual periods, $248–296M per year, plus other data-centre, software and payment-processor commitments. (2) Unquantified settlement commitments: the 10-Q states that public-service campaigns, law-enforcement liaison headcount, platform and policy change costs, and liquidated damages under the state settlements have not been accrued and will be expensed as incurred. (3) Operating leases running through 2037 (2041 for leases not yet commenced), with right-of-use assets of $689M and lease liabilities of $669M long-term — on balance sheet, but easy to miss when computing net debt (and, note, ROIC.ai’s published enterprise value includes them as debt).

How conservative is the accounting? Conservative in the places that matter and clearly disclosed everywhere. Revenue recognition over a 27-month estimated paying-user life is genuinely conservative relative to recognising bookings on receipt. Bookings, Adjusted EBITDA and free cash flow are all reconciled to GAAP in the release. The one item to watch is the “legal settlement expenses” add-back in Adjusted EBITDA — $34M in Q2, $91M in 1H26 — characterised as arising “outside of the ordinary course of business” with “limited historical precedent.” With ten more states litigating and negotiations ongoing, that characterisation becomes harder to defend each quarter. The other estimate to watch is the 27-month paying-user life itself: guidance explicitly assumes “no material changes in estimates used in revenue recognition.” A shortening of that estimate would pull revenue forward; a lengthening would push it out.

How CapEx-hungry is the business? It was not; it now is. Capex was $180M (FY24) and $441M (FY25). Q3 2026 alone is guided at $170M against $24M in Q2 2026. More important than the level is the character: the CFO stated that historically “infrastructure investments have been a function of hours and users on our platform” — variable and self-funding — whereas the new spend is “largely being driven by model training,” i.e. fixed cost incurred ahead of revenue. Roblox is converting from an asset-light platform into a semi-capital-intensive AI operator while its top line declines.

Capital Allocation & Management

How much FCF does the business generate, how does management use it, and what is the philosophy? Trailing free cash flow is $1,641M, but approximately 83% of trailing operating cash flow is the deferred-revenue float build, which is deflating: Q3 2026 free cash flow is guided to $(60)M to $5M against roughly $443M in Q3 2025. The philosophy, stated plainly by the CFO: organic investment first (“we are gonna keep a lot of dry powder”), then buyback sized to offset dilution, then potentially M&A. Uses in practice: DevEx payouts to creators, infrastructure and trust & safety, R&D, and now capex.

Significant acquisitions recently? Only Morpheus AI (Self Forcing / pixel-latent world models), which added $20M of goodwill and $6M of net cash consideration for the half. The M&A record remains genuinely disciplined — no large acquisitions, no metaverse-era write-downs, in an industry littered with both. The CFO’s remark that Roblox might do “bigger things in the future to the extent there is strategic or industrial logic” is the risk to monitor, not a present problem.

Buying back shares? Yes, for the first time. $3.0B authorised May 2026 with ~$1B intended over twelve months; 8.2M shares retired for ~$380M in Q2 at an average of roughly $46.34, against a 2026-08-21 close of $38.40; $2.6B remains. The programme is explicitly “designed to partially offset dilution from employee equity grants.”

Issuing large amounts of new shares to insiders? Yes. Stock compensation was $1,016M (2024), $1,129M (2025) and $557M in 1H26 with $310M guided for Q3 — a rising run-rate near $1.2B, roughly 19% of bookings. Shares outstanding rose from 708,359K to 714,292K in six months despite retiring 8.2M, implying ~14.1M of gross issuance. Shares reserved for future issuance rose from 175,582K to 205,630K — 28.8% of shares outstanding. The buyback is mitigation, not reversal.

Compensation policy of directors/management? Heavily equity-weighted, with a mixed record on hurdle integrity. The 2021 CEO Long-Term Performance Award (up to 11.5 million shares on five-year stock-price targets) paid nothing — the targets were not achieved — which is a real pay-for-performance outcome. But it was cancelled on 2024-03-01 and replaced with the 2024 CEO Award: 446,534 target shares on cumulative bookings (80%) and Adjusted EBITDA (20%) over 2024–2025, of which 67% vested in Q1 2026. A stock-price hurdle that was failing was replaced with operating hurdles over a window in which bookings grew 25% then 55%. The Outside Director Compensation Policy was amended and the CFO’s offer letter was amended on 2026-06-09 (Exhibits 10.1 and 10.2 to the Q2 10-Q); the terms warrant review in the next proxy.

Motivations of management? Founder-led and long-horizon, with control. David Baszucki holds Class B shares carrying 20 votes each (44,383K Class B against 669,909K Class A), giving him and affiliates voting control regardless of economic ownership; the company reincorporated from Delaware to Nevada in May 2025. The 2026 decisions — mandatory age-checks, a discovery algorithm optimising 28-day retention over near-term spend, heavy AI investment, and withdrawal of annual guidance because “we do not believe annual guidance is a helpful tool for investors” — are all consistent with a founder optimising a decade out and structurally insulated from shareholders who disagree. Whether that is a feature or a bug depends entirely on whether he is right, and shareholders have no mechanism to test it. Insider behaviour offers no reassurance in either direction: all sales since June were under properly adopted 10b5-1 plans, and there was not one open-market purchase across a 74% drawdown.

Valuation & Market Data

Is the stock an ADR, MLP, or K-1 issuer? No. Roblox is a U.S. domestic issuer (Nevada, since May 2025), NYSE-listed, dual-class (Class A one vote, Class B twenty votes), issuing a standard Form 1099. No K-1, no partnership structure, no ADR mechanics.

Dividend policy? None, and none contemplated. Capital return runs entirely through the $3.0B repurchase authorisation.

How profitable is the business? Restated for this section: gross margin 80%; cash operating margin before stock compensation approximately 9% of bookings (roughly $140M in Q2 on $1,557M); GAAP operating margin −15.6% in Q2 2026 (improved from −29.8%) but guided to deteriorate sharply in Q3, with consolidated net loss of $(348)M to $(307)M against $(185)M in Q2 — and that guidance excludes loss-contingency accruals. Adjusted EBITDA of $152M in Q2 is guided to $0–41M in Q3.

Is net income diverging from cash from operations? Extremely, and this is the central analytical issue in the name. Trailing twelve months: net loss attributable $(1,001)M against operating cash flow of $2,100M — a $3.1B divergence. The bridge is $1,142M of stock compensation (a real transfer of ownership that GAAP expenses and the cash-flow statement adds back), ~$248M of depreciation, and ~$1,743M of deferred-revenue float build. The float is the swing factor and it is deflating on a disclosed schedule: the change in deferred revenue went from $365M in Q2 2025 to $99M in Q2 2026, with $173M guided for Q3 2026 against roughly $562M a year earlier. Roughly 96% of the guided year-over-year decline in Q3 operating cash flow is that single line.

Risks & Downside

What factors would cause the stock to decline? In rough order of probability: (1) Q4 2026 bookings below roughly $1.7B against a $2,223M comparable; (2) the monetization decline persisting into 2027 with no disclosed crossover; (3) a state-AG settlement imposing structural limits on under-13 monetisation rather than cash payments (35% of age-checked users are under 13); (4) AI infrastructure spending compounding without offsetting revenue, keeping Adjusted EBITDA near zero; (5) the deferred-revenue float turning outright negative in 2027, causing reported revenue to decline for the first time; (6) further regulatory findings echoing Australia’s that the safety programme does not work; (7) continued dilution against the 28.8% reserved-share overhang.

Risk of a catastrophic loss? Low in the financial sense. Net cash is $5,075M — 18.5% of the market capitalisation — the only funded debt is $1,009M of 3.875% notes not due until 2030, all leases are operating leases, and Adjusted EBITDA remains positive. There is no covenant pressure and no refinancing wall. The realistic catastrophic scenario is not insolvency but regulatory amputation: a settlement or statute that materially restricts monetisation of under-13 users across multiple jurisdictions. Roblox would survive that; the equity would not survive it at this multiple.

Chance of a total loss? Negligible on any reasonable horizon. A platform with 123M daily users, 29 billion quarterly hours, 27 million monthly payers, 80% gross margins and $6.1B of cash and investments does not go to zero. This is why the framing here is “avoid here,” not “short.”

Recent News & Events

Has the business environment changed recently? Yes, materially, on 2026-07-30. Roblox reported Q2 bookings of $1,557M (+8%, low end of guidance), guided Q3 bookings to a 14–18% year-over-year decline, and withdrew full-year guidance, describing it as accelerating a previously announced move to quarterly-only guidance by one quarter. The stock fell 26.9% the following session — the worst day in its listed history. The stated cause is a decline in per-hour monetization concentrated in U.S./Canada under-13 users, driven by a mix shift out of 2025-vintage viral hits and compounded by the company’s own recommendation-algorithm change.

Significant acquisitions? Morpheus AI, acquired in Q2 2026 to accelerate Roblox Reality (photorealistic real-time multiplayer). Small in financial terms ($20M of goodwill; $6M of net cash consideration for the half) and large in strategic terms — it anchors a multi-year, unquantified AI investment programme.

Change in accounting policies? None. The estimated average lifetime of a paying user held at 27 months in both Q2 2026 and Q2 2025, and guidance explicitly assumes no material change to the consumable/durable allocation or that estimate. The one presentational change worth flagging is the “legal settlement expenses” add-back introduced in Adjusted EBITDA ($34M Q2, $91M 1H26).

Recent changes — new markets, facilities, management? New markets in the product sense rather than the geographic: Roblox is removing the “self-imposed limits” on content type, extending the platform to 2D, single-player, offline and turn-based games to reach the over-18 cohort. Russia came back online in June 2026. New facilities: the $845M three-year hosting commitment signed in June and a step-change in capex. Management is unchanged at the top (Baszucki CEO, Chopra CFO), though the CFO’s offer letter was amended on 2026-06-09 and the Outside Director Compensation Policy was amended. Governance events: director Chris Carvalho terminated his November 2025 Rule 10b5-1 plan on 2026-05-05 and adopted a new one covering up to 120,000 shares through 2027-08-31.

Other material developments. Ten states plus Los Angeles County have filed youth-safety suits; five states have settled; $91M was accrued in 1H26. Australia’s eSafety Commissioner secured a court-enforceable undertaking on 2026-08-19 after testing found adult strangers could still contact children despite the new measures. A federal securities class action covers purchasers from 2024-10-31 to 2026-04-30. Age-check penetration reached 57% globally (U.S. and Australia above 70%). Kids and Select age-based accounts launched in June with roughly 30,000 games available, up nearly 50% since launch. Global Chat scaled to 30 million daily messages in June and Quick Words to over five million per day.

Appendix B — Sources

Report date 2026-08-22. All URLs accessed 2026-08-22 unless otherwise noted. Primary sources first.

A. Primary — SEC filings and company disclosures

B. Quantitative data services

C. News and third-party reporting

D. Methodological notes

1. Enterprise value is rebuilt by hand at the 2026-08-21 close from the 10-Q cover-page share counts (669,909K Class A + 44,383K Class B = 714,292K) and the balance sheet ($6,084M of cash and investments, $1,009M of notes). No vendor enterprise value is used for Roblox or for any peer.
2. Q3 2025 bookings of $1,922M are derived, not disclosed — solved from the Q3 2026 guidance percentages ($1,576M ÷ 0.82 and $1,653M ÷ 0.86 both return $1,922M). Q4 2025 bookings of $2,223M are the residual against the FY2025 total of $6,790M.
3. Q3 2025 free cash flow of approximately $443M is derived as Q3 2025 operating cash flow ($1,189M nine-month less $643M half-year = $546M) less Q3 2025 capex ($143M nine-month less $40M half-year = $103M). It cross-checks against the company’s own “(99)%” year-over-year free-cash-flow guidance at the high end of the Q3 2026 range.
4. Cash operating profit before stock compensation is computed as bookings less (total costs and expenses less stock compensation), using Q2 2026: $1,557M − ($1,698M − $282M) = $141M, annualised to approximately $560M. This is an approximation because cost of revenue is deferred on the same basis as revenue; it is used as an order-of-magnitude anchor, not a precise figure.
5. Management commentary is treated as hypothesis, not evidence. Every claim sourced to the call or the shareholder letter is labelled as such above, including the unverifiable A/B-test assertion that underpins the bull case.