Applied Digital Corporation (NASDAQ: APLD) — A 15-Year Annuity, One Fragile Tenant Deep

An independent equity research note Report date: 2026-06-12 Price reference: ~$41.47 (NASDAQ close, 2026-06-11); 52-week range ~$4.20–$42.27 Shares outstanding: ~285.8M (as of 2026-02-28 cover); fully diluted materially higher incl. $450M converts + 14.7M PSU/RSU pool Market cap: ~$10.0B (basic) | Enterprise value: ~$11.1B (basic; understates true claims — see Financial Quality and Valuation) Fiscal year-end: May 31 | CIK: 0001144879 | HQ: Dallas, TX | Former “Applied Blockchain, Inc.” (renamed Nov 2022)

⚡ Claude’s Take

This block is the author’s own independent opinion and general information only — not investment advice. The analysis that follows it deliberately carries no recommendation and no price target; the only position taken anywhere in this article is in this clearly-labeled block.

Verdict: AVOID at ~$41 — this is a great asset story wrapped around a fragile equity claim at a euphoric price. Not a short (the squeeze risk is extreme). Conviction: medium-high. Directional fair-value zone: ~$14–$24, i.e. roughly 0.6–1.0x a generously-credited, fully-diluted forward equity value that properly subordinates the common to ~$2.9B of project notes and a $0.9B+ accreting 12.75%+ preferred. I want either a much lower entry or hard, P&L-visible proof that the CoreWeave annuity is being collected — not just contracted — before paying up.

Tag: “The best-structured house in the neocloud land-grab — built on one tenant’s promise to pay for fifteen years.”

Of all the Bitcoin-miner-to-AI converts, Applied Digital has made the single smartest structural choice: it became the landlord, not the cloud. The tenant — not APLD — owns the GPUs, so APLD carries none of the 6-year-vs-economic-life GPU-depreciation time bomb that hangs over CoreWeave, Nebius and IREN. It owns a 30-to-50-year power-and-shell asset, has a genuine first-mover edge in pre-secured Dakota grid power, and — to management’s real credit — delivered its first 100MW liquid-cooled building in Ellendale on time and on budget, which is the hardest thing in this industry to do. That is why I will not short it: $15.5B of contracted 15-year leases, a 30%-of-float short interest, a 7.3 beta, and relentless positive lease/financing headlines can squeeze this far higher before any bear thesis pays.

But strip the narrative and look at the equity. The market is paying ~$10B for the residual claim that sits junior to ~$2.9B of 8–9%+ project notes and a $923M Macquarie preferred that accretes at 12.75% rising to 16.75% with a 1.80x minimum-return floor — and ~71% of the entire $15.5B backlog rests on CoreWeave, a deeply FCF-negative (–$7.25B in FY25), ~67%-Microsoft-concentrated, ~BB-standalone counterparty. The clever A3-rated SPV/lockbox/springing-guarantee structure improves recovery, but it cannot manufacture solvency: a take-or-pay lease is only as good as the lessee, and in a CoreWeave Chapter 11 the lease is rejectable. Meanwhile the “Adjusted EBITDA of $44M” is positive only because ~$49M of recurring, dilutive stock comp is added back; true economic EBITDA is roughly breakeven, free cash flow was –$1.62B over nine months, the share count is up 27% in nine months (and ~190% since 2022), and the CEO’s own open-market buying happened entirely in 2022–23 at ~$2.72 — he has bought nothing into the 10x run and has trimmed into strength. What the market is mispricing: it is capitalizing gross contracted lease revenue as if it were earned, de-risked, terminal NOI to the common, when (a) almost none has hit the P&L, (b) the senior preferred and notes eat the first ~$300–400M of annual cash flow, and © the counterparty is one funding-market freeze away from distress. This is a late-capital-cycle momentum bet dressed as real estate. Bull-flip trigger: the PF1/PF2 buildings energize and CoreWeave + the IG hyperscaler actually pay contracted rent into the P&L at high-80s% NOI margins for a few quarters, with the cost of capital refinanced toward investment-grade ABS — proof the annuity is real and collectible. Bear-flip trigger: any CoreWeave funding wobble, a construction-cost/timing slip on the ~900MW under build, or a tightening of high-yield/preferred markets — any one of which turns a $25–40B forward capital need into a dilution spiral or a covenant problem.

1. Executive Summary

Applied Digital Corporation is a Dallas-based developer, owner and operator of large-scale digital infrastructure, mid-transformation from a commodity Bitcoin-hosting business into a build-to-suit AI/HPC data-center landlord. The company designs, finances, constructs and leases purpose-built, high-density, liquid-cooled data-center campuses to hyperscale and “neocloud” tenants under 15-year (extendable to 30-year) triple-net-style leases. Its defining asset is Polaris Forge 1 in Ellendale, North Dakota — a 400MW campus leased to CoreWeave — whose first 100MW building came online in the quarter ended February 2026. A second campus (Polaris Forge 2, ~200MW) is leased to an unnamed investment-grade hyperscaler, and a third (Delta Forge 1, ~300MW, southern U.S.) broke ground in early 2026. As of the Q3 FY26 call, total contracted lease revenue stood at ~$16B (~$11B CoreWeave + ~$5B IG hyperscaler), with subsequent leases (a ~$5.2B/210MW hyperscaler deal in June 2026) lifting it further.

The financials tell two stories at once. The operating story is early and improving: Q3 FY26 (quarter ended 2026-02-28) revenue was $126.6M, up 139% YoY, the HPC-hosting segment turned a $17.6M operating profit, and the legacy crypto-hosting segment quietly earned $13.9M of operating profit on just $119.6M of assets — the highest return-on-assets line in the company. The financing story is alarming: the company generated –$1.62B of free cash flow in nine months, funded entirely by capital markets (debt borrowings of $2.5B, a $900M Macquarie preferred, and $196M of ATM equity), grew its share count 27% in nine months, and carries a capital stack — 9.25% notes, 6.75% notes, a 12.75%-accreting preferred — that is expensive even by frontier-infrastructure standards. Reported “Adjusted EBITDA” of $44.1M and “Adjusted net income” of $33.2M (+$0.09/sh) flatter a GAAP net loss to common of $100.9M (–$0.36/sh) by adding back ~$49M of stock-based compensation and excising the entire operating loss of a cloud business APLD still ~97% owns.

The bull case is genuinely attractive at the asset level: a real, contracted, long-dated annuity backed by a 30–50-year power-and-shell asset, in a power-constrained market with ~$700B/yr of hyperscaler capex chasing capacity, and — crucially — no GPU-depreciation risk because tenants own the silicon. The bear case is equally concrete at the equity level: a sub-scale (one building operating), single-anchor-tenant developer, financed with the most expensive capital in the complex, whose marquee customer is itself cash-burning and Microsoft-concentrated, run by a serial business-model pivoter and serial diluter, valued at ~30–37x sales and ~7x book (both in the ~92nd percentile of its own history) — a price that already discounts near-flawless execution of a multi-year, ~$25–40B build.

This memo takes no position and sets no price target. It frames APLD as a structure of embedded expectations: the equity is, in substance, a forward, execution- and counterparty-gated option on (1) converting a contracted backlog into collected, P&L-visible NOI, (2) CoreWeave’s multi-year solvency, and (3) continuous, cheapening capital access — at a price that assumes all three resolve favorably.

2. Business Overview

What APLD does. Applied Digital builds and operates digital infrastructure across three segments that are at very different life stages. Founded around 2020 and listed (via an existing shell, CIK 0001144879) in 2022 as “Applied Blockchain,” it renamed to Applied Digital in November 2022 to signal the AI pivot. The business today is best understood as a wholesale/hyperscale data-center developer-landlord with a melting crypto cash-cow attached and a GPU-cloud business on its way out the door.

The three segments (Q3 FY26, quarter ended 2026-02-28):

1. HPC Hosting — the future of the company (Q3 revenue $71.0M; segment operating profit $17.6M). APLD develops purpose-built, liquid-cooled (direct-to-chip) AI data-center campuses and leases the shell + power + cooling to tenants who install and own their own GPUs. Revenue has three components, of very different quality: base rent ($44.1M) — the durable, contracted, straight-lined annuity; tenant fit-out services ($18.9M) — low-margin (~5%), non-recurring, percentage-of-completion construction revenue; and power pass-through and ancillary ($8.1M) — a margin-neutral cost reimbursement. Leases are 15 years (extendable), effectively non-cancelable / make-whole, with annual escalators, at a guided ~$130/kW/month and high-80s%-to-90s% cash NOI margins. The first 100MW building at Polaris Forge 1 (the CoreWeave campus) energized in the quarter; ~700–900MW more is under construction.

2. Data Center Hosting — the legacy crypto cash engine, being wound down (Q3 revenue $37.5M, +7% YoY; segment operating profit $13.9M on $119.6M of assets). APLD hosts third-party Bitcoin miners at two sites (~286MW, primarily Jamestown and Ellendale-adjacent capacity in the Dakotas). This is, ironically, the company’s highest-return-on-assets business (~$56M annualized operating profit on ~$120M of assets ≈ ~45% asset yield), but it is structurally non-strategic: management is converting powered sites to HPC use and treats the crypto business as a transitional, self-funding bootstrap — a “melting ice cube” that throws off cash while the HPC build ramps.

3. Cloud — the abandoned strategy, being spun off (Q3 revenue $18.1M; segment loss $52.2M incl. a $59.7M non-cash write-down). “Sai Computing” was APLD’s own GPU neocloud — ordered 34,000+ NVIDIA H100s in 2023 and pitched at the October 2023 Investor Day as a co-equal third pillar. By April 2025 the Board had put it under strategic review / held-for-sale; in February 2026 APLD signed a binding agreement to combine it with EKSO Bionics in a reverse-merger to form “ChronoScale Corporation,” with the stated long-term goal of spinning it to APLD shareholders. Per the contribution agreement APLD retains ~97% economically for now; the reclassification out of held-for-sale triggered the $59.7M write-down. Management explicitly excludes the cloud’s losses from “Adjusted” metrics — a treatment worth scrutinizing (see Financial Quality) given APLD still owns nearly all of it.

Revenue model in plain terms. HPC hosting = real estate: rent per kW per month × MW leased × occupancy, on a non-cancelable 15-year contract, with the tenant bearing GPU obsolescence and (largely) power-price risk. Crypto hosting = a power-arbitrage tolling fee. Cloud = GPU-hour rental (being exited). The strategic thesis is that APLD converts cheap, early-secured grid power and proven EPC execution into a portfolio of long-dated, credit-backed lease annuities, financed ~80% with project debt + preferred so that common shareholders retain >85% of each site’s equity while putting up little incremental cash.

Recurring vs. non-recurring. The contracted HPC base rent is the most genuinely recurring revenue APLD has ever had — but as of Q3 FY26 only one 100MW building was generating it. A large share of reported HPC revenue is still non-recurring fit-out construction, and the bulk of total company revenue today is still the melting crypto business plus the exiting cloud. The annuity is contracted; it is not yet collected.

Verdict. APLD is a mid-transition infrastructure developer with a genuinely differentiated, lower-risk business model than its miner-turned-neocloud peers, but an enterprise whose quality rests almost entirely on a contracted-but-largely-unrecognized backlog arriving on time, on budget, and from a fragile anchor tenant — while the legacy cash engine melts and the cloud misadventure is spun away at a loss.

3. Industry Dynamics

APLD operates at the wholesale/hyperscale data-center leasing rung of the AI infrastructure value chain. It must be judged on the structure of that rung, plus the melting-ice-cube economics of Bitcoin hosting.

3a. AI/HPC data-center leasing — booming demand, late-cycle capital, weak terminal barriers

The demand side is a genuine secular boom. Aggregate 2026 hyperscaler capex guidance has been reported at ~$635–725B (management cited “nearly $700B,” up from ~$400B just a few quarters earlier), with power and “time-to-power” the binding constraints. Industry commentary (and APLD’s own anecdote of Elon Musk warning that “all available excess power on the grid… will still not be enough”) points to a multi-year capacity shortfall. On the demand axis, this is one of the most attractive backdrops any infrastructure developer has ever faced — and APLD’s core scarce input (pre-secured, interconnected grid power in low-cost, cold-climate, business-friendly states) is exactly what is short.

But the profit pool concentrates away from APLD’s rung. In the AI value chain, NVIDIA captures ~75% gross margins at the silicon layer; the capacity/landlord layer where APLD sits is the most capital-intensive and most commoditizable rung, sandwiched between an oligopolist supplier and sophisticated, self-build-capable customers. The wholesale data-center business is a good real-estate business when it is scaled and diversified (Digital Realty, Equinix, QTS, Vantage) — but those qualities come from thousands of tenants and decades of interconnection density, neither of which APLD has.

Capital-cycle position (Marathon lens): mid-to-late euphoria. The single most important framework point for this name: high current returns are precisely the signal that attracts capital and mean-reverts returns. Capital is flooding into AI data-center capacity from every direction — hyperscaler self-build, scaled REITs, private equity and infrastructure funds, sovereign money, and essentially every Bitcoin miner pivoting (IREN, MARA, RIOT, CIFR, WULF/Terawulf, Cipher, Hut 8). Today’s scarcity rents (~$130/kW/month, high-80s% NOI margins) are observable now; the supply catch-up, ASP compression and credit-spread normalization that the capital cycle predicts typically arrive later. APLD is building into the bull phase — correct, but the durability of the rents being capitalized into a $10B valuation is the open question.

Verdict: structurally attractive demand, structurally weak barriers, at a cyclically euphoric moment. A real, large, growing TAM — but a capacity/landlord layer prone to commoditization, where APLD competes sub-scale against players 6–10x its size and against the customers themselves. Attractive to ride with discipline; dangerous to underwrite through the cycle at today’s rent and pricing assumptions.

3b. Bitcoin hosting — structurally bad, correctly treated as transitional

Bitcoin mining/hosting is a textbook commodity industry: no barriers to entry, a difficulty ratchet that mechanically erodes each participant’s share of a fixed block reward, and a quadrennial halving that compresses hashprice. APLD’s exposure is the lower-beta hosting (tolling) version rather than self-mining, and it is genuinely high-return on a small asset base — but it is structurally non-strategic and shrinking as a share of the company. Management’s decision to use it as a self-funding bootstrap and reallocate power to higher-value compute is exactly the right capital-cycle response: harvest a mature, over-supplied business at the cost-of-capital floor and redeploy into the scarce resource (power for AI).

Verdict: structurally unattractive industry, correctly being run off. For valuation it should be treated as a transitional cash engine (~$50M/yr operating profit, declining), not a terminal business.

4. Competitive Position

Management’s claimed moat (hypothesis). Wes Cummins frames APLD’s edge as real assets “not easily replicable”: pre-secured low-cost Dakota grid power, owned/controlled land, in-house EPC and construction capability, a proven ability to “execute on time and deliver fully functional state-of-the-art facilities,” and accumulated permitting/community goodwill (the Ellendale ratepayer-savings story). The implicit moat claim is that being early to power in an under-supplied region, plus an execution track record few can match, creates a durable advantage.

Pressure-tested against the Greenwald taxonomy:

• Cost / supply-side advantage — real but locational and erodible. Cheap, early-interconnected Dakota power plus in-house construction genuinely lowers cost versus a developer paying third-party rates or waiting in an interconnection queue. But this is an asset-level advantage and a timing advantage — not a franchise. A well-capitalized rival can buy land in a business-friendly state and join an interconnection queue; APLD’s edge is its accumulated queue position and EPC learning curve, which shrink as the whole industry races to secure power. It is a head-start, not a wall.

• Customer captivity / switching costs — weak. The lock-in is the 15-year contract, not the asset. During the term, the make-whole lease is genuinely sticky (the tenant owes rent whether or not it draws power). At lease-end, switching costs are low-to-moderate: the shell is somewhat re-tenantable (a plus versus stranded single-purpose assets), but nothing structurally binds a hyperscaler to renew at APLD rather than self-build or move to a larger, more diversified landlord. The customers are the most sophisticated multi-sourcers on earth and are themselves the largest competitive threat (every hyperscaler is building its own capacity).

• Economies of scale + captivity (the only durable Greenwald combination) — absent. APLD has neither the scale (one operating building; ~$319M TTM revenue vs DLR/EQIX at $6–9B) nor the captive demand (interconnection density, ecosystem) that makes Equinix/Digital Realty defensible. It is a price-taker in a scarce-power window.

The “$15.5B annuity” framing, pressure-tested. The backlog is real and contracted, but it is (a) concentrated — ~71% CoreWeave; (b) encumbered — junior to ~$2.9B of project notes and a $923M accreting preferred at the prime SPV; © partly accrual-flattered — GAAP straight-lines escalating 15-year rents, building a non-cash “lease incentive” receivable ($201.8M, up from $84.4M) that runs ahead of cash; and (d) counterparty-gated — an annuity is only as good as the tenant’s 15-year solvency (below).

Verdict: no durable competitive advantage — a genuine, value-relevant head-start in a scarce-power window, not a franchise. APLD’s business-model choice (landlord, not cloud) is structurally superior to its peers because it sheds GPU-depreciation risk and owns a long-life asset; but the competitive position is that of a sub-scale price-taker whose edge erodes as capital floods the industry. The moat, such as it is, cannot be tied to a financial outcome that would persist if the scarce-power window closes.

5. Growth History and Forward Opportunities

Growth history — explosive but low-quality so far. Revenue compounded from essentially nothing to a real base: $8.5M (FY22) → $55.4M (FY23) → $165.6M (FY24) → $144.2M (FY25, lower as the cloud business was reclassified to held-for-sale) → $352.6M (9M FY26). Quarterly revenue more than doubled YoY to $126.6M in Q3 FY26 (+139%). But the composition of that growth is the caveat: a large share of the HPC ramp to date is low-margin, non-recurring tenant fit-out construction ($18.9M of the $71M HPC line) plus non-cash straight-line rent — recurring base-rent NOI ($44.1M/quarter) has only just begun as the first building energized. The headline growth rate overstates the underlying earnings-power ramp.

Capacity trajectory. Crypto hosting ~286MW (stable, not expanding). HPC: 0MW operating until late 2025 → 100MW operating (PF1 building 1) → ~700–900MW under construction (PF1 buildings 2–4, PF2). Management describes a development pipeline of four marketed sites (~1GW), advanced-stage discussions, and a long-term vision of 5GW+ of critical IT load across campuses by the early 2030s. Contracted backlog built from $0 (pre-May 2025) → $7B (CoreWeave 250MW) → $11B (full 400MW) → ~$16B (+PF2) → higher still with Delta Forge and the June 2026 210MW/~$5.2B hyperscaler lease.

Forward opportunity — large, real, and entirely prospective. The path to management’s self-set “$1B of NOI within 5 years” (with internal $1B and $2B NOI targets) rests on: (1) energizing the ~900MW under construction on schedule (the PF1 RFS date is ~July 2026, ramping data-hall-by-data-hall through the September and November quarters; PF2 first building later in FY27); (2) signing the marketed ~1GW pipeline; and (3) expanding existing campuses (each claimed scalable to ≥1GW). The economics are credible if delivered: at ~$130/kW/month and high-80s% NOI margins on, say, 1GW of leased critical IT load, ~$1B of site-level NOI is arithmetically reachable. The gating items have shifted from power (largely secured) to construction-scale execution and financing.

Quality-of-growth caveat. APLD has changed its central growth narrative repeatedly (see ): the GPU cloud that was pitched as a co-equal growth pillar in 2023 is being spun off at a write-down in 2026. The current landlord growth story is the most coherent and best-financed yet — but the company’s history counsels treating five-year, self-set NOI targets as promotional anchors, not guidance.

Verdict: high potential, low realized quality so far. The growth is real and the runway is large, but to date it is dominated by one-time fit-out and non-cash accruals, and the durable-NOI ramp is still a forward promise gated on execution, leasing, financing, and counterparty solvency.

6. Financial Quality

This is where the bull narrative and the financial reality diverge most sharply. The honest summary: APLD is a deeply cash-consumptive, capital-markets-dependent build whose “Adjusted” metrics materially flatter a loss-making enterprise.

Income statement (Q3 FY26 / 9M FY26). Q3 revenue $126.6M; GAAP operating loss –$85.7M (9M: –$111.6M); GAAP net loss –$70.6M, of which –$100.9M (–$0.36/sh) is attributable to common (the difference is the $31.9M loss allocated to the Macquarie preferred NCI — see below). Management reports “Adjusted EBITDA $44.1M” and “Adjusted net income $33.2M (+$0.09/sh).” The bridge is the story:

• +~$49M of stock-based compensation added back. SBC was $50.1M in Q3 alone and $94.7M over nine months (~27% of revenue). This is recurring, dilutive, cash-economic compensation — adding it back to reach positive “EBITDA” is the single largest flattering adjustment.
• The entire cloud operating loss (–$52.2M, incl. the $59.7M write-down) is excised from Adjusted figures — for a business APLD still ~97% owns.
• Net of an honest treatment of SBC and the retained cloud, economic EBITDA is roughly breakeven, and the enterprise is GAAP loss-making.

Cash flow — the crux of the financial-quality verdict (9M FY26):

• Operating cash flow: –$42.9M. (Net loss –$101.9M; non-cash add-backs SBC $94.7M, non-cash interest $75.0M, D&A $33.6M, the $59.7M write-down; offset by working-capital drains — AP –$171.6M, prepaids/deposits –$88.2M.)
• CapEx: –$1,576.7M (with another ~$564M of PP&E sitting in accruals/AP, non-cash).
• Financing: +$3,620.9M — debt borrowings $2,504.9M, Macquarie preferred +$900M, other preferred +$740M, common ATM +$196.4M (15.3M shares).
• Free cash flow ≈ –$1.62B. The entire ~$2B cash build is financed, not earned. The model works only if capital markets stay open at sub-9% and the contracted leases convert to cash on schedule.

Balance sheet. Cash ~$1.73B + restricted $178.8M; total debt face $2,943M / net $2,693M after $250M deferred financing costs; stockholders’ equity $1.581B; total assets $6.25B. The debt stack:

The $923M Macquarie “preferred equity” is the most important non-obvious item. Raised at the Ellendale SPV (APLD HPC TopCo 2), it sits in temporary equity on the balance sheet but is economically expensive structured debt dressed as equity: a 12.75%/yr dividend (daily accrual, semi-annual compounding, PIK-able through 2035, stepping to a maximum 16.75%), with a 1.80x minimum-MOIC liquidation-preference floor (2.00x on IPO), investor put redeemable after April 2028. It is a senior, accreting claim ahead of the common on the best asset in the portfolio. The $31.9M of 9M net loss allocated to NCI — and $32.0M of PIK dividends — is why the loss to common exceeds the total net loss. Treating this as “equity” flatters both the leverage optics and the apparent residual value of the common.

Depreciation is artificially low — and a margin headwind is coming. Gross PP&E is $3,068.6M but accumulated depreciation is only $56.8M because $1,544.7M (~50% of gross PP&E) is construction-in-progress that is not yet depreciating. As the ~900MW energizes, D&A will ramp hard, pressuring GAAP margins precisely as revenue scales — today’s reported segment profitability is flattered by undepreciated assets.

Lease accounting — a relative positive, with a caveat. HPC hosting is recognized as operating leases, straight-lined under ASC 842 — not sales-type leases, so there is no aggressive upfront revenue front-loading (a genuine relative positive versus what was feared). The caveat: straight-lining escalating 15-year rents builds a large non-cash straight-line-rent receivable that runs ahead of cash collections, and ~38% of Q3 HPC “revenue” is lower-quality fit-out (percentage-of-completion) and power pass-through.

Going concern — cured. APLD carried a going-concern warning in FY24 (cash fell to ~$31.7M against a $1B+ build). The Q3 FY26 10-Q and FY25 10-K carry no substantial-doubt qualification; $2.1B of cash and no near-term maturities have removed the immediate risk — though the parent provides completion guarantees on each project (cost-overrun exposure) and ~$337M of unrecognized SBC plus a $25–40B forward capital need keep the financing treadmill running.

Returns. ROE/ROIC are negative and not yet meaningful (the asset base is largely pre-revenue CIP). The one genuinely high-return line is crypto hosting (~45% asset yield), which is being run off. The HPC franchise’s prospective returns depend on whether ~$130/kW/month leases out-earn a blended cost of capital that includes 9.25% notes and a 12.75%+ preferred — a much higher hurdle than the “high-80s% NOI margin” headline implies once the full capital stack is charged.

Verdict: poor current financial quality, improving but unproven economics. The economics could improve dramatically with scale and refinancing — but today the company does not earn its cash, the “Adjusted” profitability is an SBC-and-exclusion artifact, the capital stack is expensive and senior to the common, and the reported margins are flattered by undepreciated assets and non-cash rent. Economics that “improve with scale” is the thesis, not yet the evidence.

7. Capital Allocation

Philosophy: a high-velocity, high-dilution land-grab. APLD’s capital allocation is the textbook late-capital-cycle behavior the Marathon framework warns about — marshaling every available form of capital to build ahead of anticipated AI-data-center returns. The question is not whether management is aggressive (it plainly is) but whether the aggression builds per-share value or merely scale and market cap.

The dilution record is stark. Share count has roughly tripled since 2022: ~99M (May 2022) → ~157M (Aug 2024) → ~215M (Oct 2024) → ~285.8M (Apr 2026) — up ~82% across the ~20-month AI pivot and +27% in the last nine months alone. The funding history reads as a near-continuous capital raise: FY25 common equity $191.6M + a $450M convertible + $198.2M of preferred; FY24 common $130.8M; a Yorkville $250M SEPA; Series E–G preferred/PEPA instruments with floor-price reset mechanics; the Macquarie project-equity JV (up to ~$5B at ~$2.25M/MW); the SMBC-led $2.35B secured notes at a ~97 OID; Macquarie equipment notes; and a $20M B. Riley-affiliated bridge loan that converted into 8.42M shares. Cumulatively, APLD has marshaled well over $3–3.5B of external capital against a FY25 GAAP loss of ~$234M. There have been no buybacks and no dividend (appropriate for a builder, but it means 100% of per-share value depends on the build out-earning the dilution).

The structural innovation deserves credit. The Macquarie preferred + project-debt model is genuinely clever: by pushing site-level financing into bankruptcy-remote SPVs funded ~80% with non-recourse debt and preferred, APLD can build multi-billion-dollar campuses while keeping >85% of each site’s common equity and limiting parent recourse. The CoreWeave lease restructuring (March 2026) — an A3-rated SPV with a springing parent guarantee, $50M letter of credit, and a cash lockbox that puts APLD first in the payment waterfall — is well-engineered structured finance that genuinely improves the credit and should lower refinancing cost. If the project debt can later be refinanced into investment-grade ABS as Cummins envisions, the cost of capital falls and per-share value is “unlocked.” That refinancing is the explicit bull catalyst — and it is unproven.

Insider behavior — the tell. Wes Cummins made 48 open-market purchases (code P) totaling ~2.03M shares for ~$5.5M at an average of ~$2.72 — but every single buy was between May 2022 and May 2023, when APLD was a sub-$6 Bitcoin miner. There has been zero open-market insider buying during the 10x run from ~$4 to ~$42. Into the rally, insiders have been net sellers via routine code-S sales (Cummins sold tranches at $9.20, $15.26 and $36.42 for ~$10.9M gross; the CFO and several directors also sold into strength). His large code-“J” transfers are non-economic moves among entities he controls (Cummins Family Ltd., 272 Capital LP, his IRA), and his core ~21.9M-share / 7.5% stake is intact — so this is not a flashing-red insider exit. But the signal is unambiguous: the only conviction buying was trough-era and stale; no insider has put fresh cash to work at $30–40.

Compensation — mixed, leaning negative. Cummins’s FY2025 pay was ~$27.7M (versus ~$1.24M in FY2024 — a ~22x jump, almost entirely a $25.46M stock award) in a year the company lost ~$234M. The January 2026 mega-grant (4.5M PSUs + 1.5M RSUs, ~$122M grant-date fair value, intended to replace five years of awards) vests purely on share-price hurdles of $50 / $75 / $100 (90-day average, within five years). The price hurdles are a genuine market condition and do align Cummins with share-price upside — but there is no ROIC, ROE, per-share, NOI, or dilution metric anywhere in the plan. A serial diluter can drive market cap (the metric that pays) via narrative and scale without creating per-share value — exactly the misalignment to fear here. The proxy also sought +15M plan shares and +200M authorized shares, pre-loading further issuance.

Related-party web. A dense B. Riley / 272 Capital nexus surrounds the company: Cummins founded 272 Capital (acquired by B. Riley) and was President of B. Riley Asset Management; a director is ex-CEO of B. Riley Wealth Management; B. Riley was both a 5% holder and a lender (the bridge that converted to equity). Family-member employment and payments to a CFO-family firm also appear. None of the surfaced transactions is individually large, but the cluster warrants ongoing scrutiny.

Verdict: aggressive, creative, and not yet proven to create per-share value. Management has shown real skill at financing (keeping the build alive through a near-death moment and engineering an A3 lease structure) and at executing a first building on time. But the capital allocation is, to date, a dilution-funded scale-grab incentivized by a share-price-only comp plan, with the only insider conviction buying years stale. Whether the serial dilution and expensive capital out-earn their cost — i.e., build per-share value rather than just market cap — is the central, still-open question, and it hinges entirely on the contracted leases collecting at high margins for many years.

8. Changes and Headwinds — Last Two Years

A serial business-model pivoter. APLD’s strategy has shifted repeatedly, which is itself thesis-relevant — evidence of either rare adaptability or chronic narrative-chasing:

• 2021: Ethereum mining in China → killed within ~2 months by China’s crackdown.
• Mid-2021: Pivot to U.S. Bitcoin hosting (~500MW built in ~24 months; Marathon a key customer).
• 2022–2023: Pivot to building HPC data centers; launched its own GPU cloud (“Sai Computing,” 34k+ H100s ordered), pitched at the Oct-2023 Investor Day as a core pillar.
• 2024: Re-cast as a build-to-suit HPC landlord; signed the CoreWeave Ellendale LOI.
• 2025: Put the cloud business under strategic review / held-for-sale — exiting the segment pitched as core 18 months earlier.
• 2025–2026: Cloud spun into ChronoScale via the EKSO reverse-merger; explored REIT conversion; launched the Base Electron power play.

The CoreWeave saga and the near-death moment. The defining two-year arc: FY2024 cash fell to ~$31.7M against a >$1B build (the going-concern period), rescued by a cascade of dilutive/expensive financings; the CoreWeave 250MW lease signed in May 2025, expanded to the full 400MW ($11B) by mid-2025; an NVIDIA relationship and the Macquarie $5B JV (closed Oct 2025) stabilized the funding; the $2.35B 9.25% 2030 notes (Nov 2025) and $2.15B 6.75% 2031 notes (Mar 2026) termed out the project debt; and the March 2026 lease restructuring lifted the CoreWeave SPV from BB to A3.

Recent-events timeline (key dates): CoreWeave 250MW signed 5/28/2025 → full 400MW mid-2025; auditor change (Marcum → CBIZ, 6/2025); Macquarie $5B JV closed 10/2025; $2.35B 9.25% 2030 notes (11/2025); ChronoScale/EKSO term sheet 12/2025 → binding agreement 2/2026; Cummins mega-grant 1/6/2026; PF2 ~200MW IG-hyperscaler lease (~1/2026); $2.15B 6.75% 2031 notes (3/2026); CoreWeave SPV restructuring to A3 (3/30/2026); Delta Forge 1 lease (4/23/2026); Goldman ~$350M revolver (5/2026); ~210MW / ~$5.2B IG-hyperscaler lease (6/8/2026) and $1.59B 7.00% 2031 notes (6/9/2026). Sell-side price targets were raised sharply into the June news (Lake Street $90, Needham $83, Craig-Hallum $79). Short interest has hovered ~30% of float throughout. No short-seller report or material litigation surfaced in the corpus reviewed (open question).

Headwinds. (1) Cost of capital — APLD’s own admission that its capital is “higher than it should be”; the entire bull case hinges on refinancing toward IG ABS, which is not yet done. (2) Counterparty concentration — CoreWeave’s solvency over 15 years (see Risk Analysis). (3) Execution at scale — ~900MW under construction through a North Dakota winter; any cost-overrun hits the parent via completion guarantees. (4) Permitting/moratoria — several target sites face county moratoria; the South Dakota site was paused after a tax exemption failed. (5) Dilution overhang — $337M unrecognized SBC + a $25–40B forward capital need. (6) Macro/rate sensitivity — a 7.3 beta name whose model dies if high-yield/preferred markets freeze.

Verdict: the last two years strengthened the financing and asset base materially while exposing the strategy-stability and equity-claim weaknesses. The company is dramatically better capitalized and de-risked operationally than at the FY24 nadir — but the same period reveals a management team that follows capital-market enthusiasm, an equity stacked beneath expensive senior claims, and a thesis now riding on a single tenant.

9. Risk Analysis

Risk of a catastrophic loss. Lower than for a self-mining or GPU-owning peer, because APLD owns re-tenantable, long-life power-and-shell assets rather than depreciating silicon. But the equity (junior to ~$2.9B notes + $0.9B+ accreting preferred) could be severely impaired in a combined CoreWeave-distress-plus-funding-freeze scenario. A total loss is unlikely given asset coverage; a large permanent impairment of the common is a live, non-trivial tail.

10. Valuation Discussion (Embedded Expectations)

No price target. No recommendation. This section frames what the ~$10B equity value implies.

Where APLD trades. At ~$41.47 (2026-06-11): market cap ~$10.0B (basic), EV ~$11.1B, on ~$319M TTM revenue and ~$44M of quarterly “Adjusted EBITDA.” That is ~30–37x trailing sales and ~7x book — both around the 92nd percentile of APLD’s own valuation history (per own-history valuation percentiles: P/B 91.6th, P/S 91.7th, composite 91.7th). Against the scaled, diversified, investment-grade data-center REITs it aspires to become, APLD trades at ~3.5x their price/sales:

Per-MW and per-NOI lenses. On the ~750MW of capacity backing the $15.5B backlog, EV/contracted-MW is ~$12–18M/MW — at or above the top of the ~$8–15M/MW private-market range for operating assets, for capacity that is still largely under construction and single-anchor. On management’s self-set $1B NOI in five years, EV ~$11.1B implies ~11x / a ~9% forward NOI yield — which would be fair-to-cheap versus REITs at ~4–5% cap rates if the NOI is achieved and if that NOI accrued to the EV holders cleanly. It does not: the $923M preferred (12.75–16.75% accreting, 1.80x MOIC floor) and the 8–9% notes consume the first several hundred million dollars of annual cash flow, so the common’s claim on that $1B NOI is far thinner than the headline yield suggests.

Embedded expectations — what must be true to justify ~$10–12B of equity:

1. Convert the $15.5B backlog on schedule and budget — energize ~900MW of CIP without material cost-overrun or delay.
2. CoreWeave stays solvent and pays full-rate rent for 15 years — ~71% of the annuity.
3. The economics survive the capital stack — $1B+ of site NOI must clear the 12.75–16.75% preferred and the notes and ~$125M/yr of ongoing SBC before the common earns a return.
4. Capital markets stay open and cheapen — fund the ~$25–40B needed for 5GW, and refinance project debt into IG ABS, without ruinous dilution (already +27% shares in nine months).
5. Scarcity rents hold through a capital cycle that is flooding the industry with capacity.

Scenario sketch (illustrative, not a target):

• Bear: a CoreWeave wobble or funding freeze mid-build; re-leasing gap, covenant stress, emergency dilution; the equity re-rates toward asset value net of senior claims — a severe (but, given asset coverage, survivable) impairment.
• Base: the buildings energize and CoreWeave + the IG hyperscaler pay; NOI ramps toward ~$1B over ~5 years; refinancing partially de-risks; but the preferred/notes and continued dilution cap per-share value, and the multiple compresses from ~30x sales toward a scaled-REIT range as growth normalizes.
• Bull: flawless execution + IG-ABS refinancing + a string of new IG-hyperscaler campuses pushes the platform toward 5GW and $2B+ NOI; the cost of capital collapses; the equity compounds as the “unlock” Cummins describes materializes.

The 30.4%-of-float short interest is the explicit variant-perception signal — a large cohort is betting the price already discounts the bull scenario as near-certain. The crowded short also makes APLD dangerous to short outright: real contracts, power scarcity, and positive-news flow can squeeze it sharply.

Verdict: the market is paying a scaled-REIT-plus-hyper-growth multiple for a sub-scale, single-anchor developer whose contracted NOI is real but unrecognized, whose common is deeply subordinated, and whose value depends on a flawless multi-year build and one fragile tenant. The asset story can be true and the equity still expensive — that is the central tension.

11. Variant Perception

Consensus view. The sell-side and the momentum bid see APLD as a uniquely-positioned AI-infrastructure compounder: ~$16B+ of contracted 15-year leases backed by improving credits (CoreWeave now A3 at the SPV, plus an IG hyperscaler), a proven on-time/on-budget builder, a clever capital structure that keeps shareholders’ equity claim high while limiting recourse, a clear path to $1B+ NOI, and 5GW+ of long-term optionality — with price targets recently raised to $79–$90.

Strongest bull case. APLD made the right structural choice (landlord, not cloud), sheds GPU-depreciation risk, owns the scarce input (power) in a multi-year shortage, has contracted annuities most peers lack, and is executing. If the buildings energize, CoreWeave pays, and the cost of capital refinances toward IG, this is an under-appreciated infrastructure annuity in its earliest innings — and the short interest is fuel.

Strongest bear case. The equity is a deeply subordinated, dilution-funded claim on a single-anchor backlog that is contracted but not collected; “Adjusted” profitability is an SBC-and-exclusion artifact; FCF is –$1.6B/9mo and entirely financed; the capital is the most expensive in the complex (9.25% notes, 12.75–16.75% preferred); the anchor tenant is itself FCF-negative and Microsoft-concentrated; the CEO’s only conviction buying was years ago and stale; and the valuation (~30–37x sales, ~7x book, 92nd-percentile own history) prices flawless execution at a likely capital-cycle peak.

The 3–5 assumptions that actually matter:

1. CoreWeave’s 15-year solvency (the SPV improves recovery, not solvency).
2. On-time/on-budget energization of ~900MW (and the next ~1GW).
3. Cost of capital falling toward IG-ABS (the “value-unlock” hinge).
4. The common’s residual after the preferred + notes are charged — i.e., does $1B NOI translate to meaningful per-share value.
5. Rent durability through the capital cycle.

What would falsify each side. Bull falsified: a CoreWeave funding wobble, a construction-cost/timing slip, or a high-yield/preferred market freeze — any one breaks the model. Bear falsified: multiple quarters of CoreWeave + IG-hyperscaler rent flowing into the P&L at high-80s% NOI margins, plus an actual refinancing of project debt into investment-grade ABS — proof the annuity is collectible and the cost of capital is falling as promised.

12. Fact vs. Interpretation

13. Open Questions

1. CoreWeave’s standalone 15-year payment capacity — what is the look-through credit beyond the A3 SPV, and what is APLD’s re-leasing economics if CoreWeave rejects/renegotiates?
2. What share of the $1B target NOI actually reaches the common after the Macquarie preferred (12.75–16.75%), the notes, and ongoing SBC? The single most important undisclosed number.
3. IG-ABS refinancing — concrete timing, size and achievable spread of the promised migration from 9.25%/6.75% project notes to investment-grade ABS.
4. Fully-diluted share count at the relevant NOI horizon, incl. converts, the 14.7M PSU/RSU pool, ATM usage, and any further preferred conversion — what is per-share value on a fully-loaded base?
5. ChronoScale/EKSO — carrying value, spin mechanics, and whether the ~97%-retained loss-maker leaks value or overhangs the stock.
6. Construction cost per MW actual vs. the ~$10M/MW guide across PF1/PF2/Delta Forge, and completion-guarantee exposure on overruns.
7. The unnamed IG hyperscaler(s) — identity/credit, and whether the June 2026 210MW lease is a new campus or a PF2/Delta Forge fill.
8. Power (Base Electron) — guarantee exposure, the ~10% stake’s value, and whether off-balance-sheet power dependence becomes a constraint or a liability.
9. Any short-seller report or litigation — none surfaced in the corpus; confirm.

14. What Must Be True

For the bull case to be right:

• CoreWeave (and the IG hyperscaler) pay contracted rent into the P&L at high-80s% NOI margins for multiple quarters, and the ~900MW under construction energizes on time and on budget.
• APLD refinances project debt toward investment-grade ABS, cutting the cost of capital and demonstrating that $1B+ NOI converts into meaningful per-share value above the preferred and notes.
• Capital markets stay open and the platform scales toward 5GW with decelerating dilution.
• Falsification test: if, four quarters out, recurring HPC base-rent NOI is not visibly ramping at high margins, or no IG-ABS refinancing has occurred, or the share count has again jumped double-digits to fund the build — the “annuity unlock” thesis is broken.

For the bear case to be right:

• A CoreWeave funding wobble, a construction slip, or a capital-markets tightening forces a re-leasing gap, covenant stress, or emergency dilution; the deeply-subordinated common re-rates toward asset value net of ~$2.9B notes + $0.9B+ accreting preferred.
• The “Adjusted” profitability remains an SBC-and-exclusion artifact, economic FCF stays deeply negative, and the cost of capital does not fall.
• Falsification test: if CoreWeave’s credit visibly strengthens (sustained positive FCF, M&A by an IG acquirer, or an IG standalone rating), and APLD prints positive economic free cash flow with the project debt refinanced to IG-ABS spreads, and dilution slows — the bear thesis is broken and the asset story is winning.

15. Source Appendix

Primary sources (full citations in the Source Appendix below):

• APLD Form 10-Q, quarter ended 2026-02-28 (filed 2026-04-08) — financials, debt stack, lease backlog, SBC, Macquarie preferred, segments.
• APLD Form 10-K, FY2025 (ended 2025-05-31) — going-concern history, segment basis, capital structure.
• APLD Q3 FY2026 earnings call transcript (2026-04-08); Analyst/Investor Day (2023-10-12) and prior quarterly calls — strategy, guidance, NOI targets.
• APLD DEF 14A (FY2025 proxy) — compensation, PSU hurdles, related parties, share authorizations.
• SEC EDGAR XBRL (revenue, op income, SBC, equity, assets, cash, share counts); Form 3/4/5 corpus (insider transactions).
• Market-data fundamentals snapshot + own-history valuation index (own-history percentiles, short interest, ownership); Financial news aggregators (June 2026 lease + notes + analyst PT changes).
• Public filings and disclosures of CoreWeave (CRWV), IREN, Nebius (NBIS) and Vertiv (VRT) — counterparty-credit and peer/comp context.

The body of this article carries no investment recommendation and no price target. The only position taken is in the clearly-labeled “Claude’s Take” block, which is the author’s own independent opinion and general information only — not investment advice. Do your own research.

APPENDIX A — Standard Diligence Questionnaire

Applied Digital Corporation (NASDAQ: APLD) — as of 2026-06-12

Supplemental to the research memo. Fact (F) / Interpretation (I) / Assumption (A) labels applied where it matters.

General

What thoughtful questions have other investors asked about this company? The recurring institutional questions: (1) How safe is CoreWeave as a 15-year tenant, and does the A3 SPV structure really de-risk ~71% of the backlog? (2) When does recurring base-rent NOI actually show up in the P&L (vs. one-time fit-out + non-cash straight-line rent)? (3) What is the true cost of capital once the 12.75–16.75% Macquarie preferred and 9.25% notes are charged, and when does the promised refinancing to investment-grade ABS happen? (4) How much more dilution funds the ~$25–40B build to 5GW? (5) Is Wes Cummins an operator or a promoter, given the serial business-model pivots? (6) Why is short interest ~30% of float? (F)

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? There are no GAAP earnings — the company is loss-making (Q3 FY26 net loss to common –$100.9M). The crypto-hosting segment is near a cyclical high (post-halving BTC strength); the HPC segment is at the very start of its ramp (one 100MW building operating). (F/I)

Driven by external environment or internal actions? Both: the AI-capex super-cycle (external) created the demand; the landlord pivot, the contracted leases, and the financing engineering are internal. (I)

How stable are revenues? Contracted HPC base rent is highly stable once collected (15-yr non-cancelable, make-whole, escalating) — but only ~$44M/quarter is yet flowing; ~38% of HPC revenue is non-recurring fit-out + pass-through; crypto revenue is commodity-volatile but small. (F)

Outlook for products/services / market size? Large and growing: hyperscaler capex ~$700B/yr, multi-year power shortage. APLD targets 5GW+ and $1B–$2B NOI. The market is real; APLD’s share of it is the question. (F/A)

Business Quality & Competitive Moat

Is the industry getting more or less competitive? More — capital is flooding the AI-data-center-capacity rung from hyperscalers (self-build), scaled REITs, PE, and every pivoting miner (Marathon late-capital-cycle dynamics). (I)

How profitable is the business (ROIC/ROE)? Currently negative/not meaningful (asset base is ~50% pre-revenue CIP). The crypto line earns ~45% on assets but is being run off. Prospective HPC returns depend on ~$130/kW/mo leases out-earning a 9–13%+ blended cost of capital. (F/I)

How profitable is the industry / barriers to entry? Scaled, diversified DC-REITs are good businesses (Equinix/DLR); the sub-scale, single-anchor developer rung is capital-intensive and commoditizable. Barriers (power interconnect queues, land, EPC capability) are real but erodible head-starts, not franchises. (I)

Can the business be easily understood? Moderately — the landlord model is simple, but the capital structure (project SPVs, 12.75–16.75% preferred dressed as equity, converts, completion guarantees, off-balance-sheet power via Base Electron, the ChronoScale spin) is genuinely complex and obscures the common’s true claim. (I)

Undermined by foreign low-cost labor? No — physical U.S. infrastructure tied to domestic grid power; the binding constraint (interconnected power) is local. (F)

Do brands matter? Nature of competition? Brand is minor; competition is on power availability, time-to-power, execution speed, and cost of capital. APLD competes on early-secured Dakota power + build speed. (I)

Customers’ switching costs? High during the 15-yr term (make-whole contract); low-to-moderate at renewal (shell is re-tenantable, but nothing binds a hyperscaler to renew vs. self-build). (I)

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet? The contracted-lease backlog ($15.5B of future minimum payments) is off-balance-sheet value; the ~10% Base Electron stake and B&W warrants are cost-method/de minimis. (F)

Off-balance-sheet liabilities? Parent completion guarantees on each project (cost-overrun exposure); the Base Electron power guarantee (terminates on $50M raise/IPO); operating-lease and power-purchase commitments. The $923M Macquarie preferred sits in temporary equity but is economically a senior accreting liability. (F/I)

How conservative is the accounting? Mixed. Positive: operating-lease (straight-line) recognition, not aggressive sales-type front-loading; going-concern cured. Aggressive/flattering: “Adjusted EBITDA/net income” add back ~$49M SBC and excise a ~97%-owned cloud loss; D&A artificially low (~50% CIP undepreciating); large non-cash straight-line rent receivable; preferred classified as equity. (I)

How CapEx-hungry? Extremely — –$1.58B capex in 9 months; ~$10M/MW gross build cost; a ~$25–40B forward need for 5GW. This is among the most capital-intensive models in public markets. (F)

Capital Allocation & Management

FCF generation / use / philosophy? Deeply FCF-negative (–$1.62B/9mo), funded by debt + preferred + ATM equity. Philosophy is a high-velocity, dilution-funded land-grab; all capital goes to the build. No buybacks, no dividend. (F)

Significant acquisitions recently? No operating M&A; the cloud (Sai Computing) is being divested into ChronoScale via the EKSO reverse-merger. (F)

Buying back / issuing shares? Issuing aggressively — shares +27% in 9 months, ~+190% since 2022; +15M plan shares and +200M authorized requested in the proxy. No buybacks. (F)

Compensation policy / management motivation? Cummins FY25 pay ~$27.7M (vs. a ~$234M loss); Jan-2026 mega-grant of 6M PSU/RSU vesting purely on $50/$75/$100 share-price hurdles — a real market condition but no ROIC/per-share/NOI/dilution metric. Motivation is aligned to share price/market cap, not capital efficiency — a concern for a serial diluter. Dense B. Riley/272 Capital related-party web. Core ~7.5% Cummins stake intact; but no fresh insider buying into the run. (F/I)

Valuation & Market Data

ADR / MLP / K-1? No — a U.S. C-corp common stock (NASDAQ: APLD); no K-1. (F)

Dividend policy? None (growth-stage builder). (F)

How profitable is the business? Not yet (GAAP loss); “Adjusted” profitability is an SBC-and-exclusion artifact (see Financial Quality). (F/I)

Net income diverging from cash from operations? Both are negative; OCF (–$42.9M 9M) is less negative than net loss due to large non-cash add-backs (SBC, non-cash interest, write-down), but capex makes FCF deeply negative (–$1.62B). (F)

Risks & Downside

What would cause the stock to decline? A CoreWeave funding/credit wobble; a construction cost-overrun or timing slip; a high-yield/preferred market freeze; further heavy dilution; multiple de-rating from ~30–37x sales; a disappointing NOI ramp; revelation that the common’s residual after the senior stack is thinner than the bull case assumes. (I)

Risk of catastrophic loss? The equity could be heavily impaired in a CoreWeave-distress-plus-funding-freeze scenario (it is junior to ~$2.9B notes + $0.9B+ accreting preferred). (I)

Chance of total loss? Low — asset-backed by re-tenantable, long-life power-and-shell assets that limit a zero; but a large permanent impairment of the common is a live tail risk. (I)

Recent News & Events

Has the business environment changed recently? Yes, favorably on the surface: June 2026 brought a ~210MW / ~$5.2B IG-hyperscaler lease and a $1.59B 7.00% 2031 notes deal, plus sharp sell-side PT hikes ($79–$90). The CoreWeave SPV was lifted to A3 (Mar 2026). Underlying tension (subordinated equity, expensive capital, single-anchor concentration) is unchanged. (F)

Significant acquisitions / accounting changes / new markets? ChronoScale/EKSO cloud divestiture; auditor change (Marcum → CBIZ, 6/2025); new geography (Delta Forge 1, southern U.S.); Base Electron power venture (1.2GW gas IPP). (F)

APPENDIX B — Source Appendix

Applied Digital Corporation (NASDAQ: APLD) — Research as of 2026-06-12

Sources prioritized primary-first.

Primary — SEC filings (EDGAR, CIK 0001144879)

1. Form 10-Q, fiscal Q3 2026, quarter ended 2026-02-28 (filed 2026-04-08; apld-20260228.htm). Income statement, segment results (HPC / Data Center / Cloud), cash-flow statement, debt footnote (Note 6), lease footnote (Note 14 — future minimum lessor payments / $15.5B backlog), SBC (Note 12), Macquarie redeemable-NCI preferred, PP&E/CIP, ChronoScale/EKSO contribution agreement, related parties.
2. Form 10-Q, Q2 2026, quarter ended 2025-11-30 (filed 2026-01-08). Prior-quarter comparatives; held-for-sale treatment of cloud.
3. Form 10-K, FY2025, year ended 2025-05-31. Going-concern history, segment basis, capital structure, risk factors.
4. DEF 14A (FY2025 proxy, filed 2025-09-22). Executive compensation, Jan-2026 PSU mega-grant ($50/$75/$100 hurdles), share/authorization requests, related-party disclosures, director backgrounds (B. Riley / 272 Capital nexus).
5. Form 3/4/5 corpus (~191 filings). Insider transactions: Wes Cummins open-market purchases (code P) 2022–23 at ~$2.72 avg; code-S sales into strength 2024–26; code-J intra-entity transfers; officer/director activity.
6. 8-K corpus (financing + lease events). $2.35B 9.25% 2030 notes (Nov 2025); $2.15B 6.75% 2031 notes (Mar 2026); CoreWeave SPV restructuring/A3 (Mar 2026); Delta Forge 1 (Apr 2026); $1.59B 7.00% 2031 notes (Jun 2026); 210MW/~$5.2B IG-hyperscaler lease (Jun 2026).
7. SEC EDGAR XBRL (scripts/edgar.sh concept): revenue (RevenueFromContractWithCustomerExcludingAssessedTax), operating income, share-based compensation, stockholders’ equity, total assets, cash, common shares outstanding — multi-year series FY22–9M FY26.

Primary — earnings calls & investor events (company earnings-call transcripts)

8. Q3 FY2026 earnings call, 2026-04-08 — primary for current strategy, segment detail, $16B backlog ($11B CoreWeave + $5B IG hyperscaler), $1B NOI target, financing model, Base Electron, lease restructuring, cloud spin.
9. Analyst/Investor Day, 2023-10-12 — original strategy; the GPU-cloud “co-equal pillar” pitch (since abandoned).
10. Prior quarterly calls Q2 FY2023 through Q2 FY2026 — strategic-pivot timeline, guidance-vs-delivery scorecard, CoreWeave saga, going-concern period.

Secondary — data & news

11. Market-data fundamentals snapshot + own-history valuation index (2026-06-11): price $41.47, mkt cap ~$10.0B, EV ~$11.1B, P/S ~30x, P/B ~7x, beta 7.27, short interest 80.1M sh (30.4% of float), insiders 11.1%, institutions 66.4%; valuation-index own-history percentiles (P/B 91.6, P/S 91.7, composite 91.7).
12. Financial news aggregators (June 2026): $1.59B 7.00% senior secured notes; 15-yr 210MW IG-hyperscaler lease (~$5.2B); analyst PT changes (Lake Street $90, Needham $83, Craig-Hallum $79).
13. yfinance / fetch.py — price, EV, comp multiples (cross-check only; reconciled to filings).

Analytical frameworks

14. Greenwald & Kahn, Competition Demystified — moat taxonomy (used to classify APLD’s edge as an erodible locational/supply-side advantage without customer captivity).
15. Chancellor (ed.), Capital Returns (Marathon Asset Management) — capital-cycle analysis (used to locate the AI-data-center industry in mid-to-late euphoria).
16. CoreWeave (CRWV), IREN, Nebius (NBIS), Vertiv (VRT) public filings and disclosures — counterparty-credit and peer/comp context.

All URLs accessed June 2026 via SEC EDGAR (sec.gov), market-data providers, and company filings/transcripts. Quantitative figures reconciled to SEC filings; third-party aggregator data used for orientation and own-history percentiles only.