Applied Digital

Refresh delta (since 2026-07-06)

What changed — six material moves:

1. The backlog nearly doubled, de-concentrated, and got audited. ~$23.5B / ~1 GW / 4 campuses / one HPC tenant → $36.2B / 1,410 MW / five campuses / three HPC tenants. Two entirely new southern-region campuses appeared: Delta Forge 1 (300 MW, ~$7.5B, leases signed 2026-04-20/22) and Delta Forge 2 (210 MW, ~$5.2B, signed 2026-06-05), both in Louisiana. Critically, the ASC 842 lessor note discloses audited minimum contracted lease payments of $35.845B — corroborating the marketing figure within ~1%.
2. Cost of debt collapsed by 225bp. New project debt priced at 6.750% ($2.15B, Mar 2026) and 7.000% ($1.59B, Jun 2026, refinancing a $300M Goldman bridge) against the original 9.250% notes; weighted-average term-loan rate 7.1% at year-end. The revolver was upsized to $430M at SOFR+2.25%. This directly answers management question #3 from the prior dossier.
3. Revenue quality is materially worse than the headline. FY26 revenue $611.3M, but $270.6M of it is tenant fit-out construction billed at $258.1M of cost — a 4.6% margin pass-through. The actual landlord line was $114.7M of rental revenue producing $90.4M of NOI at a 91% margin. At year-end only ~100 MW of the 1,410 MW contracted was operating and revenue-generating.
4. Stock-based compensation went to $219.3M from $22.5M — a 10x jump, 41% of adjusted revenue, and the dominant driver of SG&A's +208% rise to $332.1M. GAAP EBITDA was $(65.5)M; adjusted EBITDA of $107.2M is arithmetically what you get after adding that back.
5. Two governance overhangs cleared. The FY2025 material weakness in controls over complex financial instruments was remediated as of 2025-08-31, and CBIZ issued an unqualified ICFR opinion at 2026-05-31. The Weich derivative suit was dismissed 2024-06-05 (without prejudice; never re-filed). The prior dossier flagged neither — both were genuine shelf gaps.
6. The market did not pay for any of it. Price $33.23 (2026-07-05) → $29.40 (2026-08-10), −11.5%; market capMarket capitalisationThe share price multiplied by the number of shares. What the market says the equity is worth. ~$9.45B → $8.57B — despite Q4 revenue of $258.7M against a ~$95.3M consensus. Sell-side also softened: 31 Buy / 0 Hold / 0 Sell at the boundary → 3 Strong Buy / 11 Buy / 2 Hold / 1 Sell, consensus PTConsensus price targetThe average of what published analysts think the share price should be. An opinion poll, not a forecast. $68.15 (range $37.50–$90.00).

What held — does the structural thesis stand?

Yes, and it is stronger than it was. The prior dossier's core frame — "a build-to-suit, take-or-pay data-center landlord whose entire game is tenant credit and cost of capital, not GPUs" — survives intact and every FY26 disclosure confirms it. Three things that were assertions at the boundary are now evidenced: the backlog is a contractual receivable schedule, not a press release; the tenant base is no longer single-name; and the capital cost is falling, not rising. The bear case did not break either — it moved. It is no longer "can they fund it and will anyone rent it." It is now "only 7% of the contracted megawatts are earning rent, cost-to-complete is undisclosed, and FY2027 contractual interest of $377.5M sits against FY2027 contractual rent of $451.1M."

---

Phase A — Understand the business

Company Overview — REFRESHED (materially changed)

The company now runs two reportable segments, not three: HPC Hosting and Data Center Hosting (legacy blockchain). Cloud Services ceased to be a segment after the May 2026 ChronoScale transaction.

The lease portfolio as disclosed:

Every lease is take-or-pay and non-cancellable — "a termination for convenience would require payment of the full remaining contractual value". Each carries three five-year renewal options (PF2 has two); management puts the all-options figure at ~$86B.

Pipeline beyond the contracted book: ~1.5 GW contracted-and-under-construction; an active pipeline of >3 GW of gross utility power; an extended pipeline of >5 GW; plus a partnership with Base Electron (a related party) pursuing ~1.2 GW of front-of-the-meter gas generation in the Dakotas in the MISO region. Management is "actively marketing an additional 1.7 GW" with 250 MW of expansion deals in advanced negotiation.

The CoreWeave lease restructuring (2026-03-30) — read this carefully. The ELN-02 lease was amended to suspend the term for two of four data halls, which were moved to a new lease with CoreWeave Compute Acquisition Co. VIII, LLC ("CoreWeave SPV"); the ELN-03 lease was assigned outright to that SPV, releasing CoreWeave from ELN-03. CoreWeave provided Unconditional Springing Guaranties on both and a $50M letter of credit. On its face this looks like credit weakening — the obligor moved from parent to SPV and the parent guaranty became springing rather than direct. Management's account is the opposite: the SPV structure is what enabled the bond placements at 7%, "225 basis points inside" the 9.25% pricing. Both readings are defensible from the record and I am not resolving them — see Lens 13.

ChronoScale: APLD retains ~97% of ChronoScale, which completed a holdco reorganisation on 2026-07-01 (now ChronoScale Holdings Corporation, Nasdaq: CHRN). ChronoScale committed on 2026-06-04 to divest the legacy Ekso Bionics business, now held-for-sale and reported as discontinued operations.

Supply Chain — REFRESHED (new named counterparties)

The FY26 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes. names its suppliers for the first time, which the prior dossier could only describe generically:

• ABB — electrical equipment. BASX — mechanical and cooling systems. Caterpillar — backup generation.
• The stated mechanism: a standardized "AI factory" design lets APLD consolidate purchasing across a multi-gigawatt pipeline and place bulk orders, supporting repeatable 12–18 month build cycles per 100–150 MW facility.
• The 10-K still concedes the chokepoint: "we rely on a limited number of vendors… some of our contracts provide a single source of materials".
• Power remains the binding constraint, per management directly: "power is definitely top of the list" as the growth governor, with supply chain historically capping the industry at ~700 MW annually and APLD now targeting 1.5 GW.
• New competitor names the 10-K now lists: Digital Realty, Equinix, plus IREN, Cipher, TeraWulf, Hut 8, Riot, CleanSpark, HIVE, Core Scientific, Bitdeer, Galaxy Digital, Fermi Inc., Keel Infrastructure, MARA.

Competitive Advantages — CARRIED with one amendment

Unchanged in substance from the previous dossier (a land-and-power + contract-duration moat, not a technology moat). One amendment: the prior dossier judged bargaining power to be "shifting toward APLD but started from weakness." FY26 settles it — APLD signed three campuses in four months with investment-grade counterparties, and the newest leases carry no tenant warrant of the kind CoreWeave extracted ($85.7M CoreWeave warrant, $121.2M Building-4 warrant, both booked as lease-incentive assets). Paying equity for anchor leases appears to have been a CoreWeave-era cost, not a standing one. That is a real moat datapoint.

Segments — REFRESHED

FY2026 vs FY2025 vs FY2024, from the audited segment note (in $000s):

The single most important line in this refresh is the split inside HPC revenue. Decomposed:

• Tenant fit-out services: $270.602M revenue − $258.147M cost = $12.455M gross profit, a 4.6% margin. This is construction pass-through. It inflates the revenue line and tells you almost nothing about the landlord economics.
• Rental: $114.702M revenue − $56.773M cost = $57.929M; on the company's own NOI definition, $90.388M of NOI on $99.8M of base rent = 91% margin.

So the prior dossier's read of "HPC segment profit margin ~25%" was structurally right but conflated two businesses with a 20x margin difference. The correct frame: APLD is a 91%-margin landlord with a 4.6%-margin general contractor bolted on, and in FY26 the contractor was 70% of HPC revenue. As megawatts energize, mix shifts toward the 91% line — that is the operating-leverage story, and it is real. But it also means reported revenue growth will look worse than economics in years when fit-out tapers.

Data Center Hosting (crypto) is now a single customer with ~1.5 years of contract left (was four customers at the boundary), across 286 MW at Jamestown and Ellendale. Segment profit fell 24% to $48.3M, but $25.0M of the FY25 comparative was a one-off Garden City escrow gain — underlying performance actually improved on a $10.2M revenue increase.

Customer concentration (total revenue):

Customer A is almost certainly CoreWeave (the only HPC tenant recognizing revenue in FY26). Note the tension worth holding: backlog de-concentrated while realized revenue concentration went up — CoreWeave is ~59% of FY26 revenue. Both are true, because the new tenants have not started paying yet.

---

Phase B — Measure performance

Earnings Result — REFRESHED (FY2026 + Q4, quarter ended 2026-05-31)

Q4 FY2026:

Q4 revenue mix: $208.2M services (of which $152.4M tenant fit-out) + $50.6M rental and other. Note that 56% of the entire fiscal year's fit-out revenue landed in Q4 — the beat versus a $95M consensus is largely analysts not modelling construction pass-through, not a demand surprise.

Full-year FY2026:

Conflict surfaced, not resolved: the prior dossier cited FY2025 revenue of $142.3M (continuing operations as then reported). The FY26 10-K states FY2025 revenue of $228.569M. This is not a restatement of economics — it is a recast for the ChronoScale transaction, which pulled the cloud business back into continuing operations. The like-for-like comparative is the adjusted line: $144.193M, which reconciles to the prior dossier's $142.3M within rounding/recast noise. Use $228.6M for GAAP comparatives and $144.2M for core-operations comparatives; do not mix them.

Balance-sheet flags — the important ones:

• Total assets $9.929B (from $1.870B) — a 5.3x expansion in twelve months.
• Cash $1.592B unrestricted + $2.381B restricted (from $43.95M + $72.37M). The restricted balance is escrowed project money, not free liquidity — the prior dossier's "$1.73B war chest" framing should not be carried forward without that split.
• PP&E net $4.236B (from $1.252B).
• Total debt $4.976B carrying / $5.307B face principal (from $0.688B). $4.195B of that principal matures in FY2031 — a genuine refinancing cliff, ~79% of the stack in one fiscal year.
• Redeemable NCI (Macquarie) $1.956B sitting in temporary equity, between liabilities and equity. Total funded under the A&R UPA is $1.8B of an available $5.0B; MIP holds 13.5% of TopCo 2's fully diluted common. Management describes the cost as mid-teens IRR — this is expensive capital ranking ahead of common.
• Total equity $1.725B (from $0.498B). Balance ties: $6.186B liabilities + $2.019B temporary equity + $1.725B equity = $9.929B. ✓
• Shares outstanding 287.88M (from 224.91M) — +28%, materially slower than the +47% the prior dossier flagged. Overhang: 22.35M unvested RSU/PSU + 28.1M warrants + 1.5M preferred-linked ≈ 52M shares, ~18% of current count.
• Cash flow: operating +$89.685M (first positive year; FY25 was $(115.402)M), investing $(2.936)B, financing +$6.877B. CapexCapital expenditureMoney spent on long-lived things — buildings, machines, servers — rather than on running costs. $3.042B.
• Unbilled receivables $43.149M (from $0) — $124.5M of revenue recognized against $81.3M of billings. This is exactly the straight-line/cash-rent divergence the prior dossier said to watch, now quantified. It is modest relative to revenue but the direction is the one to track.

Guidance / tone: no formal financial guidance. But management set a hard, falsifiable target on the call: $1B of NOI within one year, "three years ahead of schedule". Against FY26 NOI of $90.4M that implies an 11x increase in twelve months. Also disclosed: ~$600M of capex expected next quarter. These are the two numbers to hold management to at the FY27-Q1 print.

Market reaction: shares fell ~3% intraday to $26.375 on results day, then rose 6.16% after hours to ~$28. As of 2026-08-10 the stock is $29.40, market cap $8.57B, 52-week range $13.16–$50.73, shares 291.47M. Net of the whole refresh window the stock is down 11.5% while the contracted book grew 54%.

Earnings Calls — REFRESHED (sentiment shift)

transcripts/ still does not exist — no primary transcript was ingested this pass (fool.com and insidermonkey.com URLs both 404'd; Seeking Alpha is paywalled). Content below is from a secondary transcript summary and is labelled as such. This remains the shelf's most persistent gap, now four dossiers running.

The tone arc across the last four quarters:

• FY26-Q1/Q2: "we secured the capital, now we build."
• FY26-Q3 (Apr 2026): validation — first HPC revenue recognized.
• FY26-Q4 (Jul 2026): operator confidence, and a new vocabulary. The pitch has moved from capital access to execution reliability: "all of our construction projects are on time and on budget," contrasted against an industry where "roughly 90% of large projects run late or over budget"; "we signed leases for five campuses, including three in just the past four months".

The genuinely new note is the CFO's, and it cuts against the bulls' framing. Saidal Mohmand said HPC financials "primarily reflect only the initial 100 megawatts that are online" and to expect "a significant step-up." That is a bull statement, but read literally it is also the bear's strongest sentence in the whole filing: after $4.36B of cumulative capex, 100 MW of 1,410 MW is earning rent. Management is telling you the P&L is 7% of the story — which is true in both directions.

Phrases that appeared: "on time and on budget," "$1B of NOI," "sub seven times leverage," "225 basis points inside," "1.7 GW actively marketed." Phrases that disappeared: "going concernGoing concernAn auditor’s formal doubt that the company can fund itself for the next year. It is a warning about survival, not about performance.," "held for sale," "Cloud Services." Sentiment trend: still improving, and now grounded in delivered assets rather than signed paper — which is a materially better kind of confidence than the "arguably euphoric" tone the prior dossier flagged.

Conflict surfaced, not resolved: the summary reports Q4 adjusted EBITDA of $42.4M "from $1 million prior quarter," but the prior dossier recorded Q3 FY26 adjusted EBITDA of $44.1M. The "$1M" figure is most plausibly the prior-year quarter (FY25 full-year adjusted EBITDA was only $19.6M), but I cannot confirm that from the sources on hand. Do not cite the $1M without checking a primary transcript.

Comps — CARRIED, not refreshed

Peer multiples were not sourced this pass — the WebSearch budget was exhausted before the comp sweep. See the previous dossier Lens 7 for the peer set (CORZ, IREN, NBIS, CIFR, WULF, HUT). The one comp figure that updates cleanly: APLD's own market cap is now $8.57B against $35.845B of audited contracted base-term rent, i.e. roughly 0.24x gross backlog on equity, or ~0.35x on an EV of ~$11.95B. At the boundary the equivalent EV/backlog was ~0.51x. The backlog got cheaper by roughly a third in five weeks. All peer multiples remain n/a; do not infer any.

Stock-Price Catalysts — REFRESHED

New events inside the window (2026-07-06 → 2026-08-10):

• 2026-07-27 — FY26-Q4 print. Revenue $258.7M vs ~$95.3M consensus; adj. EPS $0.04 vs $(0.19). Shares −3% intraday to $26.375, then +6.16% after hours to ~$28. A 171% revenue beat that did not produce a sustained up-move — the clearest signal in this refresh that the market is not valuing this name off reported revenue.
• 2026-07-29 — FY2026 10-K filed. Audited $35.845B lessor minimum lease payments; clean ICFR opinion; material weakness remediated.
• 2026-08-07 — director sold $2.34M of stock; insiders bought $0 over the trailing three months.
• 2026-08-09 — an analyst raised a target to $37.50 — notably the low end of the range, and the consensus structure degraded to 3 Strong Buy / 11 Buy / 2 Hold / 1 Sell (from 31/0/0 at the boundary), consensus PT $68.15.
• Short interest 27.02% of float, 4.11 days to coverDays to coverHow many normal trading days it would take for everyone betting against the stock to buy back in. A high number means the exit is crowded., down 1.58% month-on-month — essentially unchanged; the shorts have not covered into the good news.
• Newly sourced, filling a prior gap: insider ownership 9.50%, institutional 65.67%. The prior dossier could not source Cummins's stake.

Cumulative move over the window: $33.23 → $29.40, −11.5%. What the market reacts to is unchanged — lease signings, tenant credit, and financing events — but the FY26-Q4 non-reaction adds a new datapoint: the market has stopped rewarding backlog announcements, which is what you would expect once ~$36B is already in the price and the question rotates to execution and cost-to-complete.

⚠️ Incomplete. A general news sweep for this window was not run (WebSearch budget exhausted). Any event that did not surface through the four fetches above is missing from this list.

---

Phase C — Judge people & books

Management — PARTIALLY REFRESHED

Carried from the previous dossier on archetype and history (Cummins as founder/financial-engineer with B. Riley ties; combined Chairman + CEO; CFO transition to Saidal Mohmand). Three updates:

• New President. Jason Zhang, co-founder and Chief Strategy Officer, was appointed President on 2026-01-15. The prior dossier did not have this. It partially addresses the "one dealmaker running 1+ GW of live infrastructure" concern by adding an executive layer — though Zhang is a strategy, not operations, background.
• Skin in the game now sourced: insider ownership 9.50% — meaningful for a founder-CEO. Offset: a director sold $2.34M on 2026-08-07 with zero insider buying in three months.
• The capital-allocation record improved this year, on the evidence. Refinancing from 9.25% → 6.75%/7.00%, repaying the 8% MEC and DevCo notes in full, retiring the SMBC loan, converting all Series G to common (zero outstanding at year-end), and funding $1.8B from Macquarie rather than common equity — Mohmand's stated framework is to avoid dilutive common and convertibles and hold "sub seven times leverage". DilutionDilutionIssuing new shares, so each existing share owns a smaller slice of the same company. decelerated from +47% to +28%. That is a real, measurable improvement over the FY25 record.
• Employees: 256 full-time at year-end. Worth sitting with: $219.3M of stock compensation across 256 employees is ~$857k per head. The 10-K attributes it to headcount growth and performance awards; the tax note discloses $37.952M of "excess officer's compensation" as a non-deductible reconciling item. That is a large number for a company with a $90M NOI.

Forensic Red Flags — REFRESHED

Resolved since the boundary — credit where due:

• Material weakness remediated. FY25 had a material weakness in controls over complex financial instruments (the prior dossier missed this entirely). Remediated as of 2025-08-31 via additional accounting hires and a third-party consultant; CBIZ issued an unqualified opinion on ICFR at 2026-05-31.
• Derivative suit gone. Weich v. Cummins (Nevada) was dismissed 2024-06-05 for failure to plead demand futility; plaintiff has not sought leave to amend.
• Backlog corroborated. The marketing number is now backed by an audited ASC 842 lessor schedule. This retires the strongest version of the "backlog is marketing" bear point.

Still live, and in two cases sharper than before:

• Stock-based compensation is the dominant non-GAAP adjustment, and it is enormous. $219.3M, 41% of adjusted revenue, 10x YoY. The entire gap between GAAP EBITDA of $(65.5)M and adjusted EBITDA of $107.2M is $172.8M of add-backs, of which SBC is the bulk. Adjusted EBITDA of $107.2M is not cash generated; it is the loss with the compensation added back. Anyone quoting the 18% adjusted EBITDA margin without that sentence is quoting a number that means less than it appears.
• Fit-out revenue flatters the top line. $270.6M of FY26 revenue at a 4.6% margin. Revenue "+167%" is true and materially overstates the economic progress.
• A June lease inside a May table. The Item 1 portfolio table is captioned "As of May 31, 2026" and totals 1,410 MW / $36.2B — but that total includes Delta Forge 2, whose lease Note 21 says was signed 2026-06-05, five days after year-end. Immaterial to valuation; material as a tell about which direction presentation errors run.
• "No critical audit matters." CBIZ determined there were none on a balance sheet carrying Level 3 warrants, a $1.96B redeemable NCI with embedded derivatives valued by binomial lattice, $199.4M of lease-incentive assets, and first-year lessor accounting. That is a surprising conclusion and worth noting, though it is an auditor judgment, not a finding against the company.
• Lease-incentive assets grew to $199.4M (from $84.4M), including the $121.2M Building-4 warrant. These amortize against revenue over the lease term — the prior dossier's watch item, confirmed and growing.
• Related-party web widened. Base Electron is a related party and an unconsolidated VIE: APLD holds a $2.0M equity investment plus a $58.6M promissory note (max exposure $60.6M) and had guaranteed Base Electron's obligations to Babcock & Wilcox under a 1.2 GW design-build agreement — guarantee terminated 2026-05-29 upon a $37.0M payment by or on behalf of Base Electron. Separately, $75.8M of FY26 "gain on change in fair value of derivatives" is largely a $89.2M mark on a Babcock & Wilcox warrant — i.e. a meaningful chunk of the year's non-operating income came from marking a warrant in a related-party-adjacent counterparty. This is not an accusation; it is a structure that deserves a direct question (Lens 14).
• The FY2031 maturity wall: $4.195B of $5.307B principal due in one fiscal year.

Regulatory findings (required):

• SEC Litigation Releases / AAERs: none. regulatory/regulatory-findings.md is unchanged since 2026-07-06 — it was not re-run this pass, so its coverage window still ends 2026-07-06. Nothing in the FY26 10-K suggests an SEC matter.
• Securities class action — unchanged. McConnell v. Applied Digital, No. 3:23-cv-01805 (N.D. Tex.), naming the Company, Cummins and then-CFO Rench. Still stayed and administratively closed since 2025-09-08 pending the motion to dismiss. Company states it cannot estimate a range of loss.
• Derivative action — dismissed (see above).
• Non-SEC enforcement: none disclosed in the 10-K. A fresh web check was not performed (budget exhausted) — carried from the boundary as "none found."

---

Phase D — Project & stress-test

Forward Projection — REFRESHED

Model integrity — read before any number below. our model was re-run after populating our figures. It reports: opening balance sheet sourced = yes, but only 1 quarter of history, and our model returns "No computed values. The workbook has 51 formulas and no cached results." It also seeded gross margin at 50%, operating margin at 25%, revenue growth at 8%, and capex at 5% of revenue — against an actual FY26 capex of 498% of revenue. The workbook produces no citable output and its seeded assumptions are wrong for this business by an order of magnitude. Nothing below comes from it. Every figure is either or with arithmetic shown. No our model create was run (unattended).

This is the projection that matters, and it is contractual rather than estimated:

These exclude reimbursements, variable rent escalators, and all renewal options. They are the floor, not the expectation.

The crunch is FY2027, and it is arithmetic, not opinion:

• FY2027 minimum rent $451M against FY2027 scheduled interest $378M → 1.19x coverage before a single dollar of operating expense, SG&A, or D&A.
• Apply the 91% NOI margin: NOI ≈ $410M. Less $378M interest = ~$32M, before corporate SG&A (which ran $332M in FY26, ~$113M ex-SBC), before D&A, and before the Macquarie preferred's mid-teens return.
• FY2027 is therefore a GAAP loss year with near-certainty, and cash-tight, bridged by the $1.59B unrestricted cash, the $2.38B restricted (already committed to construction), the remaining ~$3.2B of Macquarie capacity, the $2.0B PEPA, and the $430M revolver.
• From FY2028 the picture inverts: $1.454B rent → ~$1.32B NOI vs $399M interest = 3.3x. From FY2029, ~$2.05B NOI vs $359M interest = 5.7x. The model is not "does this work" — it is "can they get from here to FY2028 without a dilutive raise."

**EPS path ** — shares ~288M outstanding, ~52M of overhang:

• FY2027: $(0.70) to $(1.10), base ~$(0.90).
• FY2028: $(0.15) to +$0.35, base ~+$0.10 — the crossover year, one year later than the prior dossier's estimate, because the contractual rent schedule is now known and FY27 rent of $451M is well below the prior dossier's $800M–$1.0B FY27 revenue estimate. This is the most consequential correction in the refresh.
• FY2029: $1.00 to $2.20, base ~$1.60.

Bull / base / bear on FY2029 GAAP EPS:

• Bull ~$2.20+: all five campuses energize on schedule, the 1.7 GW being marketed converts, the FY31 wall refinances inside 6%, options push backlog toward $86B.
• Base ~$1.60: contracted 1,410 MW energizes roughly on the disclosed schedule; no new large campus; modest further dilution.
• Bear <$0 through FY2029: an interconnection or transformer slip pushes two campuses right; cost-to-complete overruns force PEPA draws at depressed prices (the Series G floor mechanism reprices downward at the board's discretion); a tenant defers.

The number nobody has disclosed, and it is the biggest one. Cost-to-complete for the contracted 1,410 MW appears nowhere in the 10-K. The filing says only that "significant investments in property and equipment will remain throughout fiscal year 2027." Bounding it: PP&E net is $4.236B for a portfolio ~7% energized; management guided ~$600M for the next quarter alone and said spending rises from there. At a rough $8–12M/MW for shell-and-power (tenant supplies GPUs), 1,410 MW implies $11–17B of total build cost, against ~$4.4B of cumulative capex spent. That leaves roughly $7–13B to fund, versus identified capital of ~$9.6B. It is plausibly funded and it is plausibly not — and the honest statement is that the disclosure does not let an outsider tell. This is management question #1 now.

Bull vs Bear — REFRESHED

Bull case. APLD now owns 1,410 MW of non-cancellable, take-or-pay, 15-year contracted capacity across five campuses and three investment-grade-or-better tenants, backed by an audited $35.845B receivable schedule — and termination for convenience requires payment of the full remaining contractual value. FY26 proved four things that were assertions a year ago: the buildings get delivered on time, the rent recognizes at a 91% NOI margin, operating cash flow turned positive (+$89.7M), and the capital markets will fund it at 7% and falling rather than 9.25%. Controls were remediated with a clean ICFR opinion, the derivative suit is gone, dilution decelerated from +47% to +28%, and Macquarie is funding the build at the project level rather than the common. Steady-state contracted rent is ~$2.3B/year against ~$300M of interest. The stock is down 11.5% while the contracted book grew 54%, putting EV at roughly 0.35x gross contracted backlog.

Bear case — things that can permanently impair:

1. Seven percent. 100 of 1,410 MW is earning rent. Everything else is a construction project with undisclosed cost-to-complete and $377.5M of contractual interest due in FY2027 against $451M of contractual rent. The entire equity value sits in the gap between what is built and what is promised.
2. The FY2031 wall. $4.195B of $5.307B principal matures in one fiscal year. That refinancing happens in whatever credit market exists in 2030–31, and the whole structure assumes it is open.
3. The compensation. $219.3M of SBC on 256 employees, $37.95M of it non-deductible excess officer compensation, converting a $(65.5)M GAAP EBITDA into a $107.2M "adjusted" one. If the AI-infrastructure tape cools, that expense does not fall — but the currency it is paid in does, forcing cash comp or more shares.
4. Realized concentration got worse, not better. CoreWeave is ~59% of FY26 revenue, and its lease obligor was moved to an SPV with a springing — not direct — parent guaranty.

Pre-mortem (18 months out, thesis broke): an interconnection or transformer slip pushed Polaris Forge 2 and Delta Forge 1 delivery right by two to three quarters; cost-to-complete came in above plan; the company drew the PEPA into a falling tape (the Series G floor price is board-adjustable downward), reversing the dilution improvement; FY2027 rent of $451M failed to cover interest plus opex; and a market that had already stopped paying for backlog announcements re-rated the equity from "contracted landlord" to "half-built developer." The buildings were still real. The equity was still repriced.

Are multiples too high? On GAAP earnings, meaningless (losses). On EV/contracted backlog the stock got ~30% cheaper during a window in which the business measurably de-risked. The bull/bear gap is now almost entirely execution and cost-to-complete, and much less counterparty credit than it was five weeks ago — that is the single largest change in the shape of the argument.

Contrarian view (what the market refuses to see). At the boundary the contrarian point was "this is an industrial REIT wearing an AI multiple." That trade has partly happened — the multiple compressed 30% on backlog while the fundamentals improved. The live contrarian point is now the inverse and subtler: the market is treating a 91%-NOI-margin, non-cancellable, 15-year contractual receivable schedule as though it were an equity growth story, and pricing it off a GAAP loss line that is dominated by stock compensation and by depreciation on assets that are not yet earning. If you underwrite the audited $35.8B and the tenants pay, the correct comparison is a levered net-lease REIT, and 0.35x gross contracted rent is not obviously expensive. The catch — and it is a real one — is that a net-lease REIT owns finished buildings, and APLD owns a $10B-ish construction project it has not told you the cost of.

Devil's Advocate — REFRESHED

Dismantling the bull case. Short interest is 27.02% of float and has not covered through a doubled backlog, a 407% revenue quarter, a clean audit, and a 225bp funding improvement. Shorts staying put through that much good news is information.

• The strongest short argument is now a schedule, not a story. Bulls cite $36.2B. The audited table says $451M arrives in FY2027 — 1.3% of the backlog — against $377.5M of interest. 97% of the backlog is FY2029-and-beyond, which means the equity is being asked to underwrite three-plus years of flawless construction, on undisclosed cost, funded by capital that does not yet exist on the balance sheet.
• The revenue beat was mostly construction. $258.7M vs $95.3M consensus reads as a blowout; $152.4M of it was tenant fit-out at ~5% margins. The market saw through this — the stock fell 3% intraday. Anyone quoting +407% is quoting a general contractor's revenue.
• Adjusted EBITDA is a compensation adjustment. GAAP EBITDA $(65.5)M → adjusted $107.2M. The company also defines its non-GAAP measures to exclude ChronoScale entirely — a consolidated subsidiary it owns 97% of. Two layers of exclusion between GAAP and the number in the headline.
• The SPV restructuring cuts both ways and management only tells one side. CoreWeave was released from ELN-03; ELN-02's obligor moved to an SPV with a springing guaranty and a $50M LC. Management frames it as the enabler of 7% bond pricing — plausibly true. But a springing guaranty on an SPV is structurally weaker recourse than a direct parent obligation, and the bondholders got the benefit while the equity got the counterparty change. The 10-K does not quantify what "springing" requires.
• Related-party density is increasing, not decreasing. A $58.6M loan to Base Electron, a $2.0M equity stake in it, a terminated $37M-settled B&W guarantee on its behalf, and an $89.2M mark-to-market gain on a B&W warrant driving most of the year's non-operating income. Layer that on the standing B. Riley history and the live McConnell allegation that the board is not independent under Nasdaq rules.
• The dilution improvement may be a bull-market artifact. Shares +28% (better than +47%). But Series G's floor price was raised four times during FY26 as the stock ran ($4.25 → $12.50 → $22.00 → $34.00), and the board holds sole discretion to lower it. The PEPA commitment was just raised to $2.0B. In a drawdown, that mechanism converts at lower prices into a weaker tape — precisely when it is most dilutive.
• What must hold for today's price: (1) 1,410 MW energizes roughly on the disclosed 2H26–1H28 schedule; (2) cost-to-complete lands inside identified capital; (3) all three tenants pay take-or-pay for 15 years; (4) $4.195B refinances in FY2031. Break any one and the equity de-rates, and the FY27 coverage of 1.19x leaves no cushion for the first one.
• Single scenario that permanently impairs: a two-to-three-quarter delivery slip across two campuses while cost-to-complete overruns, forcing a large PEPA/equity draw into a weak tape. Note this is a different bear thesis than the boundary's, which was a CoreWeave credit event. Tenant credit genuinely improved this year; execution and funding risk did not. A short who is still short the old thesis is short the wrong thing.

Conflict flagged, unresolved: the transcript summary reports ">80% of capacity leased to a single high-grade hyperscaler". The 10-K's own campus table cannot support that: the second hyperscaler holds PF3 (300 MW) + DF1 (300 MW), plus DF2 (210 MW) if it is the same entity — 57% at most, and PF2's 200 MW is attributed to a different investment-grade hyperscaler. Either the summary is a garbled paraphrase or tenant identities overlap in ways the filing does not disclose. This is a material discrepancy about the single most important risk in the name and it should be resolved against a primary transcript before any position is sized.

Management Questions — REFRESHED priority order

The prior dossier's 15 questions stand (see the previous dossier Lens 14). Three were answered by FY26 — Q3 (refinancing the 9.25%: done, at 6.75%/7.00%), Q8/Q9 partially (derivative suit dismissed, material weakness remediated), and the backlog verification. The reordered top five for the FY27-Q1 call:

1. What is total remaining cost-to-complete for the contracted 1,410 MW, campus by campus, and does identified capital fully fund it without a common-equity or PEPA draw? (The single largest undisclosed number in the company.)
2. FY2027 minimum contracted rent is $451M against $377.5M of scheduled interest. Walk through FY2027 liquidity month by month. What is the trigger that would cause you to draw the PEPA, and at what share price does that stop being accretive?
3. $1B of NOI within one year implies an 11x increase from $90.4M. Which specific megawatts, at which campuses, on which dates, produce it?
4. Define "springing" in the CoreWeave ELN-02/ELN-03 guaranties. What conditions cause the parent guaranty to attach, and why was CoreWeave released from ELN-03 rather than remaining jointly liable?
5. $219.3M of stock compensation across 256 employees, $37.95M of it non-deductible excess officer compensation. What is the FY2027 SBC plan in dollars, and what happens to it if the share price halves?

---