TeraWulf

Phase A — Understand the business

Lens 1 · Company Overview

TeraWulf is, on paper, a bitcoin miner. In economic reality as of Q1 2026, it is a vertically-integrated power-and-datacenter landlord that has pivoted to renting critical-IT capacity to AI/HPC tenants, with a shrinking bitcoin-mining tail attached. The Q1 2026 10-Q states it plainly: "During the three months ended March 31, 2026, HPC leasing revenue represented a majority of total revenue for the first time".

How it makes money — two segments (10-Q Note 17):

• HPC Leasing — long-term datacenter leases (10–25 yr) where TeraWulf delivers power + cooling + space to AI tenants. Q1 2026 revenue $21.0M, segment profit $10.2M. Accounted for as ASC 842 operating leases.
• Digital Asset Mining — hash-computation services sold to the Foundry USA mining pool (FPPS model), bitcoin earned and routinely converted to USD. Q1 2026 revenue $13.0M, segment profit $10.2M. Down from $34.4M a year ago — the company is curtailing and repurposing miner buildings to make room for HPC.

The asset base. Flagship is the Lake Mariner Data Campus in Barker, NY — a retired coal plant repurposed in 2022, sitting on NYISO Zone A low-cost/low-carbon power, scalable to ~500 MW near-term and up to 750 MW subject to NYISO approval. 90 MW is contracted from the New York Power Authority on a 10-yr deal.

Key customers (extreme concentration):

• Core42 (a G42 affiliate) — 60 MW critical IT, fully delivered as of Q1 2026.
• Fluidstack — the anchor. 378 MW at Lake Mariner (Akela subsidiary) + 168 MW at the Abernathy, TX JV = ~546 MW contracted to a single counterparty. Backstopped by Google.
• For Q1 2026, "the Company's HPC lease revenue was generated from one customer".

Key supplier / pool: one ASIC supplier (Bitmain) and one mining pool (Foundry USA) — also single-source.

Contract structure & payment terms. HPC leases are long-dated (10-yr Lake Mariner with two 5-yr options; 25-yr at Abernathy with escalators). Power is a pass-through at cost, no markup, grossed up into lease revenue. Core42 prepaid 12 months' base rent ($90.0M). The economic kicker: leases "benefit from investment-grade credit support" — i.e., the Google backstop — which the company explicitly says is what makes the contracted revenue bankable.

Headline contracted backlog ``: 522 critical MW under contract for >$12.8B of revenue. Note the gap vs. the filing: only $1,029.0M of lease payments is "commenced" (Core42's 60 MW live) per the maturity schedule — the other ~$11.8B is contracted-but-not-yet-commenced (Fluidstack), entirely contingent on construction completing and leases commencing in 2026+.

Lens 2 · Supply Chain

Map the chain upstream → TeraWulf → end-demand, names attached:

Upstream (inputs):

• Power — NYISO Zone A grid; NYPA (90 MW, 10-yr); on-site generation being added via the Morgantown, MD (210 MW power plant, FERC approval pending) acquisition. Chokepoint: grid interconnection + NYISO approval to scale Lake Mariner past 500 MW.
• ASIC miners — sole supplier Bitmain (S19/S21 family; 54,100 owned, 35,500 operational; avg 17.2 j/th). Single-source.
• Land — owned or long-term ground-leased. Lake Mariner ground lessor is Somerset Operating Co. / Cayuga (Prager-affiliated related parties); Cayuga site (183 acres, NY) on an 80-yr lease.
• GPUs — TeraWulf does not buy the GPUs; its tenants (Fluidstack, Core42) own the compute. TeraWulf is the shell + power. (Fluidstack's GPUs are financed up to $10B via Macquarie.)
• Capital — the real input. Lenders/counterparties: Cantor Fitzgerald, Morgan Stanley (note underwriters), Google (warrant-pledge collateral + backstop), the 2030 Secured Notes holders, convertible holders.

TeraWulf (the midstream): converts advantaged power + land + interconnection into energized critical-IT MW. Operates via subsidiaries La Lupa and Akela (Lake Mariner), Big Country Wulf / FS CS I (Abernathy JV), Hawesville Developer.

Downstream (end-demand):

• TeraWulf → Fluidstack (neocloud) → Anthropic (Fluidstack's reported $50B customer) and other AI labs.
• TeraWulf → Core42 / G42 → enterprise/sovereign AI.
• Mining side: TeraWulf → Foundry USA pool → bitcoin network.

Single-source dependencies (the whole risk in one line): one anchor tenant (Fluidstack), one credit backstop (Google), one flagship site (Lake Mariner), one ASIC vendor (Bitmain), one mining pool (Foundry). This is a chain of single points of failure held together by Google's balance sheet.

Lens 3 · Competitive Advantages (moats)

What's genuinely advantaged:

1. Power + interconnection at a retired-plant brownfield. Lake Mariner's existing transmission infrastructure and NYISO Zone A position are scarce and hard to replicate — interconnection queues are multi-year. This is the real moat: controllable, low-cost, low-carbon power at scale with a live interconnect..
2. In-house power/grid expertise from the Beowulf E&D acquisition (94 employees, power-development DNA) — differentiates vs. pure-play datacenter developers who must outsource energy.
3. Credit-enhanced contracts. The Google backstop converts a speculative tenant (Fluidstack) into bankable, 10–25-yr, project-financeable revenue. That financial engineering is a moat of sorts — it's why WULF can borrow $3.2B at 7.75% against half-built buildings.

What is NOT a moat:

• There is no technology moat, no switching cost on the mining side, no network effect, no brand. The bitcoin fleet is 0.9% of global hashrate — a price-taker.
• Bargaining power is inverted: TeraWulf needs Fluidstack/Google far more than they need TeraWulf. Google holds ~14% pro-forma equity and the warrant-pledge collateral on the senior notes; Fluidstack has optionality (term-contraction rights in the Abernathy lease). The customer holds the whip.

Verdict on moat: real but narrow and site-specific (the power position), wrapped in financial engineering that is only as durable as the counterparties. Not a franchise — an infrastructure position.

Lens 4 · Segments

`` — Q1 2026 vs Q1 2025 ($000s):

The trend — a clean handoff: total revenue is flat YoY ($34.0M vs $34.4M) but the mix inverted entirely. Mining revenue fell −62% (172 BTC mined vs 372 BTC, fewer machines running) while HPC went from zero to the majority. Segment profit nearly tripled ($20.4M vs $7.0M) — HPC leasing carries far better unit economics than mining once the power is pass-through.

Geography: 100% US. Lake Mariner (NY) is essentially all current revenue; Abernathy (TX) and Hawesville (KY) are pipeline. The geographic diversification narrative is forward-looking, not yet in the P&L.

The catch the segment table hides: segment profit ($20.4M) is a rounding error against the $196.2M of total costs below it. The two segments are profitable at the gross level; the company lost $427.7M in the quarter (see Lens 5). The segments measure the operating business; the holdco measures the financing scheme.

---

Phase B — Measure performance

Lens 5 · Earnings Result (Q1 2026, the latest print)

`` — three months ended 3/31/2026 ($000s):

• Total revenue $34,012 (Digital asset $12,990 + HPC lease $21,022) — flat YoY.
• Operating loss $(162,142) vs $(59,628) — blown out by:
• SG&A $127,605 (vs $46,573) — of which ~$101.4M is stock-based comp. SBC alone is 3x revenue.
• Impairment of PP&E $25,697 — writing down miner buildings being repurposed for HPC.
• Depreciation $28,477 (incl. $11.9M accelerated on shortened-life miners).
• Below the line — where the real loss lives:
• Interest expense $(67,071) vs $(4,049) — the $5.3B debt load arrives.
• Change in fair value of warrants $(216,325) — the Google warrants are a mark-to-market liability; as the stock ripped, the warrant liability ballooned and flowed through as a non-cash loss. This single non-cash line is the largest driver of the loss.
• Interest income $29,411 (on the $2.6B cash pile).
• Equity in net loss of investee $(11,548) — the Abernathy JV.
• Net loss $(427,703); net loss attributable to TeraWulf $(427,634); EPS $(1.01) vs $(0.16).

Balance-sheet flags (the heart of the matter) ``:

• Total assets $7,008.8M; total liabilities $7,086.4M → negative stockholders' equity $(77.6)M (was +$140.4M at 12/31/25). The company flipped to a deficit in one quarter, driven by the $427.7M loss.
• Cash $2,630.0M + restricted $462.7M = $3,092.7M total.
• PP&E net $2,582.2M — nearly doubled from $1,507.7M QoQ (the HPC buildout).
• Gross debt ~$5.29B on B/S (Long-term $3,060.2 + Convertibles $1,597.3 + Short-term convertibles $490.4 + Short-term debt $98.6 + current LTD $43.6). Total principal maturities $5,725.0M. Plus the off-balance-sheet $1.3B Abernathy JV notes (equity-method).
• Warrant liabilities $1,061.0M — the Google warrants.

Cash flow: operating cash flow $(17.6)M (the business consumes cash); investing $(712.8)M (capex $522.9M + asset acquisition $201.4M); financing +$100.4M. Net cash fell $630.0M in the quarter — they are spending the war chest fast.

Market reaction ``: the stock is up 187% LTM and trades near 52-wk highs ($28) — the market is pricing the contracted backlog and the HPC narrative, not the GAAP loss. The $427M loss is being correctly read as mostly non-cash (warrants + SBC + impairment). What's priced in: execution on 522 MW.

Unusual vs. own history: negative equity, $216M warrant swing, $101M quarterly SBC, and a negative net cost-to-mine (see Lens 6) are all firsts. This is no longer the same company it was a year ago.

Lens 6 · Earnings Calls / operational sentiment trend

No transcripts on the shelf (transcripts/ empty); I read management's tone from the MD&A across the three filings and supplement.

The narrative arc (FY25 10-K → Q3'25 10-Q → Q1'26 10-Q): a steady, deliberate hardening of the pivot language. By Q1 2026 the MD&A leads with "vertically integrated owner, developer, and operator of digital infrastructure… purpose-built to support HPC workloads, including AI." Bitcoin is reframed as "legacy." Recurring phrases: "infrastructure-first," "Tier-1 counterparties," "credit-enhanced," "long-dated contracted revenue," "disciplined capital allocation," "project-level financing.".

What they stopped saying: the bitcoin-maximalist, hashrate-growth, "lowest-cost miner" framing has been demoted to a paragraph. Mining is now described in terms of curtailment and repurposing, not expansion.

Operational color worth flagging:

• Q1 2026 net cost to mine one bitcoin: $(381) — negative. Winter polar-vortex curtailment + $14.1M demand-response credits exceeded gross power cost. Demonstrates the power-asset optionality beautifully — but it's a weather artifact, not a run-rate.
• Mining: 1 BTC/day, 0.9% of global hashrate, fleet avg age 1.6 yr.

Tone: confident, institutional, "we are an infrastructure company now." The risk is the gap between the polished narrative and the half-built reality (only ~25% of industry-leased AI capacity has actually been delivered ``).

Lens 7 · Comps

Peer set = the bitcoin-miner-to-AI cohort. Multiples are `` with heavy dispersion across sources and dates (June 2026); I label rather than fabricate.

EV/Sales, EV/EBIT, P/E, div yield, 5-yr avg ROE: n/a. These names are pre-profit on AI revenue (HPC only began contributing in mid-2025), carry huge convertible loads (debt makes EV ≠ market cap by billions), and have negative or meaningless trailing earnings — so a clean P/E or EV/EBIT multiple is not reliably sourceable and a forward number would be a fabrication. The honest framing: the cohort trades on contracted-MW and $/MW-of-backlog, not on earnings. On that basis WULF's ~$14B cap against 522 MW / $12.8B contracted backlog is roughly in line with IREN and richer than CIFR/CORZ per delivered MW — but with the least delivered (60 MW live) and the most balance-sheet leverage.

The comp that matters most: IREN — same pivot, but anchored by Microsoft (a hyperscaler tenant, not a venture neocloud) and with AI revenue actually ramping ($33.6M in FQ3'26, +94% QoQ) ``. WULF's anchor is Fluidstack-backstopped-by-Google — a structurally weaker direct counterparty at a similar valuation. That is the relative-value flag.

Lens 8 · Stock-Price Catalysts (>5% moves, last ~2 yr)

`` — the tape is event-driven on HPC announcements:

• Aug 14, 2025 — Fluidstack 200+ MW, 10-yr leases announced → +19% intraday.
• Aug 18, 2025 — Fluidstack CB-5 expansion (+160 MW).
• Oct 27–28, 2025 — Abernathy 168 MW JV, ~$9.5B contracted revenue, Google backstop raised to ~$3.2B / ~14% equity.
• Oct 22, 2025 — $3.2B 7.75% Senior Secured Notes priced (the HPC build financing).
• Dec 18, 2025 — Abernathy JV $1.3B project financing priced.
• Mar 11, 2026 — Wall Street cuts WULF + CIFR targets (a bitcoin/sector wobble).
• Apr 16, 2026 — $1,035.7M equity raise at $19.00/sh; Bridge facility repaid.
• Keefe later more than doubled its target; Morgan Stanley to $66.50.

Pattern: the market reacts almost exclusively to (a) new contracted MW with credit enhancement and (b) financing events that de-risk the build. It does not react to bitcoin-mining metrics anymore. The single most powerful catalyst class is "another Fluidstack/Google MW tranche." The corollary risk: any crack in the Fluidstack/Google relationship would be the inverse catalyst.

---

Phase C — Judge people & books

Lens 9 · Management

CEO — Paul Prager (Co-founder, Chairman & CEO since Feb 2021) ``:

1. Track record: US Naval Academy grad, ex-anti-submarine-warfare officer, then 1980s Wall Street oil trader. Founded Beowulf Energy (1990) — a private power developer. Beowulf revived a coal plant in Hardin, MT in 2020 to power a Marathon Digital bitcoin mine — the template he repeated at Lake Mariner. So the power-development competence is real and demonstrated.
2. Tenure & skin in the game: founder-CEO, 5 yr in seat. Still a meaningful holder, but selling — disposed of 275,000 shares for ~$4.48M over Mar 24–25, 2026. Routine diversification at a 52-wk high, not alarming in size, but worth tracking as the stock runs.
3. Capital allocation: aggressive and debt-financed — $5.7B principal raised, a $1.0B equity raise at $19, $200M buyback authorization (paradoxically running a buyback while issuing converts and equity). The bet is "build contracted infrastructure with cheap-ish credit and let the equity re-rate." ROE/ROIC are negative and not yet meaningful given the build phase.
4. Red flags — the Beowulf related-party web (material): critics allege $100M+ has flowed from public TeraWulf to Prager-controlled Beowulf E&D since the 2021 RTO; $20.3M of management/service fees in 2023 (~38% of opex); land leased from Somerset Operating Co. (99.9% Prager-owned), with 8.5M shares (~$11.5M) issued in a 2022 lease amendment. Mitigant: TeraWulf acquired Beowulf E&D in May 2025 ($54.6M, 94 employees) and terminated the Services Agreement — internalizing the single largest related-party stream. The ground-lease relationship (now the A&R/Somerset-Brookings leases) and a board-observer right for the lessor persist. Net: governance improved materially in 2025, but the founder's private entities are still woven through the land and history.
5. Archetype: founder-operator / dealmaker, not a caretaker. Implication: high execution intensity and aggressive financing — good for a build-out, but the related-party history means governance must be watched, not assumed.

Lens 10 · Forensic Red Flags

Acting as a forensic analyst ``:

• Revenue recognition (HPC): ASC 842 straight-line over non-cancellable term, with power passthrough grossed up into revenue and "generally with a mark-up" on subcontracted fit-out costs. This inflates the revenue line relative to economic margin (power is a wash). Watch: $17.6M of the $21.0M HPC revenue is "rent," $1.1M power passthrough, $2.4M maintenance — the genuinely high-margin piece is the rent. Deferred-rent liability ($70.7M) means cash (Core42's $90M prepay) was collected ahead of straight-line recognition — a positive cash dynamic but a non-cash revenue accrual to monitor.
• Warrant accounting (the big one): the $1,061M warrant liability (Google warrants) is marked to market through the P&L every quarter. As the stock rises, the loss grows — $216.3M non-cash hit in Q1 alone. This makes GAAP net income nearly uninterpretable and will keep producing large, volatile, non-cash losses while the stock is volatile. Not fraud — but it means "net loss" is the wrong metric for this name.
• SBC flattering nothing — it's enormous: $101.4M SBC in Q1 (3x revenue); $368.0M of unrecognized RSU/PSU comp to be expensed over ~1.2 yr. 32.99M RSUs granted in Q1 at $12.37 WAGD fair value. This is massive real dilution dressed as a non-cash expense.
• Cash-flow vs. earnings divergence: operating cash flow is negative $(17.6)M while the business shows segment profit — the gap is working capital + the cash cost of running a holdco with $67M/qtr interest. The company is not self-funding; it lives on capital markets.
• Capitalized interest: $13.5M of interest capitalized to PP&E in Q1 (plus $6.7M of amortized issuance costs capitalized). Standard for a build, but it understates reported interest expense vs. true cash interest burden.
• Goodwill $55.5M (mostly allocated to HPC) — small, low risk.
• Effective interest rates >> coupons: the 2031 converts carry a 1.00% coupon but a 10.9% effective rate (the conversion-feature bifurcation); the 2032 converts are 0.00% coupon. The cash coupon understates the true cost of capital — the dilution is the cost.
• Going concern: NOT flagged. With $2.6B cash, no going-concern doubt is raised. But negative equity + negative operating cash flow + a $2.8B 2030 maturity wall means liquidity is a future question if capital markets close.

Regulatory findings (required):

• SEC Litigation Releases / AAERs: None. `` — 0 LR, 0 AAER naming TeraWulf in the 2021–2026 search window (SEC EDGAR EFTS).
• 10-K Item 3 (Legal Proceedings): "The Company is not a party to any material legal proceedings and is not aware of any pending or threatened claims.". Q1 2026 Note 12 repeats this verbatim.
• Non-SEC (FTC/DOJ/etc.): web search surfaced short-seller allegations re: Beowulf related-party dealings (2023–24) but no government enforcement action, consent decree, fine, or penalty. Allegations ≠ findings.
• Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (LR, AAER), 10-K Item 3, and web search as of 2026-06-21. The governance risk is real (related-party history), but it has not produced a legal/regulatory action.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection

WULF is mid-transition and GAAP-loss-making on non-cash drivers, so an EPS line is the wrong target — the value driver is contracted-MW energized × lease economics, financed by debt that converts to equity. I frame three paths on revenue and delivered MW and note why an EPS forecast would be fabricated.

Inputs (all labeled):

• Live today: 60 MW (Core42) generating ~$21M/qtr HPC + ~$13M/qtr mining.
• Contracted-not-live: ~462 MW more (Fluidstack 378 MW Lake Mariner + 168 MW Abernathy, less timing) → 522 MW total / >$12.8B lifetime contracted revenue.
• Industry reality check: only ~25% of leased AI capacity delivered sector-wide; "structural de-ratings" for missed milestones.

Base case ``: Fluidstack Lake Mariner tranches commence through 2H 2026 as guided; Abernathy delivers 2H 2026–2027. Energized critical-IT rises from 60 MW → ~200–250 MW by end-2026. Run-rate HPC revenue scales toward ~$300–500M annualized as tranches commence. Still GAAP-negative on D&A + interest + warrant marks through 2027.

Bull case ``: full 522 MW delivered on schedule, Lake Mariner expansion to 750 MW unlocks more Fluidstack/new tenants, Hawesville (480 MW) + Cayuga (320 MW) lease up. The $12.8B backlog becomes a ~$1B+ annualized project-EBITDA stream by 2028 → the equity re-rates to an infrastructure (not miner) multiple.

Bear case ``: Fluidstack hits funding/credit trouble (it's raising $1B at a valuation that swung from $7.5B to $18B in months — venture-fragile ); construction slips; a lease commencement is delayed or a tranche is right-sized via the term-contraction option. Google's backstop covers a payment default but a delay still starves cash against the 2030 wall. Equity de-rates hard.

Three-year EPS: n/a / not modelable without fabrication. Non-cash warrant marks (a function of the unknowable future stock price) and the SBC schedule make a GAAP EPS forecast meaningless; a "clean" number would be invented. The scoreable bet is delivery, not EPS.

No forecast.ts create in --watchlist mode (per skill). The natural Brier line for a future pass: "WULF energizes ≥200 critical-IT MW of HPC by 2026-12-31, p≈0.55."

Lens 12 · Bull vs Bear

Bull case. TeraWulf controls something genuinely scarce — utility-scale, low-cost, low-carbon power with live interconnection at a retired-coal brownfield, in an environment where AI compute is gated by power, not chips. It has converted that into 522 MW / $12.8B of contracted, credit-enhanced, 10–25-yr revenue with Google standing behind the anchor tenant and holding ~14% of the equity — a hyperscaler-grade endorsement. The segment economics are real (HPC seg. profit ~$10M on $21M rev in quarter one). If management delivers the build on anything close to schedule, WULF re-rates from a bitcoin-miner multiple to a contracted-infrastructure / data-center-REIT-like multiple — a step-change the ~$14B cap only partially reflects. The pivot is, in essence, "we own the power, the hyperscalers need the power, and Google is co-signing the lease."

Bear case (2–3 permanent-impairment risks).

1. Counterparty fragility — the thesis IS Fluidstack. The anchor is a venture-stage neocloud ($180M 2024 revenue, raising $1B at a valuation that tripled in months, $10B of GPU debt via Macquarie, leaning on a single $50B Anthropic contract). Google backstops payments, but a Fluidstack insolvency mid-build, a delayed commencement, or an Anthropic wobble cascades straight into WULF's cash flows and the 2030 maturity wall. The equity is a levered call on a startup's solvency.
2. Balance-sheet leverage into a build. Negative equity, ~$5.7B principal debt + $1.3B JV debt, $2.8B due in 2030, operating cash flow negative. The model only works if capital markets stay open and construction stays on budget. A rate/risk shock or a missed milestone ("structural de-rating") could force dilutive raises at the worst time — and the company already issues equity + converts continuously.
3. Commodity tail + dilution. The bitcoin segment is shrinking and weather-dependent; SBC is 3x revenue ($368M unrecognized) and converts strike from $8.48 to $19.94 — relentless dilution that caps per-share upside even if the enterprise succeeds.

Pre-mortem (18 months out, thesis broke): Fluidstack failed to raise its next round / lost Anthropic capacity → one or more Fluidstack tranches were delayed or contracted via the term option → WULF's energized-MW ramp stalled at ~100 MW → with $300M+/yr cash interest and negative operating cash flow, WULF tapped equity markets repeatedly at falling prices → Google's backstop covered a default but not the timing → the stock de-rated from an "infra" multiple back to a "distressed miner with a half-built campus" multiple. The coal-plant-to-AI story became a coal-plant-with-empty-buildings story.

Are multiples too high? At ~$14B for 60 MW delivered and $12.8B contracted-but-mostly-unbuilt, the market is paying a near-full infrastructure price for execution that is ~12% complete. The multiple isn't insane if you underwrite Fluidstack + Google + the build; it's very rich if you don't.

Contrarian view (what the market is refusing to see): the bull narrative has quietly substituted "Google" for "Fluidstack" in everyone's mental model. The market is pricing WULF as if Google is the tenant. Google is not the tenant — it is a backstop guarantor with a 14% equity option and a warrant pledge, whose interest is in its own optionality, not in rescuing WULF's timeline. The gap between "Google-backed" (true) and "Google-tenanted" (false) is where the risk hides.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case:

• What structurally breaks the money machine? A single counterparty (Fluidstack) is ~546 of 522+ contracted MW. Lose it — to insolvency, an Anthropic loss, or a financing freeze — and the backlog headline ($12.8B) is fiction. The Google backstop is a payment guarantee on commenced leases; it does nothing for un-commenced, delayed, or right-sized capacity, and Google can let the warrants lapse.
• Revenue concentration: "for the three months ended March 31, 2026, the Company's HPC lease revenue was generated from one customer". 100% single-customer on the segment that justifies the entire re-rating.
• Why the moat is weaker than bulls think: the moat is power+interconnection at one site. Every miner peer (IREN, CIFR, CORZ, Riot) is racing the same playbook on the same brownfield-power logic; IREN already has Microsoft (a real hyperscaler) and CIFR has AWS — TeraWulf got the venture neocloud. The "best power" is replicable; the quality of the tenant is where WULF is comparatively weaker.
• Most dangerous competitor bulls underestimate: IREN — same cap (~$13B+), genuinely ramping AI revenue (+94% QoQ), hyperscaler-anchored, zero net debt vs. WULF's $5.7B, zero-BTC-treasury cleanliness. On a relative-value basis a short-WULF/long-IREN pair expresses the whole thesis.
• Worst capital-allocation / governance: the Beowulf related-party history ($100M+ to a Prager entity pre-2025, Somerset land, board-observer rights) + a $200M buyback run concurrently with $1B equity issuance and continuous convert issuance — issuing and buying at the same time is value-agnostic financial activity. CEO selling into the high.
• Assumptions that must hold for today's price: (1) Fluidstack stays solvent and funded through 2027; (2) construction delivers 522 MW roughly on time and on budget; (3) capital markets fund the gap without crushing dilution; (4) the 2030 $2.8B wall refinances cleanly; (5) bitcoin doesn't collapse the legacy segment's cash. Five things, all must hold.
• If growth disappoints 20–30% (a few tranches slip a year): energized-MW stalls, cash interest swamps thin operating cash flow, forced dilutive raises, de-rate to "distressed builder." Plausibly −50% to −70% equity given the leverage and the negative equity starting point.
• Single permanent-impairment scenario: Fluidstack insolvency before leases commence + a tight financing window → WULF cannot service $300M+/yr interest from operations → restructuring of the Wulf Compute secured notes → equity substantially impaired. Plausibility: low-to-moderate but non-trivial, and it is the entire left tail.

Lens 14 · Management Questions (ordered by information value)

1. Fluidstack is ~100% of HPC revenue and a venture-stage company raising at a valuation that tripled in months. What is your direct visibility into its funding and its Anthropic capacity, and what specifically happens to each un-commenced tranche if Fluidstack misses a financing round?
2. The Google backstop covers payment defaults on commenced leases. Walk through, tranche by tranche, what protects you against delay or use of the term-contraction option before commencement — and what that does to your 2030 cash needs.
3. You have ~$2.8B of debt maturing in 2030 against negative operating cash flow today. What is the refinancing plan, and at what energized-MW run-rate does the company become operating-cash-flow positive?
4. Of the 522 contracted MW, how many critical-IT MW will be energized and revenue-generating by 12/31/2026, and what is the single biggest gating item (power, interconnection, construction, GPUs)?
5. SBC was $101M (3x revenue) with $368M unrecognized; converts strike from $8.48–$19.94. What is the fully-diluted share count at full conversion, and how do you think about per-share value against that dilution?
6. Why run a $200M buyback while simultaneously issuing $1B of equity and multiple convert tranches? Reconcile those for me.
7. Post-Beowulf-E&D acquisition, what related-party arrangements remain (Somerset/Cayuga ground leases, board-observer rights), and what is the annual dollar value flowing to entities affiliated with management?
8. What is the blended lease rate ($/MW/yr) across the Fluidstack and Core42 leases, ex-power-passthrough, and how does it compare to IREN's Microsoft and CIFR's AWS economics?
9. Lake Mariner can scale to 750 MW pending NYISO. What's the realistic timeline and probability of that approval, and is the incremental capacity pre-spoken-for?
10. How do you intend to diversify the tenant base beyond Fluidstack — are you in active talks with hyperscalers directly, and what would it take to land one?
11. The bitcoin segment posted a negative net cost-to-mine on winter demand-response. What is the normalized run-rate, and at what point do you fully exit mining?
12. Morgantown (210 MW power) needs FERC approval. What is the gating timeline, and how central is owned generation to the HPC strategy vs. grid power?
13. Capex was $523M in one quarter and total project cost across four sites is large. What is the all-in capex to deliver 522 MW, and what's the funding bridge to get there?
14. What covenants on the $3.2B 2030 Secured Notes and the JV notes could trip on a construction delay or a liquidity dip, and how much headroom do you have?
15. If you had to name the one event that would most damage this company in the next 18 months, what is it — and what have you done to hedge it?

---