Fermi Inc. (NASDAQ/LSE: FRMI) — A $4.7 Billion Option on a Tenant Who Has Never Signed

Published: 24 July 2026 · Coverage: Initiation Price: $7.40 (24 Jul 2026 close) · Shares outstanding: 637,574,239 (10-Q cover, 11 May 2026) · Market cap: ~$4.72bn Sector: Real Estate / Power Infrastructure — AI data-center development

Standing disclaimer. The analysis in Sections 1–15 below carries no investment recommendation and no price target. The single deliberate exception is the Claude's Take block immediately below, which is clearly labelled as the author’s own subjective opinion. This article is general information, not investment advice.

⚡ Claude’s Take

Claude’s own subjective opinion. General information only — not investment advice, and not a recommendation to buy or sell any security. The analytical body (Sections 1–15) below carries no position and no price target.

AVOID — and specifically, do not buy this dip. Conviction: HIGH.

Directional zone: I would not underwrite equity here at ~$4.7bn market cap / ~$5.2bn EV. A price at which the asset base alone justifies the entry is roughly $1.50–$2.50/share (~$1.0–1.6bn market cap, i.e. at or below the ~$1.43bn of gross construction-in-progress, net of ~$968mm of debt) — and even that requires a signed tenant first.

Tag: “They deleted the tenant from the pay plan two days before the IPO.”

This is not a fallen angel and it is not contrarian value. It is a falling knife with a real but heavily encumbered asset floor, and the single most important fact in the file is a compensation detail: on 28 September 2025 — two days before the IPO went effective — the vesting conditions on 26,303,400 of 27,450,000 “performance” units (96% of the pool) were stripped out and replaced with pure calendar vesting. The conditions removed were signing the first tenant lease and delivering 1 GW of power to the Matador substation. Management thereby ensured it would be paid in full on the passage of time whether or not the company ever did the two things the entire equity story depends on. Ten months later, the company has done neither, and the executives’ stock vests anyway. Every other governance fact in this report — the 40%-of-all-employee-equity “Top-Up Grant” to the founder, the $450.5mm of FY2025 NEO compensation in a zero-revenue year, the 1,500,000 RSUs granted to a new General Counsel vesting on the grant date, the board granting itself 1.02mm shares five weeks after firing the founder for cause — is a variation on the same theme. Founders and a sitting director have sold ~$62.6mm of stock. Not one insider has bought a single share on the open market since the IPO — not at $37, not at $4.47.

The business case fails independently of the governance case, which is why I am comfortable at high conviction. Fermi has zero revenue, zero contracted megawatts, zero megawatts energized, and no binding lease with anyone, ever. It IPO’d on a non-binding LOI that expired on 9 December 2025; the tenant — reported by Business Insider and corroborated by Bloomberg as Amazon — terminated the associated $150mm construction-advance agreement on 11 December with $0 drawn. Bloomberg’s reporting that Amazon concluded Matador would “reliably supply less electricity than Fermi claimed” is the most dangerous sentence in the file, because it implies the failure is about the asset, not merely the founder’s conduct. Meanwhile the market Fermi sells into is pre-leasing 74.3% of North American capacity before it is built at a record-low 1.4% vacancy. In that market, ten months of zero pre-leasing is not evidence of weak demand — it is evidence of counterparties selecting against this specific developer. The company’s own Chief Commercial Officer explained why: a 15–20-year lease requires a counterparty you can trust “over the long haul.” Fermi has had four CEOs-or-equivalents in ten months and still has no permanent CEO.

And it cannot afford to wait. Unrestricted cash fell from $408.5mm (31 Dec) to $207.5mm (31 Mar) to roughly $62.8mm (30 Jun) before the July convertible. The Q1 10-Q carries an explicit going-concern “substantial doubt” finding — six weeks after the 10-K said resources were sufficient — alleviated principally by an undrawn, stock-settled Yorkville note, i.e. a remedy that dilutes harder the lower the stock goes. Fermi’s secured money costs 12.0–12.9% (its own incremental borrowing rate is 16.1%; the bridge it retired carried a 48.9% effective rate) against 6.0–7.75% for Cipher, Applied Digital, TeraWulf and Core Scientific and ~3–4.5% for the investment-grade REITs. At a 12–13% marginal cost of secured capital, a developer earning a market 12–15% unlevered yield on cost earns nothing. The economics do not require a demand failure to break; they break at the financing cost. The $431mm convertible bought roughly two quarters, not solvency: management’s own plan calls for ~$2bn of capex in the next twelve months against ~$760mm of identified liquidity — a ~$1.24bn gap — and the full program at $70–90bn, of which all capital ever raised is 2.2–2.9%.

Three hard deadlines converge inside that runway: MUFG’s amortisation doubles on 10 Nov 2026 absent a ≥400 MW lease; Keystone demands prepayment at 105% absent a customer agreement by 31 Dec 2026; and the Texas Tech ground lease — the land itself — requires an executed Phase 1 tenant lease plus evidence of fully funded Phase 1 before year-end or the landlord gets an early termination right. The first annual meeting must also be held by 31 December, with an ousted founder controlling ~22.7% directly (a bloc the company puts at ~40.3%), two of three proxy advisers already on record supporting him, and an unadjudicated 70% entrenchment bylaw the company abandoned its own lawsuit rather than defend under oath.

What keeps this from being a zero, and why the bear case is not a short at this price: the assets are real and genuinely scarce. ~2.2 GW of gas turbines with the OEMs sold out through 2030 and pricing up 10–20 points, a 6 GW air permit (the second-largest of its kind), water, gas, fiber and ~4,523 contiguous acres. That is a saleable package, and management has said it may monetise turbines to fund itself. But note what that admission means: an asset you plan to sell to stay solvent is inventory, not a moat. A buyer of the equity here pays ~$5.2bn of enterprise value — roughly 3.6x the gross cost of the construction in progress — for a development option on which the reproduction cost is ~$1.43bn. I want the ~$3.5bn premium explained before I pay it, and the honest answer is that it is the residual of an October 2025 IPO priced on a 6% float and an LOI that no longer exists.

Conviction: HIGH. What flips me bullish: a single genuinely binding, filed lease with a named investment-grade hyperscaler for ≥400 MW at disclosed economics, together with closed project financing at a single-digit rate — that would satisfy the covenants, validate the asset against the Amazon report, and re-rate the equity violently. What flips me more bearish still: the 13 August Q2 call passing with the 90-day plan unmet and the Yorkville equity line drawn, or any indication the Texas Tech notice-to-proceed conditions will not be met by year-end. What I am watching: 12 August (the 90-day deadline), 13 August (Q2 call), 10 November (MUFG step-up), 31 December (Keystone, Texas Tech, and the annual meeting, all at once).

📈 Stock Price Action — Since-IPO Event Map

FRMI listed on 1 October 2025, so a five-year map is not available; this is the complete ~10-month public history. Price moves are FACT; attributed drivers are INTERPRETATION unless a same-day filing exists.

Fermi priced its IPO at $21.00 on 30 September 2025, opened at $25.13 and closed its first session at $32.53 (+54.9%) on a float of only ~6% of the capital. It touched an all-time intraday high of $36.99 the next day and has not been near it since. The all-time intraday low is $4.47 (9 April 2026). At the 24 July 2026 close of $7.40, the stock sits –80.0% from its intraday high, –64.8% below its IPO price, and +65.5% off the low, with a 52-week closing range of $4.80–$32.53. It trades below both its 21-day ($6.94) and 50-day ($7.04) EMAs on a rising-volume base, and far below its 200-day EMA ($11.80). Beta has risen from ~1.0 at listing to 2.69, with alpha of –1.19.

Cycle narrative. (1–2) The listing was engineered for scarcity: ~6% of the capital floated on a pre-revenue story, which mechanically produced both the +55% first day and the violence of everything after. (3) 12 December 2025 is the day the thesis broke — the anchor tenant walked, and the securities class action’s class period ends on precisely that date. (4) February’s rally shows the market rewarding financing headlines without pricing their cost; Keystone’s 12.90% coupon was the actual news. (5) The April low combined the lock-up expiry with reporting on the CEO’s conduct at Nvidia GTC and CERAWeek, which triggered Texas Tech to call the conduct a potential lease default. (6) The short report landed the same day as the CFO’s resignation. (7) The market paid +22.8% for a promise of a binding tenant — that promise expires 12 August 2026. (8) Tellingly, the best tape of 2026 came from the dissident’s campaign, not from company progress; the stock rose as the prospect of a sale rose. (9–10) Both reversed: the campaign was suspended, and the convertible added up to 58.9mm shares of dilution. (11) The most recent session is a 17% move with no identifiable cause — a structural feature of a name with a tiny original float, a ~40% insider bloc, no institutional 5% holder, and a fresh convert being hedged.

1. Executive Summary

Fermi Inc. is a 35-employee, pre-revenue Texas corporation that has spent $1.43bn assembling a development option on an AI data-center and on-site power campus (“Project Matador”) on ~4,523 acres of Texas Tech University System land in the Texas Panhandle. It listed on Nasdaq and the London Stock Exchange on 1 October 2025 at $21.00 per share, raising $784.9mm gross on approximately 6% of its capital. At $7.40 it is capitalised at ~$4.72bn, with ~$968mm of pro-forma debt for an enterprise value of roughly $5.2bn.

The company has never signed a binding lease with any tenant. It has zero revenue, zero contracted megawatts and zero megawatts energized; its property, plant and equipment “consisted entirely of land and construction in progress,” with nothing placed in service. The only counterparty ever identified was an unnamed “First Tenant” — reported as Amazon — bound by a non-binding LOI that expired on 9 December 2025 and an Advance in Aid of Construction Agreement the tenant terminated on 11 December with $0 drawn. The most recent disclosure describes “preliminary discussions with seven potential tenants and twelve potential joint-venture partners.”

The financial position is acute. Unrestricted cash fell from $408.5mm at 31 December 2025 to $207.5mm at 31 March 2026 to approximately $62.8mm at 30 June 2026, against Q1 capex of $441.2mm. The Q1-2026 10-Q disclosed substantial doubt about the ability to continue as a going concern — a conclusion absent from the 10-K filed six weeks earlier — alleviated principally by an undrawn, stock-settled Yorkville note. A material weakness in internal control over financial reporting was disclosed at 31 December 2025 and confirmed still existing at 31 March 2026; disclosure controls were deemed not effective. The July 2026 issue of $431.25mm of 5.00% convertible notes restored liquidity to roughly where it stood in March: management’s own plan requires ~$2bn of capex in the next twelve months against ~$760mm of identified liquidity, a ~$1.24bn gap, within a total program the company sizes at $70–90bn.

Capital allocation and governance are the weakest we have examined in this sector. Two days before the IPO, vesting conditions requiring a signed tenant lease and 1 GW of delivered power were removed from 96% of the executive performance-equity pool and replaced with calendar vesting. FY2025 named-executive compensation totalled $450.5mm in a zero-revenue partial year against $500,000 of aggregate salary. Share-based compensation has consumed 75–81% of all G&A. The founder-CEO held a “Top-Up Grant” entitling him to 40% of all equity granted to every other employee, semi-annually and before any change of control. Related-party arrangements include a Gulfstream dry lease from a founder affiliate costing $758,998 for two months of flight activity. Founders and a sitting director have sold ~$62.6mm; there have been no open-market purchases by any insider since the IPO.

Governance is also a commercial liability. The founder-CEO was removed on 17 April 2026 and terminated for cause on 30 April; the CFO resigned in April; a director resigned in July stating that no board minutes existed and that he learned of the $431mm convertible from the public announcement. The company adopted a 70% supermajority bylaw hours after losing a temporary restraining order, then abandoned its own suit rather than submit to court-ordered discovery. A securities class action is pending in the Southern District of New York. The company’s own risk factors concede this “may inhibit potential customers from transacting with us.”

We reach a negative verdict on all five substantive tests. The industry is structurally attractive but hostile to a subscale, late, capital-constrained entrant — 74.3% of North American capacity under construction is already pre-leased, so ten months of zero leasing reflects counterparty selection, not weak demand. There is no moat under any Greenwald category. Growth is entirely prospective. Financial quality cannot improve with scale because nothing is at scale. Capital allocation has been poor and, in the compensation design, adverse to outside shareholders. What Fermi owns is a genuinely scarce physical package — ~2.2 GW of turbines against sold-out OEM order books, a 6 GW air permit, land, water and gas — held by an issuer paying roughly twice its peers’ cost of secured capital, with three counterparty deadlines converging in November–December 2026 inside its cash runway. The valuation question is therefore not what the campus is worth when built; it is whether the equity is the right instrument through which to own an option that the company may have to sell assets to keep alive.

2. Business Overview

2.1 What Fermi is today

Fermi Inc. is a Texas corporation — not a Delaware entity, and with no Up-C structure, no operating-partnership units and no dual-class stock — formed on 10 January 2025. It operates publicly as “Fermi America.” Its registered offices are in Amarillo, Texas, with a second corporate headquarters opened in Dallas during Q2 2026. Fiscal year ends 31 December. The stock is registered under Section 12(b) and trades under the symbol FRMI on both Nasdaq and the London Stock Exchange — an unusual dual listing for a company of this age, which explains the presence of UK and European brokers (Ocean Wall, Rothschild & Co, Liberum, Berenberg) on its earnings calls and widens the retail base across two jurisdictions.

At 31 December 2025 the company had 35 full-time employees. Against that, it describes a build program it sizes at $70–90bn. This is the central disproportion of the business: a headcount appropriate to a development-stage promoter attached to an ambition appropriate to a national utility.

The REIT election — the premise of the IPO — has been deferred. Fermi registered its offering on Form S-11, the registration form for real-estate companies, and both the 10-K and the Q1 10-Q stated an intention to elect REIT status for the short taxable year ended 31 December 2025. The 8-K of 9 July 2026 disclosed that the company “has determined to defer its REIT election” and that it “was taxable as a C corporation … through its short taxable year ended December 31, 2025.” The structure sold to IPO buyers is not yet a fact, and the tax-efficiency rationale for the whole arrangement is in abeyance.

2.2 The land and the ground lease

The asset sits on land leased from the Texas Tech University System (TTUS) under a 99-year ground lease signed 14 May 2025 in Carson County, Texas. The headline acreage has moved and requires care:

• The original lease covered 5,769 acres.
• An August 2025 first amendment reduced the commenced area to 4,523 acres, with a further 713 acres contingent on a federal land transfer that had not commenced as of 31 December 2025.
• The frequently quoted “5,236 acres” is 4,523 commenced plus 713 pending; “7,570 acres” adds roughly 2,000 further acres acquired or contracted separately.

Economics: base rent of $1.2mm in year one with a 3% annual escalator, totalling $1,661.6mm undiscounted over 99 years. Variable rent begins only on subleasing — up to 1.0% of appraised shell value (0.5% above $3bn), plus 1.0% of gross power revenues and 25% of gross water revenues. That 25% water royalty is a material and under-discussed claim on a business whose cooling economics depend on Ogallala water.

The most revealing number in the lease accounting is the discount rate: the right-of-use asset is discounted at the company’s incremental borrowing rate of 16.1% — an auditor-reviewed, self-reported cost of secured debt, and a far more honest signal of Fermi’s cost of capital than anything in its investor materials.

The land can be lost. TTUS’s notice to proceed is conditioned on a package of six items: (i) evidence of full and unconditional Phase 1 funding; (ii) an executed lease with a Phase 1 tenant; (iii) insurance; (iv) energy services agreements; (v) a $5mm letter of credit; and (vi) all Phase 1 permits. The 10-K states these “must occur before the end of 2026,” failing which the landlord obtains an early termination right. On 30 March 2026 Fermi agreed to pre-pay $2.0mm of rent and escrow $9.0mm before year-end — the price of forbearance, not a milestone achieved. On the FY2025 call the then-CEO characterised the tenant condition as requiring “a tenant and an agreement with that tenant, and it has to be 200 megawatts,” with the CFO confirming “that’s by 12/31/26.”

Separately, the 10-K discloses that TTUS is itself a prospective sublessee of a powered shell — the landlord is also a customer, which raises the question of whether a friendly related-party arrangement could be used to satisfy the covenant rather than a genuine third-party hyperscaler lease. That remains an open question (Section 13).

2.3 How Fermi intends to make money

The model is landlord, not power merchant. Fermi intends to sign triple-net powered-shell and ground leases with rent expressed in dollars per kilowatt per month of reserved power capacity, plus a return on invested capital and pass-through costs as additional rent. Power is deliberately structured as “an incident of tenancy”: the 10-K states the model “does not anticipate nor rely upon any material marketing of or revenues from the sale of power to the grid.” This matters for two reasons — it is what makes REIT qualification conceivable, and it means Fermi is not an independent power producer with merchant upside; it is a landlord whose product happens to include electrons.

Revenue would be 100% contracted and recurring once leases exist, with 15–20-year terms. Today it is 0% of anything.

2.4 Contracted position: zero, and always has been

This is the single most important fact in the report and it deserves to be stated without hedging. Fermi has never had a binding contract with a tenant. The chronology:

Business Insider identified the First Tenant as Amazon on 17 December 2025; Fermi and the then-CEO denied it; Bloomberg corroborated the identification on 1 May 2026, reporting that talks broke down when Amazon sought to shorten the term from 20 to 15 years and concluded that Matador would reliably supply less electricity than Fermi claimed. If that reporting is accurate, the failure is an asset problem, not merely a personality problem — and the entire bull case rests on the opposite assumption.

2.5 Power: what actually exists

At 31 March 2026 the Q1 10-Q states PP&E “consisted entirely of land and construction in progress. No depreciable property, plant, and equipment had been placed in service.” Energized capacity is zero.

A definitional caution is required here, because the headline capacity numbers do not mean what they appear to mean. Management’s “more than 2 GW secured” is not 2 GW of ordered equipment: it aggregates the ~1,258 MW of firm-ordered and leased simple-cycle capacity above with combined-cycle uprates that have been neither ordered nor financed, plus the utility tariff contract. Similarly, the frequently cited “11 GW permitted” is materially overstated against the regulator’s record — TCEQ has issued 6 GW and the incremental 5 GW is a pending application. Fermi’s own 10-K states the position correctly: “currently permitted for up to approximately 6 GW.” The “17 GW” figure introduced on the Q1 call is aspiration, explicitly conditioned on land purchases not yet closed and permits not yet filed.

Two further gaps deserve naming. Solar and battery storage — both prominent in the IPO-era description of the “HyperGrid” — have nothing procured: no vendor, no megawatts, no contract, and the July 2026 investor deck’s generation page lists four gas platforms and zero solar or storage. And the gas supply is the one genuinely de-risked leg: firm supply from Energy Transfer, scalable to ~300,000 MMBtu/day with ~220,000 MMBtu/day secured, and 4.6 miles of on-site gas line installed.

The TM2500 disclosure is a materially negative item that has attracted little attention. These seven mobile units were the “rapidly deployable” leg of the story. As of 31 March 2026 the lease had not commenced because Fermi’s pick-up preconditions were unmet; a 13 April 2026 amendment moved the pick-up window to 1 July – 30 September 2027 and permits the lessor to rent the units to third parties in the interim, with automatic further extension. Fermi nonetheless owes monthly base rent through 2045 with no termination for convenience. The fastest-to-power asset slipped roughly fifteen months and may be rented out from under the company.

Permits are real: TCEQ granted final approval on 25 February 2026 for ~6 GW of gas generation, with an incremental 5 GW application filed 27 March 2026 and pending. The “11 GW” and “17 GW” figures cited in investor materials are planning parameters, not permitted or financed capacity — and the headline figure rose from 11 GW to 17 GW between the two earnings calls even as construction was paused and contracted capacity remained zero.

Nuclear is an application, not a project. A combined operating licence for four AP1000 units was filed 17 June 2025 and accepted 5 September 2025, with EIS scoping on 20 March 2026 — but only Parts 1 and 2 are filed; Part 3 (site-specific) remains outstanding, with site-characterisation data due October/December 2026. The 10-K concedes roughly 60 months of construction per unit after approvals and provides a fallback of 5 GW of additional gas “in the event that the anticipated expansion of domestic nuclear energy demand does not materialize.” Front-end engineering with Hyundai E&C and reactor-pressure-vessel forging dies at Doosan Enerbility are genuine pre-development steps; they are not licences.

2.6 Verdict

Fermi is not yet a business; it is a well-assembled, heavily encumbered development option with a landlord’s revenue model and no tenants. The physical package — contiguous permitted acreage, water, gas, fiber, and scarce turbines — is real and was assembled quickly. But every element of the revenue model remains prospective, the fastest-to-power assets have slipped, the structure sold at IPO (a REIT) has been deferred, and the one commercial relationship the company ever had ended with the counterparty walking away and drawing nothing.

3. Industry Dynamics

3.1 Demand is real, and that is not the issue

The AI infrastructure build is not in doubt. Calendar-2026 hyperscaler capital-expenditure guidance aggregates to roughly $630–750bn, up 36–77% year over year: Amazon ~$200bn, Google $175–185bn, Meta $115–135bn, Microsoft ≥$120bn, Oracle ~$50bn. CBRE reports North American primary-market vacancy at a record-low 1.4%, asking rates of $195.94/kW/month (+6.5%), with Northern Virginia at 0.3% and Dallas–Fort Worth at 1.8%. Microsoft has disclosed an Azure backlog of roughly $80bn that it cannot fill for want of power.

3.2 The decisive statistic

74.3% of North American capacity currently under construction is already pre-leased.

This single number governs the entire analysis. The market clears capacity before it is built, on the strength of the developer’s balance sheet, relationships and delivery record. A developer ten months past its IPO with zero pre-leasing is therefore not facing a weak market — it is being selected against by a strong one. Management’s repeated framing that “demand remains strong” is true and beside the point. The scarce commodity is not tenants; it is credible counterparties, and that is precisely what Fermi’s balance sheet and governance record deny it.

3.3 Where Fermi actually sits: SPP, not ERCOT

Almost all commentary on Fermi assumes ERCOT. It is wrong. The 10-K is explicit that Project Matador sits in Southwestern Public Service (Xcel) certificated territory inside the Southwest Power Pool (SPP), subject to FERC, NERC and the Midwest Reliability Organization. The consequences run in both directions:

• Favourable: Texas SB6’s ≥75 MW large-load regime — the $100k screening fee and the “kill switch” mandating ERCOT-directed curtailment for post-2025 loads — does not apply. Nor does ERCOT’s new Batch Zero interconnection process, whose first transmission plan is not due until autumn 2027.
• Unfavourable: SPP is FERC-jurisdictional where ERCOT largely is not. Fermi’s co-located configuration therefore sits inside the unresolved FERC co-location docket; the 10-K concedes FERC “is in the process of determining the policies it will apply.”
• Structural: the Panhandle has no approved multi-gigawatt transmission. The approved 765-kV programme is Permian-focused, ~$13.8–14bn, and not in service until roughly 2031; a Panhandle extension remains “a potential expansion concept, not an approved project.” The existing backbone is CREZ-era 345-kV built to export wind.

Behind-the-meter generation is therefore not a clever strategic choice — it is the only option. Everything rides on turbines.

For completeness: the widely cited 438 GW ERCOT interconnection queue is disowned by ERCOT itself (CEO Pablo Vegas: “we believe this forecast to be higher than expected future load growth”), and ERCOT filed a downward adjustment on 18 May 2026. It is not relevant to Matador in any event.

3.4 The real chokepoint: turbines

Of roughly 12–16 GW slated for 2026 across ~140 North American projects, only ~5 GW is actually under construction and ~7 GW has been cancelled or delayed; Sightline’s base case is that 30–50% of the 2026 pipeline never materialises. Announced US pipeline capacity of ~780 GW exceeds total US peak load of 759 GW.

Gas turbines are the binding constraint, and the OEMs are sold out:

This is Fermi’s one genuine asset — and it is inventory, not a moat. The company’s own going-concern mitigation says so, citing “significant equity in its power generation … equipment” and the option “to monetize all or any portion of these assets in markets where demand currently exceeds available supply.” An asset a company plans to sell to remain solvent is by definition not a barrier to entry. Worse, competitors are routing around the frame-turbine queue in 12–18 months using aeroderivatives and reciprocating engines: ProEnergy is supplying Crusoe with 13 × PE6000 (~650 MW); Caterpillar G3500s support 2 GW at Nscale’s Monarch; Meta’s Socrates mixes Titan 250s, SGT-400s and recips.

3.5 Nuclear is a decade away

Part 52 combined licences at Vogtle and V.C. Summer took roughly 46–48 months from docketing to issuance, excluding two to four years of pre-application work. As of mid-2026: NuScale is the only SMR with full NRC design approval and has no US construction under way; Kairos’s Hermes units are non-commercial test reactors; TerraPower received the first commercial advanced-reactor construction permit on 4 March 2026 targeting 2030; the BWRX-300 at Clinch River has a hearing on 13 August 2026 with no permit yet issued; Oklo has no COL; the AP300 is not licensed. Vogtle 3 & 4 ultimately cost roughly $36.8bn against a $14bn estimate — about 15 years, seven years late, ~$10,800/kW.

The 2025–26 reforms (the ADVANCE Act, Executive Order 14300, and the NEPA pilot programme Fermi has joined) have compressed NRC review timelines by quarters. Nothing has compressed construction, which is where Vogtle lost a decade. With Part 3 unfiled, roughly 46 months of review ahead and ~60 months of first-of-a-kind construction after that, a first Matador nuclear electron lands in the mid-2030s. The 2032 target has no precedent in the licensing record.

3.6 Capital-cycle read (Marathon)

Every late-boom marker is present simultaneously: AI data-center debt exceeded $200bn in 2025 with $250–300bn projected for 2026; ABS issuance of ~$27bn in 2025 against a projected ~$300bn take-out requirement; novel collateral invented to keep the flywheel turning (CoreWeave’s $8.5bn investment-grade GPU-collateralised facility); a wave of sector IPOs at elevated valuations — of which Fermi’s own pre-revenue $21.00 listing is a specimen; announced supply (780 GW) exceeding the entire installed base (759 GW); and load forecasts publicly disowned by the operator that produced them. The asset-growth anomaly finds an unusually pure specimen here: PP&E from $0 to $1,430.9mm in five quarters, funded by an IPO and four separate debt facilities.

The honest counter-argument is that the cycle has not turned. Vacancy of 1.4%, pre-leasing of 74.3% and Microsoft’s unfillable backlog all say demand for powered capacity remains unmet, and the lag phase of a capital cycle runs for years. The oversupply risk sits specifically in un-powered, un-anchored shells — which is exactly Fermi’s category until a tenant signs. The sharpest cautionary comparable is Poolside/CoreWeave’s 2 GW “Project Horizon” at Longfellow Ranch in West Texas: announced with a named anchor and mutually terminated within roughly twelve months.

3.7 Verdict

Structurally excellent for incumbents with signed tenants, cheap capital and a delivery record; structurally hostile to a subscale, late, capital-constrained entrant. Three facts settle it. First, 74.3% pre-leasing means tenants are not short of counterparties — they are short of credible ones. Second, the constraints that actually bind (turbines, transformers, switchgear, interconnection) are cleared by capital depth and relationships, which a 35-person developer with a going-concern flag has least of. Third, roughly half the 2026 pipeline is expected never to materialise, and in that phase the marginal entrant becomes someone else’s cheap asset. This is not a demand problem — which is precisely why Fermi’s failure to lease cannot be blamed on the market.

4. Competitive Position

4.1 There is no moat

Applying Greenwald’s taxonomy directly, no category survives contact with the facts.

No supply/cost advantage. Land in the Texas Panhandle is abundant and cheap — Galaxy Digital acquired its entire Helios site for roughly $65mm. Fermi’s ground lease is a 99-year obligation carrying a landlord termination right, discounted at a 16.1% borrowing rate. TCEQ air permits are ministerial and non-exclusive; management itself reported that the regulator’s feedback was effectively an invitation to apply for more, which is the opposite of a barrier to entry. The turbines are fully transferable — which is exactly why they appear in the company’s liquidity plan.

No customer captivity. Captivity requires customers, and there are none. Even prospectively, switching costs accrue only after a 15–20-year lease is signed. Prior published work in this series on this cohort reached the identical conclusion in each case: Core Scientific — “not a moat — it is a backlog”; TeraWulf — “a head start, not a durable advantage”; Galaxy Digital — “a moat that does not survive the next tenant negotiation is a backlog.”

No economies of scale. Fermi is the smallest operator in its peer set by every operating measure while carrying the largest stated ambition — the exact inversion of what scale economics require. Greenwald’s further caution applies with force: market growth is the enemy of scale advantage, and this market is growing 36–77% per year.

4.2 The comparison, with numbers

4.3 Cost of capital is the decisive gap

Fermi’s secured equipment debt prices at 12.00% (Beal/CSG, stepping to 14% on default) and 12.90% (Keystone), against 6.00–7.75% senior secured paper across the entire neocloud cohort and roughly 3.1–4.5% for the investment-grade REITs. Its best pricing — MUFG at SOFR+400, a ~12.1% effective rate — sits at or above the cohort’s worst. Its 5.00% convertible compares with IREN’s 0.00–3.50%. Its own incremental borrowing rate is 16.1%. The bridge it retired in February carried an effective rate of 48.9% and was extinguished at a $24.8mm loss, having cost roughly $44mm all-in over five months.

Now run the cross-read from prior published work in this series on this cohort. Equinix earns 26–27% cash-on-cash against ~3.1% debt — roughly 23 points of spread. Digital Realty earns 11.4% against 4–5% — about 7 points. The neoclouds earn ~12–15% yield-on-cost against 6.00–9.25% — 4–7 points. At a 12–13% marginal cost of secured capital, a developer earning a market 12–15% unlevered yield on cost earns approximately nothing. Fermi cannot build the campus described in its own 10-K at the cost of capital it is actually paying. The economics do not require a demand failure to break — they break at the financing cost. This is the most important analytical point in the report after the absence of a tenant.

4.4 How far ahead the Texas competition is

All of the following sit within 150–250 miles of Matador, all have a signed anchor tenant, single-digit project debt and delivered megawatts:

• Crusoe/Lancium, Stargate at Abilene — 1.2 GW, eight buildings, 4mm sq ft, $11.6bn financed May 2025; first two 200 MW buildings live since October 2025, roughly four of eight operational; anchored by OpenAI/Oracle.
• Vantage, “Frontier,” Shackelford County — 1.4 GW IT load, ten buildings, 1,200 acres, >$25bn; first building H2-2026; OpenAI-anchored.
• Galaxy Digital, Helios, Dickens County — 133 MW delivered 6 July 2026 with rent commenced; 526 MW / 15-year CoreWeave lease worth >$15bn.

4.5 The one honest point in Fermi’s favour

No other multi-gigawatt campus was identified in the Panhandle itself, and the single-site assembly is genuinely unusual: roughly 7,570 contiguous acres, a 6 GW air permit, Ogallala water rights, Energy Transfer gas, dual fiber routes, rail access and a Pantex-adjacent industrial workforce. If a hyperscaler specifically wants 1+ GW behind the meter in that geography, Fermi is close to the only door.

But scarcity of a site is not scarcity of an offer. The market has had ten months to walk through that door and has not. And the bull must explain why a tenant would accept the Panhandle’s transmission isolation and Fermi’s counterparty risk when Abilene, Shackelford and Dickens County are all closer to load, further along, and financed at half the cost.

4.6 Governance as a competitive disadvantage

This is not noise; it is a first-order commercial liability, and Fermi’s own Chief Commercial Officer articulated the mechanism: a 15–20-year agreement means “the counterparty is looking for assurance that who they’re partnering with is somebody that is going to be able to support them over the long haul.”

Against that standard: the founder-CEO was removed on 17 April 2026 and terminated for cause on 30 April; the CFO resigned on 19 April; a securities class action alleging misstatements about tenant demand and funding arrangements is pending; a full proxy war ran from May to July, waged by a founder controlling ~22.7% directly (a bloc the company puts at ~40.3%) who publicly campaigned to sell the company; reciprocal Texas Business Court litigation added Rick Perry as a defendant on 2 July; a 70% supermajority bylaw was adopted mid-contest and remains unadjudicated; and on 10 July a sitting director resigned citing the company’s failure to provide board and committee minutes, the use of special committees “as a means of avoiding full Board action,” and — of the $431mm convertible — “I wasn’t made aware of the transaction until the public announcement was made.” Permanent officers were appointed only on 20 July, and there is still no permanent CEO.

Fermi’s own risk factors concede the consequence: this “may inhibit potential customers … from transacting with us” and “may also negatively impact our ability to enter into definitive lease agreements.” As of 22 June 2026 the company itself stated that it and its adviser Broadhaven were engaged with seven tenants and twelve potential partners, and that “almost all of these parties have indicated they do not want to be involved with Fermi if Toby regains control.” Both sides of the dispute, in other words, agree that the governance fight is deterring customers; they disagree only about whom to blame.

4.7 Verdict

No durable competitive advantage exists. What Fermi owns is a well-assembled development option — permitted land, water, gas, fiber and ~2.2 GW of scarce turbine equipment — held by an issuer paying roughly twice its cohort’s cost of secured money, with a going-concern flag, 35 employees, a for-cause CEO termination, an active securities class action and a ~40% bloc that wants to sell. The bull claim that “land plus power plus permits in one place” constitutes a moat fails the simplest Greenwald test: a moat must be something competitors cannot replicate. Crusoe did it at Abilene. Vantage is doing it at Shackelford. Galaxy did it at Dickens County. Fermi’s differentiation is not capability — it is that it started later, sits in a worse grid jurisdiction, and pays far more for money.

5. Growth History and Forward Opportunities

5.1 There is no growth history

Fermi was incorporated on 10 January 2025 and has recognised no revenue in any period. This is not an interpretation: the SEC XBRL fact set for CIK 0002071778 contains no us-gaap:Revenues tag of any kind. There is no revenue by segment to decompose, no organic-versus-acquired split, no cohort or unit growth. The company reports as a single operating and reportable segment.

What has grown is the asset base and the share count:

PP&E went from zero to $1,430.9mm in five quarters. Under Marathon’s asset-growth framework this is the signature of a capital-cycle peak, not of compounding.

5.2 The forward opportunity, taken at its strongest

The bull case for growth is genuinely large if — and only if — the gating item clears. Management’s stated program is 11 GW of power (raised to a “path to 17 GW” on the Q1 call) and 18mm sq ft of data-center space, with all phases sized at $70–90bn. At market rents of roughly $195.94/kW/month, 1 GW of leased IT capacity would generate on the order of $2.3bn of annual rent before costs. Even a single 400–500 MW anchor lease at market terms would transform the income statement from nothing to roughly $1bn of annual contracted revenue and would unlock project-level financing at rates far below the 12–13% Fermi pays today.

That is the whole thesis, and it is not a foolish one. The demand exists, the site is permitted for 6 GW, and the turbines are in hand when the OEMs are sold out through 2030.

5.3 Why the gate has not opened

The gate has been shut for ten months for reasons that are documented rather than speculative:

1. The anchor walked. The First Tenant terminated in December 2025 having drawn $0, with Bloomberg reporting the counterparty concluded the site would supply less power than claimed.
2. Counterparty risk. A going-concern flag, a for-cause CEO termination, no permanent CEO, no board minutes and a live proxy fight are disqualifying for a 15–20-year commitment, as the company’s own risk factors and its CCO both concede.
3. Delivery risk. The fastest-to-power assets (the TM2500s) slipped ~15 months and may be rented to third parties. Zero megawatts are energized. Cleanview’s April 2026 satellite imagery showed minimal construction against the company’s claim that “the initial phase of construction is already complete,” concluding first buildings could not be live before roughly May 2027.
4. The tenants’ own bottleneck. Management itself identified customers’ mechanical/electrical/plumbing build capability — not Fermi’s power — as the emergent constraint, and hired a former Meta employee to diligence tenants’ execution pace. This is a striking reversal of the IPO narrative: if tenants cannot install MEP fast enough to take power, Fermi’s “speed to power” advantage is worth materially less than underwritten.

5.4 Forward milestones that would constitute real growth

5.5 Verdict

There is no growth to assess — only a binary option on a first lease, and the quality of that option is deteriorating rather than improving. The honest framing is that Fermi’s “growth” to date consists of converting $2.02bn of raised capital into $1.43bn of construction in progress and a 10.9% increase in shares outstanding, while the single commercial relationship it possessed terminated. Growth here is not high-quality or low-quality; it is entirely prospective and entirely contingent, and management’s own 90-day scorecard — set on 14 May and expiring 12 August — will be the first objective test of whether it is becoming real.

6. Financial Quality

6.1 The income statement is share-based compensation

Cumulative net loss over fifteen months: $675.1mm. Note that the accumulated deficit is only $321.9mm because pre-conversion losses were charged to members’ equity — the deficit understates the losses actually incurred and should not be read as the cumulative loss.

Share-based compensation is the income statement. FY2025: $132.7mm expensed — 74.7% of all G&A — plus $67.1mm capitalised, for $199.8mm. Q1-2026: $134.0mm expensed — 80.6% of G&A — plus $31.1mm capitalised, for $165.1mm. Cumulative SBC is $364.9mm; adding the stock donation and equity-settled charges, total equity-based charges reach $538.7mm in fifteen months. Unrecognised SBC at 31 March 2026 was $275.0mm over a 1.49-year weighted period. Against all of this, cash G&A runs only ~$32mm per quarter.

The FY2025 “other expense” of $312.3mm decomposes as: $173.8mm non-cash charitable donation (11,250,000 Class B units to the Dechomai Asset Trust); $61.0mm fair-value loss on the Series B convertible; $46.4mm fair-value loss on preferred embedded derivatives; $23.7mm conversion-inducement expense; $4.2mm Macquarie derivative; $3.3mm FX. Q1-2026’s $24.8mm is the Macquarie extinguishment loss.

6.2 Balance sheet

Equity was essentially flat quarter over quarter while liabilities more than doubled: Q1’s $441.2mm of capex was funded by drawing down cash and adding debt, not by earnings or fresh equity.

6.3 The cash flow, and where the burn actually is

A widely repeated error deserves correction: the burn is capex, not operating expense. FY2025’s $486.4mm GAAP net loss coexisted with only $34.2mm of operating cash use, because the loss was overwhelmingly non-cash. Q1-2026’s $188.7mm loss coexisted with $7.3mm of operating cash use — and even that $7.3mm was flattered by roughly $22mm of working-capital benefit from growth in accounts payable, implying an underlying operating burn nearer $29mm. Cumulative cash capex is $1,010.5mm; accrued-but-unpaid capex adds a further $158.1mm (FY25) and $108.4mm (Q1-26), so cash capex understates incurred capex.

The Q1 spend rate was roughly $147mm per month.

Q2 2026 (from the 9 July offering memorandum, not yet a reviewed 10-Q): cash fell to $92.0mm at 30 June, of which $29.2mm restricted — roughly $62.8mm unrestricted. Borrowings rose to MUFG $444.9mm, Keystone $77.3mm and Beal $14.7mm, totalling $536.9mm. Implied Q2 gross outflow was approximately $213mm, or ~$71mm per month — a deliberate ~52% deceleration reflecting the paused build.

6.4 Going concern — new, and not resolved

The Q1-2026 10-Q filed 15 May 2026 states that cash of $207.5mm plus $35.8mm restricted “are not sufficient to satisfy the Company’s financial obligations as they become due within one year … these conditions raise substantial doubt.” The 10-K filed six weeks earlier contained no such conclusion, stating instead that “we believe our resources are sufficient to satisfy our financial obligations for at least twelve months.”

That six-week deterioration is the cleanest expression of the company’s financial position: Fermi spent more in one quarter ($441.2mm) than it held in cash at the start of it ($408.5mm).

Management asserts the doubt is alleviated by (i) the undrawn Yorkville note, (ii) undrawn equipment facilities, (iii) potential monetisation of unencumbered turbines — in a resale market it has not price-tested — and (iv) deferring or renegotiating supplier and collateral obligations, while conceding that such approaches “could result in a termination of those arrangements.” Critically, the filing expressly excludes prospective tenant and strategic arrangements from the alleviation analysis: management is telling readers, in the filing, that it cannot rely on signing a tenant to solve liquidity within twelve months.

A disclosure-quality flag. On the Q1 earnings call held 14 May — one day before the 10-Q was filed — the Chairman told investors: “we have multiple levers we can pull, and we’re managing this company so that capital decisions are driven by strategy and not by pressure.” Neither he nor the interim CFO used the words “going concern” or “substantial doubt,” and no analyst asked. The statements are not literally irreconcilable — ASC 205-40 permits a conclusion that doubt is alleviated by management’s plans — but characterising the same facts as “not by pressure” one day before filing a substantial-doubt disclosure is a meaningful gap between oral presentation and written disclosure, and it goes directly to the weight management commentary can bear as evidence.

6.5 Material weakness

A material weakness in internal control over financial reporting was disclosed at 31 December 2025 and expressly confirmed still existing at 31 March 2026; disclosure controls and procedures were concluded not effective. The stated cause is newness — the company “ha[s] not yet fully developed or implemented” the necessary processes, systems and personnel.

We do not treat this as boilerplate. A 35-person company carrying $1.43bn of construction in progress, a $173.8mm non-cash charitable-contribution charge, embedded-derivative and convertible fair-value accounting, capitalised interest at rates that included a 48.9% bridge, and a deferred REIT election is precisely the setting where control deficiencies matter. Note also that as an emerging growth company Fermi is exempt from SOX 404(b) auditor attestation on internal control — no independent auditor opinion on ICFR is required.

6.6 Quality of earnings

• $128.0mm — 9.0% — of the $1,423.1mm of construction in progress is not steel or turbines: $98.2mm of capitalised share-based compensation and $29.8mm of capitalised interest, the latter at a weighted-average rate that included the 48.9% Macquarie loan. Nothing has been placed in service, so none of the CIP balance has yet been tested by depreciation or impairment.
• The REIT election was deferred; 2025 was taxed as a C corporation, contrary to the S-11 registration premise.
• Off-balance-sheet: $192.4mm of Siemens purchase obligations, $35.8mm of surety bonds, $5.3mm of letters of credit, and a net profits interest to MAD Energy of 2.5% of NOI on the first 1,000 MW (capped at $100mm NPV).
• The ground lease is discounted at a 16.1% incremental borrowing rate and grants the landlord 25% of gross water revenue from subtenants.

6.7 Which ratios are meaningful — and which are not

ROIC, ROE, ROA, gross/operating/EBITDA margin, EV/EBITDA, P/E and FCF yield are all undefined: there is no revenue, no asset in service and no depreciation. We note that ROIC.ai returns a TTM free-cash-flow-to-firm figure of +$928.7mm for Q1-2026; this is an artefact of adding capex back and was discarded.

The correct analogues:

6.8 Verdict

Economics cannot improve with scale because nothing is at scale; and the financial quality that does exist is poor. The company has no revenue, an income statement dominated by share-based compensation, a going-concern qualification that appeared within six weeks, an unremediated material weakness, roughly 9% of its principal asset composed of capitalised compensation and interest, and a cost of secured capital of 12–13% against peers at 6–7.75%. The one genuine financial strength is that the burn is discretionary capex rather than structural operating expense — the company can and did decelerate spending by ~52% — but that lever works by stopping construction, which is the same thing as not building the asset the equity is priced on.

7. Capital Allocation

7.1 The funding gap

The IPO raised $784.9mm gross (37,375,000 shares at $21.00 including the full greenshoe), $745.6mm net of a 5.0% underwriting spread (UBS, Evercore, Cantor Fitzgerald, Mizuho) and $731.5mm after offering costs. Cumulative external capital raised since inception is approximately $2.02bn.

Management’s own plan, stated identically in the 10-K and 10-Q: Phase 0 plus Phase 1 “could exceed $3 billion, of which approximately $2 billion is expected to be incurred in the next twelve months”; all phases together, “$70 billion to $90 billion.”

There is also a composition problem: roughly $225mm of that $760mm is equipment-secured and cannot be used to pay EPC contractors, site works or corporate overhead. The $431mm convertible restored the gap to approximately where it stood at 31 March — it bought two quarters, not solvency.

7.2 The debt ladder, in order of deterioration

A 0% coupon on a senior unsecured note is not philanthropy. Yorkville Advisors is a well-known provider of equity lines and share-settled notes; the economics are earned through share issuance at a discount, not through interest. The company has agreed to negotiate a committed equity line of credit with Yorkville and to register the resale of shares issuable under both the note and the line. Management’s remedy for going-concern doubt is therefore a mechanism that issues more shares the lower the stock trades — and its availability period expires 1 October 2026, roughly three months before the company’s hardest deadlines.

7.3 Three deadlines inside the runway

All three fall inside the ~6.7-month cash runway. A fourth item lands in the same window: Nasdaq rules require Fermi to hold its first-ever annual meeting by 31 December 2026.

7.4 Insider behaviour: zero purchases, $62.6mm of sales

There has not been a single open-market purchase (transaction code P) by any insider since the IPO — not at $21, not at $37, not at $4.47.

Three observations. First, the 3 June “cluster of seven” Form 4s are grants, not sales — Rick Perry, Haas, Robbin-Coker and McIntire received 175,094 each, Stein 250,000, and Kellerman and Everson 35,818 each — awarded five weeks after the founder-CEO was fired for cause. Second, Caddis filed a Form 144 on 30 March proposing 20,000,000 shares / $107.2mm, of which 11mm were sold within two days; the co-founder was selling into the lock-up expiry. Third, Caddis’s entire pre-IPO stake cost $126,382.19 in total — roughly $0.0022 per share, against $21.00 for IPO buyers, a ratio of about 9,500 to 1. Founders and a sitting director have realised roughly $62.6mm.

7.5 Compensation — and the finding that matters most

FY2025 named-executive compensation totalled $450.5mm in a zero-revenue partial year: Neugebauer $47.98mm; Everson, Ortiz Blanes and Hamilton $134.175mm each (6,300,000 units at $21.00 plus a $1.75mm discretionary “consulting” bonus each). Aggregate salary across all four was $500,000.

The central finding of this report on capital allocation: the 27,450,000 “performance-based” awards had their real conditions stripped two days before the IPO. The pool originally comprised 21,150,000 senior-management units vesting one-third on IPO, one-third on signing the first tenant lease, and one-third on delivering 1 GW of power to the Matador substation; 5,153,400 management Class B units vesting on $30bn and $50bn valuation thresholds; 900,000 nuclear-milestone units; and 249,380 December PSUs. On 28 September 2025 the conditions on 26,303,400 units — 96% of the pool — were removed and replaced with pure calendar vesting, re-struck at $21.00 (hence the $20.39 weighted-average grant-date fair value against $5.10 on service awards). Only 4% retain any operating condition.

The executives’ equity now vests on the passage of time whether or not a tenant is ever signed and whether or not a single megawatt is ever delivered — the two events on which the entire equity story depends, removed from the pay plan forty-eight hours before public investors bought at $21.00.

Other structural items compound it:

• The “Top-Up Grant.” The founder-CEO’s employment agreement entitled him to fully vested shares every 30 June and 31 December sized at 40% of all equity granted to every other employee, plus an additional grant immediately before any change of control — uncapped, with no performance condition, and preserved in stub form even on a for-cause termination.
• The 2025 LTIP evergreen: an initial 69,073,650-share reserve plus 10% of shares outstanding annually through 2035 (~63.8mm shares per year at current counts).
• Director “IPO Awards” of $2,000,000 ÷ $14.36 ≈ 139,276 RSUs each.
• July 2026: a permanent officer slate installed without a CEO, on five-year contracts, with ~4,475,000 sign-on RSUs including 1,500,000 to the General Counsel vesting on the grant date (~$11.1mm at $7.40, with no service condition).
• 12,127,558 further shares were admitted to the LSE on 23 July 2026 to settle advisers’ fees in stock and vesting RSUs — roughly 1.9% additional dilution, bringing shares outstanding to approximately 649.7mm.

The one genuinely well-aligned element deserves credit: the new Chief Commercial Officer is eligible for lease-contingent and sales-contingent equity. That is the right instrument pointed at the right problem — and it is also the company’s own confirmation that signing a tenant remains the unmet gate.

7.6 Related parties

• Aircraft: a dry lease of a Gulfstream GVI from TMNN Manager LLC, a Neugebauer affiliate, at $60,000/month plus $2,000–$5,000 per flight hour, with Fermi also paying all direct operating and fixed costs — $758,998 paid or accrued for January–February 2026 flight activity alone, against roughly $9,600 returned under the reciprocal time-share, a ratio of about 79:1. His employment agreement separately obliged the company to reimburse costs of using an aircraft he owns.
• TFC Utilities Management LP, whose CEO Larry Kellerman is Fermi’s Head of Power and now a director: $1,332,177 in 2025 including a $750,000 IPO bonus. Kellerman holds 11,700,000 shares via TFC Utilities Energy LLC.
• Family: the then-CEO’s sons Nathan and Noah received 750,000 and 3,000,000 RSUs — $68.5mm of grant-date value.
• Firebird/MAD Energy: six SGT-800 turbines acquired from MAD Energy LP for >$165mm via a Delaware divisive merger, financed with a $145mm 11% PIK convertible — now the subject of a fraudulent-transfer suit (340 Energy, trial set 24 May 2027). MAD Energy also holds the 2.5% net profits interest noted in Section 6.6. Note that George Wentz, appointed General Counsel on 20 July 2026 with 1.5mm immediately vesting RSUs, is the founder and CEO of MAD Energy.

7.7 Verdict

Capital allocation has been poor on the investment side and adverse to outside shareholders on the compensation side. On investment: $2.02bn raised and $1.43bn deployed into a single un-tenanted site, financed through a descending ladder of increasingly expensive paper — a 48.9% bridge repaid at a $24.8mm loss, 12–14% equipment facilities carrying tenant-contingent acceleration triggers and a $37mm exit fee, a 0%-coupon note that amortises in stock, and a convertible struck 29% above a stock down 80% from its high. On compensation: $450.5mm of NEO pay in a zero-revenue year against $500,000 of salary; a 40%-of-everyone-else’s-equity ratchet for the founder; a 10% annual evergreen; a board that granted itself 1.02mm shares five weeks after firing that founder for cause; and, decisively, the deletion of the tenant-lease and 1 GW conditions from 96% of the performance pool two days before the IPO. Insiders have sold $62.6mm and bought nothing. Under this playbook’s standard — that capital allocation is the bridge between business value and shareholder value — this is a failing grade, and it is the clearest single reason to doubt that value created at Matador would accrue to public shareholders.

8. Changes and Headwinds — Since the IPO

Fermi has been public for less than ten months, so the standard two-year lookback is the full corporate history. The period contains an unusual density of adverse developments.

8.1 Event timeline

8.2 The proxy contest

The dissident is not an outside activist but Toby Neugebauer, the co-founder and former CEO, acting through Vicksburg Investments Management LLC, personally, and the Melissa A. Neugebauer 2020 Trust — 146,516,035 shares, or 22.7% of the shares outstanding. The company characterises the wider aligned bloc, including family affiliates and his two board designees, at ~40.3%. (For the avoidance of doubt, the Schedule 13D filed 15 May 2026 is not the activist: it is Caddis Holdings LP / Griffin Perry, a separate co-founder and the son of director Rick Perry, who filed in support of the incumbent board.)

His allegations: that the board refuses a credible, banker-led dual-track strategic review; publicly foreclosed a sale on 21 April “without any Board deliberation”; withheld nomination materials; had “a faceless executive office” cancel a validly called meeting; threatened to confiscate his shares via the REIT charter’s excess-share provision; sued in a forum barred by its own charter’s forum-selection clause and named his nominees “to intimidate them”; and entrenched itself with the 70% bylaw — a threshold he argues is unattainable because roughly 20% is held by insiders opposed to him, requiring ~75% of everyone else.

The company’s case: that the stock fell more than 80% under his leadership because he failed to sign a tenant; that he was terminated for cause on grounds including material misrepresentations to the board, public communications violating fiduciary duties and NDAs, unauthorised third-party meetings on company transactions, and “a repeated pattern of threatening, abusive and bullying behavior”; that his claim never to have sold a share is misleading because Vicksburg sold 4,082,858 shares concurrently with the IPO at $10.50 — half the $21.00 IPO price — on conversion of $42.87mm of notes; that his founder stock cost ~$0.0067 per share against $21.00 for IPO buyers, making a quick sale asymmetric; and — the sharpest operating point — that as of 22 June it and its adviser Broadhaven were engaged with seven tenants and twelve potential partners, and that “almost all of these parties have indicated they do not want to be involved with Fermi if Toby regains control.”

Outcome: he failed a procedural threshold; he was not defeated by shareholders. No special meeting was ever held. His claim of “more than 70% of votes cast” and the company’s report of ~31.0% consents against ~36.4% revocations are both true — the first is a share of votes cast, the second of shares outstanding, and the binding test is the charter’s requirement of 50% of outstanding to call a meeting. He suspended the campaign on 3 July, one day after the judge recused, and expressly reserved the right to resume or to run a slate at the annual meeting — which must be held by 31 December 2026.

Proxy advisers split: Glass Lewis and Egan-Jones recommended consenting to call the meeting (both 29 June); ISS recommended against (~2 July).

8.3 The Everson resignation

On 10 July, director Miles Everson — the former CFO — resigned, stating that the company had failed to prepare or provide any board or committee minutes despite written requests dated 18 April and 5 May, and that “continued use of special committees to approve significant transactions is a means of avoiding full Board action — [y]esterday’s announcement of a new $350 million convertible bond is just such an example. This matter was not brought to the Board … I wasn’t made aware of the transaction until the public announcement.” His 19 July response letter adds that the approving committee excluded the Neugebauer-designated directors and demands the company “correct or supplement its public disclosure.”

Caveat, stated plainly: Everson was Neugebauer’s designee and is not a disinterested witness. But the company itself filed his rebuttal, and the allegation that no minutes exist is a factual claim about record-keeping that sits alongside an unremediated material weakness in internal control.

8.4 Short seller and litigation

One short report, not several. Fuzzy Panda Research, 20 April 2026, disclosed short, published the same day as the CEO/CFO-departure 8-K; it estimated roughly $0.33/share of cash value and put the probability of completing Phases 0–1 below 0.1%. Its verifiable claims reconcile to primary filings ($408mm of year-end cash; 12.90% turbine debt; the 30 March lock-up expiry). Its fraud and personal-conduct allegations — an “$8.7mm skim,” IP misappropriation, backdated securities, a scheme linked to the Siemens deal, an undisclosed family relationship, and testimony about substance use — are uncorroborated by any court finding or regulator and are recorded here only as short-seller allegations. We place no weight on them. Separately, Cleanview (April 2026, not a short seller) commissioned satellite imagery showing minimal construction against Fermi’s claim that “the initial phase of construction is already complete,” concluding first buildings could not be live before roughly May 2027.

Lupia v. Fermi Inc., No. 1:26-cv-00050 (S.D.N.Y.), filed 5 January 2026, names the company, directors and officers and the IPO underwriters under Securities Act §Section 11/15 and Exchange Act §Section 10(b)/20(a). The class comprises IPO purchasers and buyers from 1 October through 11 December 2025 — the class period ends on the day the tenant walked. The claim is, in essence, that the IPO was sold on a tenant relationship whose fragility was not disclosed. The company cannot estimate a loss, and its own risk factors warn that Neugebauer “has filed, and may in the future file, additional legal proceedings.”

No public evidence exists of an SEC investigation, a Wells notice, a Nasdaq deficiency notice or an auditor resignation. We record that as an absence of evidence, not a clean bill of health.

8.5 Verdict

These changes decisively weaken the thesis. In ten months the company lost its only tenant, its CEO (for cause), its CFO, a director (over record-keeping), and the REIT structure on which it registered; acquired a going-concern qualification, a material weakness, a securities class action, a short report and a proxy war; deferred its fastest-to-power assets by fifteen months; and financed itself at 12–49%. The countervailing positives are real but narrow: a 6 GW air permit granted, a 5 GW application filed, turbines physically arriving, a professionalised officer slate, and a lease-contingent incentive for the new CCO. Not one of them changes contracted megawatts, which remain zero.

9. Risk Analysis

9.1 Risk matrix

9.2 The risks that actually matter

Risks 1, 2 and 3 are one risk expressed three ways, and they share a single resolution date. Fermi must sign a binding tenant lease of sufficient scale to (a) satisfy Texas Tech’s notice-to-proceed conditions by 31 December, (b) avoid Keystone’s 105% prepayment on the same date, © avoid MUFG’s amortisation doubling on 10 November, and (d) unlock the project financing that is the only realistic source of the ~$1.24bn twelve-month gap. Every one of those falls inside the company’s own cash runway of roughly 6.7 months. There is no scenario in which the equity performs well and the tenant does not arrive.

Risk 8 is the one the market appears least focused on and which we regard as most under-priced. If Bloomberg’s reporting is accurate — that a sophisticated counterparty with full data-room access walked because the site would supply less power than claimed — then the governance narrative is a distraction and the asset itself is impaired. Every bull case, including the “new management fixes it” case, assumes the opposite.

Risk 18 warrants explicit statement because the structure invites it. The equipment facilities are nonrecourse and secured on the turbines; management has said in filings that it could be forced to “surrender assets pledged as collateral.” The $431mm convertible is unsecured and structurally subordinated to all of it. In a downside resolution, secured lenders take the turbines, the landlord can terminate the ground lease, and the equity — the residual claim on an unbuilt campus — is where the loss lands first and hardest.

9.3 Verdict

This is a high-likelihood, high-severity risk profile in which the principal risks are correlated and share a common resolution window in November–December 2026. That correlation is the defining feature: the same missing event — a signed tenant — simultaneously triggers the lender, landlord, liquidity and governance risks. Diversification of risk does not exist here; it is one binary, wearing several costumes.

10. Valuation — Embedded Expectations

10.1 Why conventional multiples do not exist

Fermi has recognised no revenue since inception; TTM EBITDA is –$343.9mm; no asset has been placed in service, so there is no depreciation, no in-place NOI and no cap rate. P/E, EV/EBITDA, EV/Sales, P/B-on-earnings and FCF yield are all undefined. Two data notes matter for anyone reproducing this work: the price-history feed’s valuation_index returns {"latest": null, "history": null} — there is no own-history percentile read, because there is no history — and ROIC.ai reports a TTM free-cash-flow-to-firm figure of +$928.7mm for a company that burned cash all year, an artefact of adding capex back. Neither should be quoted.

The absence of the own-history percentile is itself the datum. On most reports it is the single highest-signal valuation input; here the company is ten months old and there is nothing to rank against.

Capital structure — the denominator most sources get wrong. One class of common; no dual class, no super-voting, no OP units despite the S-11 registration. Shares outstanding 637,574,239 (10-Q cover, 11 May 2026), plus 12,127,558 admitted to the LSE on 23 July for advisers’ fees and RSU vesting. At $7.40 the market capitalisation is ~$4.72bn, not the $4.03bn carried by the price-history feed, ROIC and FactorsToday — those figures are struck on the prior close of $6.32 and a stale share count. Adding ~$968mm of pro-forma debt and netting cash gives a working enterprise value of ~$5.1bn.

10.2 Management’s own unit economics — the fair anchor

The IPO prospectus provides the cleanest available statement of the intended economics, and we use it rather than importing our own:

“for a normalized 1 GW, approximately (i) $600 million in top line base rent revenue ($50 per KW per month) and (ii) $900 million in power capacity rent revenue ($75 per KW per month) … we expect to incur $500 million in operating expenses attributable to such normalized GW which would result in $1.0 billion of NOI for such normalized GW.”

So $1.5bn of revenue and $1.0bn of NOI per gross gigawatt, a 66.7% margin. Two observations follow. First, only $50/kW-month is real-estate rent; $75 is power rent, of which roughly $42 is fuel passed through — Fermi is approximately 40% landlord and 60% merchant power seller, which is why the margin is 67% rather than the 80–90% the converted-miner cohort books, and which complicates the REIT logic. Second, against management’s own capex of $70–90bn ÷ 11 GW = $6.4–8.2bn per GW, $1.0bn of NOI implies a ~14% stabilised yield on cost — genuinely attractive, and better than Digital Realty’s 11.4% on a 61%-pre-leased pipeline.

The asset concept is good. The problem is entirely counterparty and financing. We want to be explicit about that, because it is the strongest version of the bull case and it survives scrutiny.

10.3 Method A — invert the cohort’s price per contracted megawatt

At $7.40 the market prices Fermi as though it had already signed roughly 200–425 MW of critical IT load on terms comparable to peers holding real leases with CoreWeave, AWS and a Google-backstopped Fluidstack. Note the coincidence that closes the circle: the MUFG facility’s amortisation test is set at 400 MW — the market is capitalising approximately the lease Fermi’s own lender demanded and Fermi has not signed.

10.4 Method B — reverse-DCF on Phase 1, every assumption stated

Equity at 2030 = $10.0bn – $4.55bn + $3.00bn – $0.90bn = $7.55bn, discounted to $3.64bn, across ~847mm diluted shares = $4.30 per share. Sensitivity: a 12x exit multiple and a 15% discount rate produce $6.45.

The headline conclusion: an investor can accept 100% of management’s own unit economics — $1.5bn of revenue, $1.0bn of NOI per gigawatt, a ~14% yield on cost, twenty-year triple-net leases — and still not reach today’s $7.40 in the base case. Reaching it requires simultaneously an investment-grade exit multiple and an investment-grade cost of equity, for an asset that has neither.

10.5 Price per megawatt, on every honest basis

Bases (a) and (b) are the finding, and no amount of framing softens them: the metric that matters most is undefined because the denominator is zero.

10.6 Against the comparable universe

Two sentences carry this table.

First, at ~$464 per announced kilowatt Fermi sits almost exactly between NuScale (~$420–500) and Oklo (~$536) — the two most prominent pre-revenue “priced on a promise” stories in this series’ coverage, both of which this series rated AVOID, both with large announced pipelines and zero binding contracts. Fermi is not priced as a power company or as a REIT. It is priced precisely as a pre-revenue nuclear developer.

Second, Solaris — which has actually contracted more than 2 GW with three investment-grade technology counterparties — commands ~$3,500 per contracted kilowatt, roughly 7.5x Fermi’s price per announced kilowatt. That 7.5x gap is not an opportunity waiting to close. It is the market’s price of a signature.

A consistency problem in management’s own capex plan. The stated build is roughly 6 GW of nuclear across two nuclear islands plus up to ~6 GW of gas. At Vogtle’s realised ~$16,000/kW, 6 GW of AP1000 alone is ~$96bn — more than the top of management’s $70–90bn estimate for the entire 11 GW campus including 15mm sq ft of data centers. Even at an optimistic nth-of-a-kind $8,000/kW the nuclear leg is ~$48bn. The $70–90bn figure is internally consistent only if the campus is built predominantly with gas — which is precisely why the incremental 5 GW gas permit filed on 27 March matters: it is the option to abandon nuclear, and it should be read that way.

10.7 Scenarios

Bear mechanics: no lease by mid-2027 triggers the MUFG amortisation step-up and Keystone’s 105% mandatory prepayment; Yorkville expires 1 October 2026. PP&E of ~$1.7–2.0bn is recovered at 85–100% into a genuine turbine shortage, less ~$0.9–1.3bn of debt, less ~$0.2bn of wind-down and litigation costs, plus ~$0.6bn for the ground lease, the 6 GW air permit and the interconnection position.

10.8 What the price implies

Solving for the probability weights that reconcile a bear of ~$1.50, a base of ~$5.35 and a bull of $16.70 to the $7.40 close gives approximately 30% bear / 45% base / 25–30% bull.

At $7.40 the market underwrites roughly a one-in-four chance of the full 2 GW hyperscale outcome and only ~30% odds of the no-tenant bear — for a company with zero contracted megawatts, an ASC 205-40 substantial-doubt disclosure, an interim office of the CEO and (until 22 July) an interim CFO, a founder-led proxy war, a securities class action, and roughly $2bn of capex due within twelve months against ~$780mm of liquidity.

The bear weight is too light. That, rather than any view on the campus’s ultimate worth, is our valuation conclusion.

10.9 Dilution — the term that dominates

Because project debt is unavailable until a lease exists, the marginal funding source is equity or equity-linked, and it prices off a depressed stock. On a $1.5bn raise against 638mm shares:

The convertible’s 45.3mm base shares (58.9mm maximum) sit on top, as do up to 40mm Yorkville shares and a 10%-of-shares-outstanding annual LTIP evergreen. The equity is short a large, price-contingent option to itself: the worse the news, the lower the price at which the required equity must be sold.

Asset floor. Pro-forma book equity of ~$0.9–1.1bn is roughly $1.40–1.70 per share; a break-up crediting the scarce long-lead equipment plus the ground lease and permits is roughly $1.90–2.80 per share. At $7.40 the stock trades at ~4.4–5.3x tangible book with no revenue — approximately $1.5bn of substance and $3.6bn of option premium.

10.10 The most informative third-party price signal

In July 2026, sophisticated institutional investors were offered this company. They bought $431.25mm of five-year paper at a 5.00% cash coupon with a $9.52 conversion price, and the company paid $34.5mm to cap its own dilution at $14.64. Given a free choice between owning the equity outright and owning a coupon plus an out-of-the-money call on it, the marginal informed buyer chose the latter. That is a more honest read on the risk-reward than any price target.

For completeness on sell-side coverage: the October 2025 initiation wave was almost entirely underwriter research (UBS $30, Evercore $37, Berenberg $37, Rothschild $31, Mizuho $27, Stifel $29, Cantor $27, Macquarie $35). The subsequent reset was severe — UBS cut $30 → $8 on 1 April 2026 while maintaining Buy, then downgraded to Neutral at $6 on 5 May; Evercore went $37 → $20 → $11. Headline “consensus” of ~$19–22 is an artefact of stale marks: the live cluster from firms that refreshed after April is $6 / $8 / $11 / $17, a median near $9.50. We found no analyst action at all on the going-concern disclosure or the July convertible. Where methodology is disclosed it is telling — Evercore applied ~13x NTM EV/EBITDA, an IPP peer multiple, to 2028 estimated EBITDA. These are probability weights on an unsigned lease presented as price targets.

10.11 Verdict

The market is pricing roughly 200–425 MW of contracted capacity that does not exist, at a valuation that requires investment-grade exit assumptions for a sub-investment-grade situation. Accepting management’s own unit economics in full produces a base case of ~$4.30, roughly 42% below the current price; reaching $7.40 requires both a 12x exit multiple and a 15% cost of equity. The asset concept is sound and the ~14% yield on cost is real if built and leased — but the equity does not capture it, because the path from here to there runs through a funding gap that must be bridged with price-contingent dilution. The one number that settles the section: $3,500 per contracted kilowatt for Solaris against $464 per announced kilowatt for Fermi. That 7.5x is what a signature costs, and Fermi does not have one.

11. Variant Perception

11.1 Consensus

The consensus view is that Project Matador is a scarce, strategically irreplaceable asset — one of the largest gas air permits ever granted, ~4,523 contiguous acres on a 99-year ground lease, four bidirectional gas pipelines, low-latency fiber to Chicago, Dallas and Phoenix, and turbine slots in a market sold out to 2030 — and that in an AI power crunch a hyperscaler must eventually take it. The debate is framed as when, not if. That framing is what supports a ~$5.1bn enterprise value on zero contracts.

11.2 The strongest bull case

Management’s economics are coherent and attractive: $1.0bn of NOI per gross gigawatt on a twenty-year triple-net structure with fuel passed through, a ~14% yield on cost, better than Digital Realty earns on a 61%-pre-leased pipeline. The turbines are in hand or financed in a market with five-to-eight-year lead times that no new entrant can replicate. One 1–2 GW signature converts an infinite EV per contracted megawatt into ~$2.5–5.1mm/MW — a deep discount to the cohort’s $12–25mm/MW — and unlocks 65–70% non-recourse project debt at a stroke. The re-rating would be violent, and the tape has twice shown what that looks like (+22.8% on the Fermi 2.0 plan, +22.6% on mere analyst speculation about OpenAI). Separately, a sale is genuinely live: the founder bloc is ~40%, he has publicly demanded a strategic review, and two of three proxy advisers backed his call for a meeting.

11.3 The strongest bear case

The company has never signed a customer. It missed the one checkable forecast in its own IPO materials — “1.1 GW of power projected to be online by the end of 2026”; the actual figure is zero, with the seven GE TM2500s now deferred to a July–September 2027 pick-up window. It lost its only prospective anchor within ten weeks of listing. It fired its founder-CEO for cause, lost its CFO twice, and ran an interim office of the CEO alongside an interim CFO. It carries a substantial-doubt disclosure whose alleviation plan includes selling the turbines it borrowed $537mm to buy. Q1-2026 G&A of $166.2mm against zero revenue was $165.1mm stock compensation.

And the decisive point: this is not a market in which nobody is signing. Solaris has contracted more than 2 GW with three investment-grade technology counterparties; VoltaGrid signed Oracle for 2.3 GW; Talen signed 1,920 MW to AWS; NRG signed 445 MW; Constellation and Vistra have multi-gigawatt PPAs. Fermi: 0 MW. It is a market where everybody is signing except Fermi.

11.4 The assumptions that actually matter

1. A binding lease of ≥400 MW within roughly four quarters. Everything else is downstream, and this is the lenders’ threshold, not merely our own.
2. That project debt at 65–70% of cost becomes available at 7–9% once a lease exists — that this is a chicken-and-egg problem rather than a permanent cost-of-capital problem. The 12.00% Beal coupon, the 12.90% Keystone coupon, the 16.1% self-reported incremental borrowing rate and the Yorkville structure all argue for the pessimistic reading.
3. That realised economics land near $50/kW-month for the shell and $75/kW-month for power. The power leg implies roughly $100/MWh — above every disclosed data-center PPA in our coverage (NRG’s $90–95/MWh for new build is the highest; Talen/AWS implies ~$62/MWh). The cautionary precedent is UAMPS walking away from NuScale when levelised cost moved from $58 to $89/MWh.
4. That the campus can be built for $6.4–8.2mm/MW — which, per the valuation comparables above, requires effectively abandoning the nuclear program.
5. That behind-the-meter co-located power sales survive regulatory scrutiny. FERC rejected Talen’s original AWS behind-the-meter interconnection agreement in November 2024 and the stock fell ~28%; in December 2025 FERC found PJM’s co-located-load rules “no longer just and reasonable.” Fermi’s SPP location is a genuine mitigant relative to PJM, but SPP is FERC-jurisdictional and the company’s own 10-K concedes FERC “is in the process of determining the policies it will apply.” This risk is under-weighted in Fermi’s disclosure relative to its importance.

11.5 Positioning and float

Short interest rose from 12.9mm shares in February to 42.5mm in mid-July, up roughly 941% year over year and spiking with the convertible — a pattern consistent with convertible-arbitrage delta hedging rather than pure directional bearishness (interpretation). Days to cover of 1.62 is thin: this is not a crowded short.

Squeeze mechanics (interpretation). With ~11.7% of a 45% float short, insiders holding 54.8%, institutions holding only 12.0%, realised volatility above 110% and a single binary catalyst, any credible tenant headline produces violent upside gaps — as the two +22% sessions demonstrated. The convexity is real. But it is a positioning fact, not a valuation argument, and the same thin float and negligible institutional base mean there is little natural support on disappointment. Note also that institutional ownership of 12% ten months after a $785mm IPO is itself an indictment: the institutions that bought the deal have largely left.

11.6 What would falsify each side

Falsifiers of the bull case: no binding ≥400 MW lease by roughly mid-2027; a Keystone covenant trigger or a lender beginning to market the equipment; an equity raise below ~$5; a second prospective tenant walking; a competitor signing the counterparty Fermi has been negotiating with; or — the Oklo test — the “11 GW” and “15mm sq ft” figures quietly disappearing from the filings, exactly as Oklo’s “14 GW pipeline” vanished from disclosure precisely as it stopped converting.

Falsifiers of the bear case: a signed, binding, disclosed lease of ≥1 GW with a named investment-grade or hyperscaler counterparty accompanied by a material prepayment; a non-recourse project financing closing at 7–9%; or a change of control at a premium.

11.7 Where we differ from consensus

Consensus treats the tenant as a timing question. We treat it as an open question of fact, for one specific reason: Bloomberg reported that the counterparty which walked concluded the site would “reliably supply less electricity than Fermi claimed.” If that is right, the market is mispricing not the timing of a signature but the existence of the product. Every bull case — including the sympathetic “good asset, bad governance, new management fixes it” case — assumes the asset performs as advertised. Nothing in the public record yet confirms that it does, because zero megawatts have ever been energized.

Our second departure is narrower and more mechanical: consensus discusses the funding gap as a financing detail. It is not a detail. It is the dominant term in the equity value, because it must be closed with price-contingent dilution while the stock is 80% below its high.

12. Fact vs. Interpretation

13. Open Questions

1. Does the site actually deliver the power claimed? The single most important unresolved question. Bloomberg reported that the departing counterparty concluded it would not. Zero megawatts have ever been energized, so there is no operating evidence either way. Everything else is secondary to this.
2. Who was the First Tenant, and why exactly did it leave? Never confirmed in any filing; the company denied the Amazon identification. The distinction between “walked over term length,” “walked over counterparty risk” and “walked over deliverable capacity” is the difference between a governance problem and an asset problem.
3. Will Texas Tech extend the notice-to-proceed conditions? UBS’s May downgrade referenced “land lease milestone extensions,” and the Chancellor has publicly discussed extending milestones — but no extension has been disclosed in a filing. The company pre-paid $2.0mm of rent and escrowed $9.0mm, which reads as consideration for forbearance.
4. Was the ground lease competitively bid? No evidence of an RFP has surfaced. TTU policy OP 01.02 requires two state-certified appraisals; none has been made public. The Amarillo Tribune obtained the lease only by public-records request, and TTUS referred remaining documents to the Texas Attorney General. Rick Perry appointed multiple TTU regents while governor. Strongly implied but undocumented — we draw no conclusion.
5. What are the Yorkville note’s actual VWAP discount mechanics? They sit in Exhibit 10.18 and are not summarised in the filings. This determines the true dilution cost of the going-concern remedy.
6. Did the July convertible cure the going-concern qualification? The next disclosure is the Q2 10-Q, expected around 13 August 2026.
7. Is the TTUS powered-shell sublease capable of satisfying the ≥200 MW covenant? If the landlord can also be the covenant-satisfying tenant, the December deadline is far softer than it appears — and correspondingly less meaningful as validation.
8. What caused the 24 July +17.1% move on 38.2mm shares? No filing, no news. Unexplained.
9. Acreage: the same 424B4 states 5,236 acres in one section and 5,769 in the audited footnote. Unreconciled.
10. Lockup length: the prospectus summary and the 10-K say 180 days; the binding underwriting section says 120 days. No waiver was disclosed.
11. Does the co-location structure survive FERC review? The 10-K concedes FERC has not settled its policy.
12. What is the status of the TCEQ permit challenge raised in Fuzzy Panda’s 7 May follow-up regarding background monitoring locations? No company rebuttal and no responsive 8-K.
13. Will the annual meeting produce a renewed proxy contest? It must be held by 31 December 2026; the founder expressly reserved the right to run a slate.

14. What Must Be True

14.1 For the bull case

14.2 For the bear case

The symmetry is the point. Both cases turn on the same event, with the same deadline. This is not a business to be analysed on trend; it is a dated binary, and the resolution window is 12 August to 31 December 2026.

15. Source Appendix

See Appendix B below for the full list of primary filings, transcripts, data sources and secondary references relied upon, with URLs and access dates.

Sections 1–15 contain no investment recommendation and no price target; the Claude's Take block is separately labelled as the author’s own opinion. This article is general information and not investment advice. The author holds no position in any security mentioned.

APPENDIX A — Standard Diligence Questionnaire

Fermi Inc. (NASDAQ/LSE: FRMI) · 24 July 2026

A standard diligence questionnaire applied to the company. Labels: FACT / INTERPRETATION / ASSUMPTION / OPEN QUESTION.

General

What thoughtful questions have other investors asked about this company?

The sell-side questions on the two earnings calls were narrow and largely deferential, but four were genuinely probing and all four remain unanswered:

• Vikram Malhotra (Mizuho) asked why management would put a “shot clock” on a binding tenant agreement given the prior year’s record — the right question, since the 90-day commitment made on 14 May expires 12 August 2026. FACT that the commitment was made; OPEN QUESTION whether it is met.
• Richard Anderson (Cantor Fitzgerald) asked what must be in place to preserve the ground lease, eliciting the single most important disclosure in the file: a ≥200 MW tenant agreement and a notice to proceed by 31 December 2026. FACT.
• An unnamed shareholder (Radway Capital) asked whether the founder’s ~40% holding could block a capital raise, and — separately and acutely — whether power revenue exceeding 25% of rental income would break REIT qualification. Management confirmed it was structuring revenue recognition to preserve REIT status; three months later the REIT election was deferred and 2025 was taxed as a C corporation. FACT.
• Stephen Gengaro (Stifel) asked what the departing first tenant did for power instead — the question that leads directly to whether the failure was about the counterparty or about the asset. It was not answered.

Bear-side investors have concentrated on three things: whether any tenant will ever sign; whether the going-concern qualification is curable without severe dilution; and whether the TCEQ air permit’s modelling is sound (Fuzzy Panda’s 7 May follow-up alleged background monitors sited in San Antonio and Lubbock rather than Amarillo/Hobbs). The third has received no company rebuttal and no responsive 8-K — OPEN QUESTION.

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low? Neither — there are no earnings. FACT: zero revenue since inception (10 January 2025); the SEC XBRL fact set contains no us-gaap:Revenues tag of any kind. Cumulative net loss of $675.1mm over fifteen months.

Driven by the external environment or internal actions? INTERPRETATION: overwhelmingly internal. The external environment is as favourable as it has ever been — 1.4% vacancy, $195.94/kW-month asking rents, hyperscaler capex up 36–77%. In that environment 74.3% of North American capacity under construction is already pre-leased while Fermi has leased nothing. The constraint is company-specific: counterparty credibility, governance and cost of capital.

How stable are revenues? Not applicable today. Prospectively, the model is among the most stable in existence if it works — triple-net leases of 20 years plus four 5-year renewals, rent expressed per kW of reserved capacity, costs passed through. ASSUMPTION: stability accrues only after signature; before signature the business has no revenue stability, it has binary risk.

Outlook for products/services? FACT: the product is powered shells plus on-site electricity. Demand for that product is demonstrably strong and unmet (Microsoft’s ~$80bn unfillable Azure backlog). OPEN QUESTION whether Fermi’s specific instance of the product performs — the Bloomberg report that the departing tenant concluded the site would “reliably supply less electricity than Fermi claimed” is unresolved and is the most important open item in this report.

How big will this market be — growing, shrinking, domestic or international? FACT: growing sharply and predominantly domestic. Calendar-2026 hyperscaler capex of roughly $630–750bn. INTERPRETATION: the market is large enough that Fermi’s failure cannot be attributed to market size; roughly 30–50% of the announced 2026 pipeline is nonetheless expected never to materialise, and announced US pipeline capacity (~780 GW) exceeds total US peak load (759 GW).

Business Quality & Competitive Moat

Is the industry getting more or less competitive? More. Capital is flooding in — AI data-center debt exceeded $200bn in 2025 with $250–300bn projected for 2026 — and new entrants are proliferating. Under Marathon’s capital-cycle framework this is a late-boom supply signature. INTERPRETATION: for the marginal, unfunded entrant, rising competition for tenants coincides with rising competition for turbines and capital, which is the worst combination.

How profitable is the business (ROIC, ROE)? Not meaningful, and we say so rather than computing a spurious number. ROIC, ROE and ROA are undefined: no revenue, no asset in service, no depreciation. The correct sector analogues are: EV per contracted MW (undefined — the denominator is zero), EV per announced MW (~$464k), gross PP&E per announced MW (~$130k, so the market carries construction in progress at ~3.6x cost), EV/book equity (~4.9x), and cash runway (~6.7 months at the observed $71mm/month). Prospectively, management’s own economics imply a ~14% stabilised yield on cost — attractive, and better than Digital Realty’s 11.4%.

How profitable is the industry — how many competitors, what barriers to entry? Highly profitable for incumbents: Equinix earns 26–27% cash-on-cash against ~3.1% debt; Digital Realty 11.4% against 4–5%. Barriers are real but capital-and-relationship barriers, not asset barriers — which is precisely why they exclude Fermi rather than protect it. FACT: Fermi’s secured debt costs 12.0–12.9% against peers at 6.0–7.75%. INTERPRETATION: at a 12–13% marginal cost of secured capital, a developer earning a market 12–15% unlevered yield on cost earns nothing. The economics break at the financing cost without any demand failure.

Can the business be easily understood? Yes — unusually so, and that is a genuine positive. Build power and shells on cheap land; lease both to hyperscalers on long triple-net terms. The complexity is not in the model but in the financing, the governance and the permitting.

Can it be undermined by foreign low-cost labour? No. The asset is physically fixed, and its value derives from US grid geography, US permits and proximity to US load. Exposure to foreign inputs is on the supply side: Siemens turbines from Germany, Doosan forgings and Hyundai engineering from Korea. FACT: management timed a customs clearance around a Supreme Court tariff ruling to save ~$27–30mm, and has applied for foreign-trade-zone subzone designation — so tariff policy is a live cost variable.

Do brands matter? Not as consumer brands — but counterparty reputation is the functional equivalent and it matters enormously. FACT: the company’s own Chief Commercial Officer identified trust over a 15–20-year horizon as the gating issue, and the company stated that “almost all” of the parties it is negotiating with “do not want to be involved with Fermi if Toby regains control.” INTERPRETATION: Fermi’s institutional reputation is currently a liability rather than an asset, which in this market functions as negative brand equity.

What is the nature of competition? Competition for a small number of very large, very sophisticated counterparties, decided on speed to power, deliverability, price and — decisively here — balance-sheet credibility. FACT: Crusoe/Lancium at Abilene has buildings live since October 2025; Vantage at Shackelford is building 1.4 GW with >$25bn; Galaxy delivered 133 MW at Dickens County on 6 July 2026. All within 150–250 miles, all with signed anchors and single-digit project debt.

Customers’ switching costs? Very high once a lease is signed — a 20-year triple-net commitment with co-located generation is close to irreversible. INTERPRETATION: this cuts against Fermi today. High switching costs make counterparties extraordinarily careful about whom they sign with, which raises the credibility bar that Fermi is currently failing. Switching costs are a moat only for the incumbent who already has the customer.

Financial Condition & Balance Sheet

Assets not fully recognised on the balance sheet? FACT: yes, and they are among the more valuable things here. The 6 GW TCEQ air permit (the second-largest of its kind) carries no balance-sheet value. The 99-year ground lease appears only as a $39.7mm right-of-use asset. Water rights, the NRC combined-licence application, and option value on the remaining ~10 GW of announced capacity are similarly uncapitalised. INTERPRETATION: a break-up analysis crediting these gets to roughly $1.90–2.80 per share — well below the current price, but well above nothing.

Off-balance-sheet liabilities? FACT: $192.4mm of Siemens purchase obligations; $35.8mm of surety bonds; $5.3mm of letters of credit; a net profits interest to MAD Energy of 2.5% of NOI on the first 1,000 MW (capped at $100mm NPV); and the ground lease’s variable-rent obligations — 1.0% of gross power revenue and 25% of gross water revenue from subtenants, plus scholarship commitments up to $200mm, ≥$1mm/year to the TTUS Excellence Fund, a reclamation sinking fund of $10mm/year escalating 3%, and $100,000/day delay rent. The seven TM2500 turbines carry monthly base rent through 2045 with no termination for convenience, despite the units not having been picked up.

How conservative is the accounting? Not conservative, and control quality is weak. FACT: a material weakness in internal control over financial reporting was disclosed at 31 December 2025 and confirmed still existing at 31 March 2026; disclosure controls were concluded not effective. As an emerging growth company Fermi is exempt from SOX 404(b) auditor attestation, so no independent opinion on internal control exists. FACT: $128.0mm — 9.0% — of the $1,423.1mm of construction in progress is capitalised share-based compensation ($98.2mm) and capitalised interest ($29.8mm), the latter at a weighted-average rate that included the 48.9% Macquarie bridge. Nothing has been placed in service, so no part of the balance has been tested by depreciation or impairment. INTERPRETATION: capitalising compensation and punitive-rate interest into an asset that has never been tested is the aggressive end of acceptable practice. Auditor: Ernst & Young LLP (Fort Worth), FY2025 fees $3.09mm. FACT: no Item 4.02 non-reliance filing, no late-filing notification and no auditor change appears in the 146-filing EDGAR history — and the 10-K/A of 30 April 2026 was a Part III filing, not a restatement.

How CapEx-hungry is the business? Extraordinarily — this is the defining financial characteristic. FACT: $1,010.5mm of cash capex in five quarters; Q1-2026 alone was $441.2mm, a rate of ~$147mm/month. Management’s own plan: Phase 0+1 “could exceed $3 billion, of which approximately $2 billion … in the next twelve months,” and all phases $70–90bn. Against ~$760–780mm of identified liquidity, the twelve-month funding gap is roughly $1.24bn. All capital ever raised (~$2.02bn) equals 2.2–2.9% of the full program.

Capital Allocation & Management

How much FCF does the business generate, how does management use it, what is the philosophy? FACT: none — free cash flow is deeply negative and always has been. A necessary correction to a widely repeated error: the burn is capex, not operating expense. FY2025’s $486.4mm net loss coexisted with only $34.2mm of operating cash use because ~92% of the loss was non-cash. Q1-2026’s $188.7mm loss coexisted with $7.3mm of operating cash use — itself flattered by ~$22mm of payables growth, implying an underlying ~$29mm. Combined operating-plus-investing burn was $604.4mm in FY2025 and $448.5mm in Q1-2026. The stated philosophy since May 2026 is to sequence capital deployment behind tenant agreements and project financing; INTERPRETATION: this is a going-concern mitigation measure — capex sequencing is expressly cited as a pillar of the ASC 205-40 alleviation conclusion — presented as strategic discipline.

Significant acquisitions recently? FACT: the material transaction is the acquisition of six Siemens SGT-800 turbines from MAD Energy LP for more than $165mm via a Delaware divisive merger, financed with a $145mm 11% PIK convertible. It is the subject of a fraudulent-transfer suit (340 Energy v. Firebird LNG, Business Court of Texas No. 26-BC11B-0016, ≥$5.985mm, trial set 24 May 2027) — corroborated in Fermi’s own 10-Q, not merely alleged by a short seller. Note: George Wentz, appointed General Counsel on 20 July 2026 with 1,500,000 immediately vesting RSUs, is the founder and CEO of MAD Energy.

Buying back shares? No. The company is a net issuer and structurally must be.

Issuing large amounts of new shares to insiders? FACT: yes, at a scale that is the central capital-allocation finding of this report. Cumulative share-based compensation of $364.9mm in fifteen months, consuming 74.7% (FY2025) and 80.6% (Q1-2026) of all G&A, against cash personnel costs of ~$12mm for ~35 employees. A $173.8mm charitable contribution of Class B units to the Dechomai Asset Trust before the IPO. FY2025 named-executive compensation of $450.5mm against $500,000 of aggregate salary — Everson, Ortiz Blanes and Hamilton each received $134.175mm. Unrecognised SBC of $275.0mm remained at 31 March 2026. The 2025 LTIP carries an initial 69,073,650-share reserve plus a 10%-of-shares-outstanding annual evergreen through 2035. Shares outstanding rose 11.4% in ten months with zero revenue, before 12,127,558 further shares admitted on 23 July to settle advisers’ fees in stock.

Compensation policy of directors/management? FACT — and this is the finding that most damns the structure. The 27,450,000 “performance-based” awards originally vested on: one-third at IPO, one-third on signing the first tenant lease, and one-third on delivering 1 GW of power to the Matador substation (plus tranches tied to $30bn and $50bn valuation thresholds). On 28 September 2025 — two days before the IPO went effective — the conditions on 26,303,400 units (96% of the pool) were removed and replaced with pure calendar vesting, re-struck at $21.00. Only 4% retain any operating condition. Management is therefore paid in full on the passage of time whether or not a tenant is ever signed or a megawatt ever delivered.

Compounding items: the founder’s “Top-Up Grant” of 40% of all equity granted to every other employee, semi-annually and again immediately before any change of control, uncapped and unconditioned; director “IPO Awards” of ~139,276 RSUs each; the board granting itself 1,022,012 shares on 1 June 2026, five weeks after firing the founder for cause; and July 2026 five-year employment agreements including 1,500,000 RSUs to the General Counsel vesting on the grant date (~$11.1mm, no service condition). The one genuinely well-designed element: the new Chief Commercial Officer’s lease-contingent and sales-contingent equity — the right instrument aimed at the right problem, and the company’s own admission that signing a tenant remains the unmet gate.

Motivations of management? INTERPRETATION, and we state it directly. The compensation architecture rewards having done the deal — forming the company, securing the land and completing the IPO — rather than operating it successfully. Founder economics of $0.0067 per unit against a $21.00 IPO price (a ~3,134x ratio) mean insiders are profitable at almost any price, while IPO buyers are down 65%. Founders and a sitting director have realised ~$62.6mm; no insider has made a single open-market purchase since the IPO — not at $21, not at $37, not at $4.47. The absence of any purchase at the all-time low is the most eloquent datum in the file.

Valuation & Market Data

Is the stock an ADR, MLP, or K-1 issuer? FACT: none of these. Fermi Inc. is a US (Texas) corporation filing 10-Ks and 10-Qs, issuing Form 1099 rather than K-1. It is dual-listed on Nasdaq and the London Stock Exchange under FRMI (LEI 529900TSHBYBCFMYZ228) — unusual for a company of this age. It registered on Form S-11 as a REIT, but the REIT election has been deferred; 2025 was taxed as a C corporation. Single share class, one vote per share; no dual-class and no OP units. Control operates instead through a Director Nomination Agreement and a REIT-charter 2.5% ownership cap.

Dividend policy? FACT: no dividend, and none expected. Management stated it does not anticipate material REIT taxable income near term given expected non-cash depreciation, and would not pay dividends until taxable income required it. With the REIT election deferred, the distribution requirement does not currently apply.

How profitable is the business? Covered above: not profitable, and no meaningful profitability ratio exists. FACT: TTM EBITDA –$343.9mm.

Is net income diverging from cash from operations? FACT: yes, enormously — and in the direction that flatters cash. FY2025: net loss $486.4mm against operating cash use of $34.2mm — a $452mm divergence, ~92% of the loss being non-cash ($173.8mm charitable contribution, $132.7mm SBC, $111.6mm derivative fair-value marks, $23.7mm inducement). Q1-2026: $188.7mm loss against $7.3mm operating cash use. INTERPRETATION: the divergence is genuine rather than manipulative — the charges really are non-cash — but it must not be read as a sign of health. The cash is leaving through the investing line at $441mm a quarter, and $31.1mm of SBC plus $11.5mm of interest are capitalised into PP&E rather than expensed, which flatters both statements simultaneously.

Risks & Downside

What factors would cause the stock to decline? In descending order of probability-weighted impact: (1) the 90-day plan expiring on 12 August with no binding tenant, confirmed at the 13 August Q2 call; (2) any equity or equity-linked raise at a depressed price, including the first Yorkville draw settling in stock; (3) MUFG’s amortisation doubling on 10 November absent a ≥400 MW lease, with the agent free to begin marketing the turbines; (4) Keystone’s mandatory prepayment at 105% on 31 December; (5) Texas Tech declining to issue or extend the notice to proceed; (6) a renewed proxy contest at the annual meeting, which must occur by 31 December; (7) adverse developments in Lupia; (8) any confirmation that the site delivers less power than claimed.

Risk of a catastrophic loss? Yes, and the structure channels it directly to the equity. FACT: the Q1 10-Q discloses substantial doubt about going concern; the equipment facilities are non-recourse and secured on the turbines, and management has stated in filings that it could be forced to “surrender assets pledged as collateral”; the $431.25mm convertible is unsecured and structurally subordinated to all subsidiary debt. INTERPRETATION: in a downside resolution, secured lenders take the turbines, the landlord may terminate the ground lease, convertible holders rank ahead of shareholders, and common equity — the residual claim on an unbuilt campus — absorbs the loss first and hardest.

Chance of a total loss? INTERPRETATION: low but not negligible — we would put it below the ~30% bear weight the market implies, because the asset floor is real. The turbines are genuinely scarce (OEMs sold out through 2030, pricing up 10–20 points), the 6 GW air permit has value, and the ground lease and water rights are saleable. A break-up plausibly recovers roughly $1.90–2.80 per share. The realistic bear outcome is therefore severe permanent impairment — on the order of 60–85% from $7.40 — rather than a zero. A true zero requires both a failed asset sale into a tight turbine market and litigation losses exhausting the residual, which is possible but not the central case. Note the offsetting consideration: a ~40% bloc actively seeking a sale makes a change-of-control outcome at some price more likely than in a typical distressed situation.

Recent News & Events

Has the business environment changed recently? FACT: the external environment has improved; the company-specific environment has deteriorated sharply. Externally: record-low 1.4% vacancy, rents up 6.5%, turbine OEMs sold out through 2030 with pricing up 10–20 points — all of which raises the replacement value of what Fermi owns. Internally, in ten months: the only prospective tenant terminated; the founder-CEO was removed and terminated for cause; the CFO resigned; a director resigned alleging no board minutes existed; a going-concern qualification appeared; a material weakness persisted; the REIT election was deferred; a securities class action was filed; a short report was published; a proxy war was fought and suspended; and the fastest-to-power assets slipped ~15 months.

Significant acquisitions? Covered above — the MAD Energy turbine transaction and its associated fraudulent-transfer litigation.

Change in accounting policies? FACT: no change in accounting policy, and no restatement. The material change is fiscal, not accounting: the deferral of the REIT election means 2025 was taxed as a C corporation, contrary to the S-11 registration premise. The going-concern conclusion changed between the 10-K (30 March: resources “sufficient”) and the 10-Q (15 May: “substantial doubt”) — a change in conclusion under ASC 205-40, driven by the facts, not a change in policy.

Recent changes — new markets, facilities, management? FACT: a second corporate headquarters opened in Dallas alongside the Amarillo site. Management was rebuilt entirely: Marius Haas became Chairman; the board expanded from five to seven (and has since lost Everson); an “Interim Office of the CEO” co-chaired by Anna Bofa and Jacobo Ortiz has run the company since April; Rob Masson became CFO (interim from 29 April, permanent 22 July); George Wentz became General Counsel on 22 July. There is still no permanent CEO, with a Heidrick & Struggles search under way — ten months after the IPO and three months after the founder’s removal. Physically: the TCEQ 6 GW air permit was granted (25 February), a 5 GW application was filed (27 March), and three Siemens SGT6-5000F turbines arrived at the Port of Houston on 21 July. Not one of these changes has altered contracted megawatts, which remain zero.

APPENDIX B — Source Appendix

Fermi Inc. (NASDAQ/LSE: FRMI) · CIK 0002071778 · Report date 24 July 2026

All sources accessed 24–25 July 2026 unless otherwise stated. Primary sources are listed first. The full trailing SEC filing corpus (120 documents) was reviewed.

B.1 Primary — SEC filings (company, CIK 0002071778)

Ownership and insider filings. Twelve Forms 3, thirteen Forms 4 and two Forms 144 were reviewed in full, together with eleven Schedule 13D/13G filings. Principal items: Caddis Holdings LP / Griffin Perry Form 4s of 30–31 March 2026 (11,000,000 shares, ~$56.3mm, not 10b5-1) and Form 144 of 30 March proposing 20,000,000 shares / $107.2mm; officer sell-to-cover transactions of 8–9 April 2026 at $4.58–$4.91; the grant cluster of 1 June 2026 (1,022,012 shares to seven directors and officers); and Rick Perry’s sale of 863,637 shares at $7.31 on 30 June 2026 via EPG Holdings. Schedule 13D filed 15 May 2026 by Caddis Holdings LP (CIK 0002123470), amended 2 July and 8 July, converted to Schedule 13G on 20 July — filed in SUPPORT of the incumbent board, not by the dissident.

Proxy-contest filings. Forty-seven DFAN14A, three PRRN14A, two PREC14A, two DEFC14A, one PREN14A, one PRER14A and four DEFA14A filings were reviewed across the period 6 May – 8 July 2026, comprising both the dissident’s solicitation (Toby Neugebauer / Vicksburg Investments Management LLC / Melissa A. Neugebauer 2020 Trust) and the company’s responses. Company DEFC14A: https://www.sec.gov/Archives/edgar/data/2071778/000121390026067954/ea0291106-07.htm

B.2 Primary — earnings-call transcripts

Both retrieved via the ROIC.ai MCP (get_earnings_call_transcript, get_latest_earnings_call), accessed 24 July 2026. Caveat applied throughout: these transcripts are machine-generated and contain audible-transcription corruptions (“Premium America”/“firming” for Fermi America/Fermi; “end of 2017” for 2027; “200 gigawatts” for 200 megawatts; “Board of Regions” for Board of Regents). Quotations were used only where meaning is unambiguous, and every figure was reconciled to a filing before use.

B.3 Primary — regulatory and legal

• Texas Commission on Environmental Quality (TCEQ): final approval of the ~6 GW clean-air permit, 25 February 2026; incremental 5 GW application filed 27 March 2026 (pending); denial of contested-case hearing requests, 24 March 2026.
• US Nuclear Regulatory Commission: combined-licence application for four Westinghouse AP1000 units, dockets 52-051 through 52-054. Part 1 filed 17 June 2025; Part 2 filed 20 August 2025; docketed 5 September 2025; EIS scoping 20 March 2026. Part 3 (site-specific) remains unfiled. 10 CFR 50.10 governs the prohibition on construction prior to issuance of a permit or licence.
• Federal Energy Regulatory Commission: co-located-load policy proceedings, including the November 2024 rejection of the Talen/AWS behind-the-meter interconnection service agreement and the December 2025 finding on PJM co-located-load rules.
• Lupia v. Fermi Inc., No. 1:26-cv-00050 (S.D.N.Y.), filed 5 January 2026 — Securities Act §Section 11/15 and Exchange Act §Section 10(b)/20(a) against the company, its directors and officers and the IPO underwriters; class period 1 October – 11 December 2025.
• 340 Energy v. Firebird LNG, Business Court of Texas No. 26-BC11B-0016 — fraudulent-transfer claim relating to the MAD Energy turbine transaction; trial set 24 May 2027. Disclosed in Fermi’s own 10-Q.
• Neugebauer v. Fermi Inc., Business Court of Texas No. 26-BC01B-0034 (filed 1 May 2026), and Fermi’s declaratory action in the Northern District of Texas (TRO denied 13 May 2026; notice of non-suit filed 1 July 2026).

B.4 Quantitative data sources

Data-integrity note. Several widely available data providers report a market capitalisation near $4.03bn, computed on the prior close of $6.32 and a stale share count. The figure used throughout this report is 637,574,239 shares × $7.40 = ~$4.72bn, taken from the 10-Q cover page of 11 May 2026, with a further 12,127,558 shares admitted on 23 July 2026 noted separately. Any analysis built on the $4.03bn figure understates enterprise value by roughly 17% on a basic basis and considerably more fully diluted.

B.5 Comparable-company research relied upon

The following companies were analysed as comparables for peer multiples, $/MW benchmarks, cost-of-capital data points, contracted-MW figures and industry framing, using their own public filings and disclosures. Attributed as prior published work in this series.

APLD · CORZ · IREN · WULF · CIFR · GLXY (neocloud / data-center converts) · OKLO · SMR · BWXT · CCJ · LEU (nuclear and SMR — the “priced on a promise” analogues) · VST · CEG · TLN · NRG (independent power) · EQIX · DLR (data-center REITs — stabilised yields and cost of capital) · GEV · POWL · AGX · SEI (turbine, switchgear and EPC supply chain).

Solaris (SEI) and Oklo (OKLO) were the two most load-bearing comparables — the former for the ~$3,500 per contracted kW benchmark, the latter for the ~$536 per announced kW benchmark against which Fermi’s ~$464/kW is measured.

B.6 Industry data and third-party research

• CBRE North America Data Center Trends — 1.4% primary-market vacancy; $195.94/kW-month asking rates (+6.5%); 74.3% of capacity under construction pre-leased; Northern Virginia 0.3%, Dallas–Fort Worth 1.8%.
• GE Vernova Q2 2026 results, 23 July 2026 — backlog plus slot reservations of 116 GW; 20 GW booked against 3 GW shipped; “mostly sold out through 2030”; 2026 order pricing 10–20 points higher per kW.
• Siemens Energy — €136bn backlog; Gas Services 65% data centers. Mitsubishi Power — sold out into 2028.
• Hyperscaler capital-expenditure guidance, calendar 2026 — Amazon ~$200bn, Google $175–185bn, Meta $115–135bn, Microsoft ≥$120bn, Oracle ~$50bn.
• ERCOT — public statements by CEO Pablo Vegas and General Counsel Chad Seely disowning the preliminary large-load forecast; downward adjustment filed 18 May 2026. (Included for completeness of Texas market context only; Project Matador is in SPP, not ERCOT.)
• Sightline Climate / Sightline Institute — base case that 30–50% of the announced 2026 data-center pipeline never materialises.
• Vogtle Units 3 & 4 cost record — ~$36.8bn against a ~$14bn estimate; ~$10,800–16,000/kW realised, used as the nuclear capex reality check.
• Peer project announcements: Crusoe/Lancium Stargate (Abilene), Vantage “Frontier” (Shackelford County), Galaxy Helios (Dickens County), and the terminated Poolside/CoreWeave “Project Horizon” (Longfellow Ranch).

B.7 Secondary press and third-party analysis

Used for triage and context. Every material claim was checked against a primary source where one exists; claims that could not be so verified are labelled as press reports or allegations in the memo and are not relied upon for any verdict.

• Business Insider, 17 December 2025 — identified the First Tenant as Amazon. The company categorically denied this.
• Bloomberg, 1 May 2026 — corroborated the Amazon identification via sources and reported that talks broke down over term length (20 → 15 years) and a conclusion that Matador would “reliably supply less electricity than Fermi claimed.” Unconfirmed by any filing; treated in the memo as a press report and flagged as the single most important open question.
• Politico, ~20 March 2026 and Bloomberg, ~25 March 2026 — reports of the then-CEO’s public conduct at Nvidia GTC and CERAWeek, which preceded Texas Tech’s characterisation of the conduct as a potential lease default.
• Washington Post, 28 April 2026 — satellite imagery showing no significant construction progress between February and April 2026, against the company’s claim that the initial construction phase was complete.
• Cleanview, April 2026 — commissioned satellite imagery reaching a similar conclusion; not a short seller.
• Fuzzy Panda Research, 20 April 2026 (“Matador Trained on Fraudulent Transfers & Drunk on AI Hype”) and 7 May 2026 (TCEQ permit modelling). Disclosed short position. Its verifiable claims reconcile to primary filings; its fraud and personal-conduct allegations are uncorroborated by any court finding or regulator, are identified as short-seller allegations in the memo, and carry no weight in any verdict. No reports on FRMI were found from Hunterbrook, Kerrisdale, Culper, Bleecker Street, Muddy Waters, Grizzly, Blue Orca, Spruce Point, Morpheus, Night Market or Iceberg.
• Amarillo Tribune — obtained the ground lease by public-records request, October 2025. Texas Tribune — reporting on Ogallala aquifer water use.
• Company press releases via PR Newswire and Access Newswire, including the LSE admission of 12,127,558 shares (23 July 2026), the arrival of three Siemens SGT6-5000F turbines at the Port of Houston (21 July 2026), and the Q2 2026 earnings date of 13 August 2026.
• Sell-side research referenced only as to published ratings and price-target changes (UBS, Evercore ISI, Cantor Fitzgerald, Mizuho, Berenberg, Rothschild Redburn, Stifel, Macquarie, Panmure Liberum, Citizens JMP). Note that the October 2025 initiation wave was almost entirely underwriter research; the coverage discussion below discusses the staleness of headline consensus.

B.8 Analytical frameworks

• Competition Demystified (Bruce Greenwald & Judd Kahn) — barriers-to-entry taxonomy (supply/cost, demand/captivity, economies of scale plus captivity), market-share-stability and ROIC tests, and asset-reproduction versus earnings-power value. Applied in Section 4 and Section 10.
• Capital Returns (Edward Chancellor / Marathon Asset Management) — supply-side capital-cycle analysis and the asset-growth anomaly. Applied in Section 3.6 and Section 5.1.

All figures in this report reconcile to the SEC filings cited above. Where a third-party source and a filing disagree, the filing governs and the discrepancy is noted in the text.