Fermi is not an AI stock that fell out of bed with the rest of the AI trade. That is the lazy read, and the tape refuses to support it. Over the twelve months to 4 September 2026, Bloom Energy rose 361% and Nokia rose 118% on the same AI-power thesis, while Fermi (Nasdaq: FRMI) fell from a $28.78 first close on 2 October 2025 to $5.12 — an 83% drawdown from its $30.01 peak close of 13 October 2025, and only 11% above the all-time closing low of $4.63 printed on 1 September 2026. The AI-power trade worked. Fermi did not. The distinction matters because it tells you what is actually being priced: not demand for gigawatts, but the risk that this particular developer runs out of money, time or patience before its first turbine spins commercially in July 2027.
Here is the part almost nobody writing about FRMI has done the arithmetic on. In July 2026 Fermi sold $431 million of 5.00% convertible notes due 2031 with a conversion price of roughly $9.52 a share, and spent $34.5 million on a capped call on top. That instrument is the company’s own, contemporaneous, cash-backed statement about where this equity is worth converting — $9.52 — struck when the stock was around $7.30. The market now pays $5.12, a 46% discount to the strike the company itself set eight weeks ago. Convertible arbitrage desks own that gap. It is the single cleanest anchor for a scenario framework on a pre-revenue business, and it is why our bull case sits at $12 (converts comfortably in the money) and our bear case at $2.50 (the converts become the fulcrum security and the equity is the residual).
Key facts: Fermi (FRMI) at a glance
- Share price: $5.12 at the 4 September 2026 close, a market capitalisation of about $3.3 billion on the 640.5 million shares outstanding at 10 August 2026; 52-week range $4.47–$36.99 — stockanalysis.com, 4 Sep 2026
- Q2 2026 revenue: $0. The company is pre-revenue — Q2 2026 results presentation, 13 Aug 2026
- Q2 2026 net loss: $26 million, or $0.04 a share; quarterly cash burn $49 million, down about 50% quarter on quarter
- Property, plant and equipment: $1.55 billion, after $185 million of capex in Q2 alone
- Balance sheet: $92 million cash and restricted cash at 30 June 2026; net debt about $520 million before the July convertible (Q2 2026 presentation)
- Convertible notes: $431 million at 5.00%, due July 2031, conversion price ~$9.52, net proceeds $417 million, capped call cost $34.5 million
- Anchor contract: TensorWave, ~$6.5 billion of aggregate contracted revenue across 222 MW, 15 years plus two five-year extensions
- Permitted capacity: 6 GW of federal air permits approved, a further 5 GW filed, on an 8,400-acre site outside Amarillo, Texas
What Fermi actually is — and why the market misfiles it
Fermi is not a technology company. Strip the AI vocabulary away and what remains is a build-to-suit industrial landlord: 8,400 acres of west Texas, a private generation fleet, and long-dated leases that bundle power with space. The company sells electrons and floor area to tenants who cannot wait in an interconnection queue. In structure, the closest analogue is not Nvidia or CoreWeave; it is a pre-let logistics developer that also happens to own the substation.
That framing changes which numbers matter. For a chip company you underwrite gross margin and design wins. For a pre-let developer you underwrite three things only: the cost to build, the contracted rent per unit of capacity, and the gap between the two funded with somebody else’s balance sheet. Fermi has now disclosed all three, which is more than most of its listed peers in this category can say.
The build is real and it is physically visible. Through Q2 2026 the company had installed 4.6 miles of natural gas lines, 11.3 miles of perimeter fencing and 7.2 miles of water distribution, and prepared roughly 300 acres — 13.1 million square feet — of land. Equipment on the ground or on order includes seven GE TM2500 aeroderivative units (126 MW, two-month delivery), three GE Vernova FR6B frames (116 MW, refurbishment complete), six Siemens SGT-800 turbines (296 MW simple cycle, 396 MW combined) and three Siemens SGT6-5000F units (728 MW simple, 1,042 MW combined). Gas supply of 220,000 MMBtu a day is secured, with five major pipelines within twenty miles.
The site sits in the Texas Panhandle for a reason that has nothing to do with tax and everything to do with molecules: Amarillo is close enough to Permian and Anadarko gas that fuel logistics stop being the binding constraint. As chairman Marius Haas put it on the Q2 call, “Power is the binding constraint on AI compute, not capital, not demand.” That is the whole thesis in eleven words, and the market currently disagrees only about who captures the rent.
The response from customers, lenders and the courts
The most important corporate response of 2026 came from a customer. TensorWave, an AI cloud operator, signed a 15-year lease with two five-year extension options covering 222 MW of Phase 1 capacity, worth approximately $6.5 billion in aggregate contracted revenue. Chief commercial officer Anna Boffa’s framing on the results call was blunt: “The market did not just validate our project and concept, it signed up for it.”
Run the arithmetic the release does not run for you. $6.5 billion across 222 MW over 15 years is roughly $1.95 million of revenue per megawatt per year, or about $29.3 million of contracted lifetime revenue per megawatt. That single ratio is the most useful number Fermi has ever disclosed, because it converts every capacity milestone into a revenue estimate. Phase 1 at 640 MW, targeted for full delivery in Q4 2027, implies roughly $1.25 billion of annual contracted revenue if leased at TensorWave economics. The 4.8 GW the company targets within 30 months, including roughly 2.6 GW through the Hillcore build-own-operate-transfer partnership, implies a number an order of magnitude larger again. Those are not forecasts; they are the company’s own contract price applied to the company’s own capacity plan.
Equipment vendors have responded with paper, not just steel. Fermi has a $500 million facility against the Siemens F-class turbines ($445 million drawn), $120 million against high-voltage equipment ($77 million drawn), $165 million against the SGT-800 turbines ($15 million drawn) and a $156 million general corporate promissory note that expires on 30 September 2026 with nothing drawn against it. Total liquidity commitments run to roughly $1.4 billion. Vendor financing at that scale is a credit judgement by GE Vernova and Siemens Energy about a company the equity market is pricing for distress — a genuine disagreement between two sets of professionals looking at the same asset.
The third response came from a courthouse. In August, Fermi disclosed that it had received a subpoena seeking documents relating to the Project Matador campus, and the shares fell about 4% on the session, as Seeking Alpha reported. No allegation of wrongdoing has been established, and a document request is not a charge. But for a pre-revenue developer whose entire equity value is a discounted stream of future rent, headline risk of that kind raises the cost of every future dollar of capital — which is exactly what a stock at $5.12 against a $9.52 conversion price is telling you.
Market impact: the numbers that set the bull and bear levels
Start with the burn, because it is the number that decides whether the story gets to 2027. Fermi lost $26 million in Q2 2026 and burned $49 million of cash, down roughly half from the prior quarter. Against $92 million of cash and restricted cash at 30 June plus $417 million of net convertible proceeds in July, that is a runway measured in years at the operating line — but capex is the real spend, and Q2 capex alone was $185 million. Property, plant and equipment now stands at $1.55 billion. The equity is a thin residual on a large, illiquid, single-purpose asset base.
Now the synthesis nobody publishes. Fermi’s contracted revenue per megawatt ($1.95 million a year) can be checked against what the listed independent power complex already earns. A merchant generator such as Vistra sells into a wholesale market and lives with hedges and basis risk; a Fermi-style tenant lease is closer to triple-net rent with a fuel pass-through. If even 55–60% of that $1.95 million per MW converts to EBITDA, 640 MW of leased Phase 1 capacity produces roughly $700–750 million of annual EBITDA. Put a 12x multiple on that — undemanding for contracted, long-dated infrastructure — and you get roughly $9 billion of enterprise value. Deduct about $1 billion of net debt post-convertible and divide by the 640,467,348 shares outstanding disclosed on the cover of the Q2 2026 10-Q, and you land near $12 a share. That is our bull case, and it needs no new customer beyond TensorWave-equivalent economics on Phase 1.
The bear case is a financing case, not a demand case. Assume first power slips beyond the 1 July 2027 target, that the undrawn $156 million promissory note lapses on 30 September 2026, and that the next tranche of capital has to be raised with the stock below the $9.52 conversion price. Every dollar then comes in as dilution or as secured debt ranking ahead of shareholders. On a $1.5 billion asset base carrying $520 million of net debt plus $431 million of converts, a stressed equity is worth something closer to $1.6 billion, or $2.50 a share. Note what that scenario does not require: it does not require AI demand to disappear, or the Amarillo permits to fail. It only requires the schedule to slip.
| Bull case ($12) | Bear case ($2.50) |
|---|---|
| Phase 1 (640 MW) leased at TensorWave economics — ~$1.25bn annual contracted revenue | First power slips past July 2027; capacity is energised but unleased |
| Converts move into the money; refinancing cost collapses | Next raise happens below the $9.52 conversion price — structural dilution |
| Vendor facilities ($785m committed) carry the build without equity issuance | The undrawn $156m promissory note lapses on 30 September 2026 |
| The 5 GW of filed air permits convert to approvals, extending the runway to 17 GW | Subpoena-driven disclosure raises the cost of every future dollar of capital |
For context on how differently the market treats contracted power once revenue is actually landing, compare Fermi with the merchant and nuclear names covered in our Vistra (VST) bull and bear analysis and the small-modular-reactor route laid out in the NuScale (SMR) prediction. Fermi is the only one of the three whose entire valuation rests on a construction schedule.
Regulation, permits and the political overhang
Fermi’s regulatory position is unusual: the constraint is not whether it may build, but how fast the paperwork moves relative to the turbine delivery schedule. The company holds federal air permits for 6 GW and has filed for a further 5 GW. That is a large approved envelope by any standard, and it is the reason the 17 GW long-range figure is quoted at all. The 30-month target of 4.8 GW includes roughly 2.6 GW under a build-own-operate-transfer partnership with Hillcore, which shifts some construction risk off Fermi’s own balance sheet.
The political dimension cuts both ways. The company was founded in January 2025 by former US energy secretary and Texas governor Rick Perry alongside Toby Neugebauer and Griffin Perry, and is now run by chief executive Lee McIntire, previously chief executive of TerraPower and chairman and chief executive of CH2M Hill. That roster opens doors in Austin and Washington. It also guarantees that any dispute involving the project becomes a political story rather than a construction story, which is precisely what the August subpoena disclosure demonstrated.
There is a broader regulatory tension worth naming. Behind-the-meter generation for data centres is the fastest way to add compute capacity to the US economy and the fastest way to add unhedged gas demand and emissions to a single county. Texas regulators have so far favoured speed. If that preference shifts — through interconnection rules, air-permit conditions or water allocation in the Panhandle — the 17 GW ambition compresses long before the 6 GW of approved permits are exhausted.
What happens next: three testable predictions
1. The 30 September 2026 promissory note is the next real signal. A $156 million general-corporate facility that expires undrawn tells you the company did not need it; an extension or a draw tells you it did. Watch that disclosure before the Q3 print.
2. A second anchor tenant — not more megawatts — is what re-rates the stock. Fermi has demonstrated it can permit and build. What it has not demonstrated is that TensorWave economics are repeatable with a second counterparty. One more 15-year lease at or near $1.95 million per MW per year would validate the ratio this entire valuation rests on, and would plausibly carry the shares back through the $9.52 conversion price. Without it, capacity announcements are cost, not revenue.
3. First power in July 2027 is the binary. Roughly 210 MW is targeted for that date, with Phase 1’s full 640 MW in Q4 2027. Delivery on schedule converts Fermi from a development story to a cash-flowing landlord and makes the bull case arithmetic above testable rather than theoretical. A slip of two quarters or more forces a capital raise at a price the current shareholder base will not enjoy.
Our own framework, having tracked this name since the October 2025 listing, is that the risk-reward is genuinely two-sided at $5.12 — which is rarer than it sounds in an AI-adjacent equity. The market is pricing schedule risk and financing risk almost to the exclusion of the contracted revenue already signed. That is defensible. It is not obviously correct. Readers tracking the same capital-flow question from the demand side will find the counterpart analysis in our coverage of IREN’s bull and bear case and the broader energy-stock complex feeding AI demand.
Frequently asked questions
What is the Fermi FRMI stock prediction for 2027?
Our framework puts a $12 bull case and a $2.50 bear case against the $5.12 close of 4 September 2026. The bull case requires Phase 1’s 640 MW to be leased at roughly TensorWave economics of $1.95 million per MW per year and first power to land on schedule in July 2027. The bear case requires only a material schedule slip that forces a capital raise below the $9.52 convertible conversion price.
Why has Fermi stock fallen 83% from its high?
Fermi listed in October 2025 at a valuation that priced years of execution in advance, peaking at a $30.01 close on 13 October 2025. The de-rating since has been driven by the passage of time without revenue: the company reported $0 of revenue and a $26 million net loss in Q2 2026, and disclosed a subpoena relating to Project Matador in August 2026. Demand for AI power has not weakened; the market has simply re-priced who captures it and when.
Is Fermi a real estate company or an AI company?
Functionally it is a build-to-suit industrial landlord that sells power alongside space under 15-year leases, which is why its economics are best measured in revenue per megawatt rather than gross margin. The AI label describes its tenants, not its business model.
How much cash does Fermi have?
The company reported $92 million of cash and restricted cash at 30 June 2026, then raised $431 million of 5.00% convertible notes in July for net proceeds of $417 million. Total liquidity commitments, including vendor equipment facilities, run to approximately $1.4 billion against quarterly capex that reached $185 million in Q2 2026.
What does the TensorWave contract mean for Fermi’s revenue?
The 15-year lease covers 222 MW for approximately $6.5 billion of aggregate contracted revenue — about $1.95 million per megawatt per year. Applied to Phase 1’s 640 MW target, that implies roughly $1.25 billion of annual contracted revenue once fully delivered in Q4 2027, though none of it is recognised until power flows.
When does Fermi expect first power?
Approximately 210 MW is targeted for 1 July 2027, with the full 640 MW of Phase 1 due in Q4 2027. The company has a 30-month target of 4.8 GW, including about 2.6 GW through a build-own-operate-transfer partnership with Hillcore.
This article is analysis, not investment advice. Prices and scenario levels are as of the 4 September 2026 close.