Stock Analysis · Bulls vs Bears

Fermi Inc. FRMI

🐂 Bulls vs Bears 🐻Report #213 Oct 2026
★ Willem owns this stock · 1.5% of the portfolio
80
very high risk▲ +3 vs last
🐂 9vs12 🐻
Bears lead
30%
▲ Chance of the biggest tailwind the forecasts miss
65%
▼ Chance of the biggest headwind the forecasts miss
The verdict

The evidence leans bearish for the next three months. Fermi's only lease still lacks the investment-grade guarantor management promised "in the coming days" on 2026-08-13. Pro-forma cash of about $445M covers less than two quarters of its own build plan, and its turbine lender's 10% quarterly amortisation starts 2026-12-31 because 222 MW is below the loan's 400 MW test. What the bulls point to is real and scarce…

The Bulls vs Bears podcast is being made
Two AI hosts debate this report, NotebookLM-style. It appears here automatically within a few hours.

Written by Claude (Anthropic’s AI) from public sources · Willem owns FRMI · AI can make mistakes — numbers in brackets point to the sources · Not financial advice

🐂 Bull arguments 9

  1. 11.5 GW of landed turbines while new ones take 3+ yearsStrong
  2. 2TensorWave rent prices at or above hyperscaler-backed peer leasesMedium
  3. 3A permitted 6 GW site outside ERCOT's large-load queueMedium
  4. 4Builders now run it, with named operators lined upMedium
  5. 5The stock already prices failure: EV ~1.7x book PP&EMedium
  6. 6Demand for deliverable AI power still outruns supplyMedium
  7. 7Hillcore could fund 2.6 GW of generation off balance sheetWeak
  8. 8Convertible debt is long-dated and has no maintenance covenantsWeak
  9. 9Nuclear licensing is moving, giving distant option valueWeak

🐻 Bear arguments 12

  1. 1The guarantor is nine weeks late; TensorWave is worth a quarter of the leaseHigh
  2. 2Cash covers under two quarters of the build planHigh
  3. 3MUFG's 10% amortisation starts 12-31: 222 MW misses the 400 MW testHigh
  4. 4Every dated milestone has slipped 6–12 months; nothing energisedHigh
  5. 5Three hard deadlines in eight weeks can cascadeHigh
  6. 6Disclosures keep running ahead of signed documentsMedium
  7. 7Federal subpoena, SEC request and a $415M claim overhang financingMedium
  8. 8A 22.5% founder at war with a board that entrenched itselfMedium
  9. 9Peers sign hyperscalers or Google backstops; Fermi signs a startupMedium
  10. 10Turnkey build puts ~$3B of construction risk on FermiMedium
  11. 11AI credit is tightening just as Fermi needs project debtMedium
  12. 12Insiders sold, nobody bought, and shorts rose twentyfoldLow

Bull arguments show how well they survive the devil’s advocate (strong / medium / weak); bear arguments show the damage they could do after the counter-argument (high / medium / low).

Bull case

🐂 Bull #11.5 GW of landed turbines while new ones take 3+ years

  • The point: the scarcest input in AI infrastructure is power that can be delivered by 2028. Fermi already owns or leases about 1.5 GW of gas turbines, while anyone ordering new turbines today waits years and pays far more.
  • Evidence:
    • Fleet [1,10]:
      • 6 Siemens SGT-800s (~300 MW), landed Feb 2026;
      • 3 Siemens SGT6-5000F F-class (up to 728 MW simple-cycle), on site in September [22];
      • 3 refurbished GE 6Bs (~114 MW), refurbishment complete per the Q2 deck;
      • 7 leased GE TM2500s (~132 MW), held by the lessor until mid-2027.
    • GE Vernova's gas backlog plus slot reservations reached 116 GW in Q2 2026. It expects at least 125 GW by year-end and output of only 20 GW a year today, rising to 24 GW in 2028 [43]. GE's CEO said H1-2026 pricing on new orders was 10–20 points higher per kW than Q4 2025 (trade-press quote, not in the 8-K) [43].
    • Siemens Energy: 69 GW firm gas-turbine backlog and lead times of "three years or more" [44].
    • Wood Mackenzie: new-build turbine prices are heading to about $600/kW by end-2027, +195% since 2019 [45].
  • Why it matters: a tenant that signs at Matador could have power in 2027–28 instead of 2030+. Even if the project stalls, the equipment has resale value in a sold-out market. That supports both the rent Fermi can charge and a floor under asset value.
  • 😈 Devil's advocate:
    • "Landed" is not "installed". The SGT-800 EPC with Primoris is still "working toward a final EPC agreement", and the F-class EPC (TSK) is "in negotiations". F-class go-live is "contingent on award" [10,11].
    • The lenders hold the equipment: MUFG (~$445M) on the F-class and CSG/Beal on the SGT-800s [1,5]. Any sale mainly repays lenders.
    • No source gives a secondary-market price for these specific units, so the resale "floor" cannot be measured.
    • Scarcity is fading slowly: GE alone plans to add 50% capacity by 2030 [43].
  • Verdict: Strong This is the one advantage rivals cannot copy quickly. It is an asset fact, not yet a business result.

🐂 Bull #2TensorWave rent prices at or above hyperscaler-backed peer leases

  • The point: the one lease Fermi signed implies rent per MW at the top of the peer range. That suggests Matador's power-first pitch commands a premium.
  • Evidence:
    • TensorWave: ~$6.5B over 15 years for 222 MW ≈ $1.95M per MW per year, about $163/kW-month (arithmetic on [1,20]).
    • Peers (division of contract value by term and critical MW): Applied Digital Polaris Forge 3 ~$1.67M, Polaris Forge 1 (CoreWeave) ~$1.83M, Delta Forge 2 ~$1.65M [40]. TeraWulf/Fluidstack ~$1.85M [41]. Cipher/Fluidstack ~$1.8M [42].
    • The variable power charge equals actual energy cost: "no commodity risk" for Fermi (CFO, Q2 call) [12].
  • Why it matters: if the template holds for future tenants, each GW leased is worth roughly $2B a year of revenue. Stifel's earlier model used $700–900M of EBITDA per GW [13].
  • 😈 Devil's advocate:
    • The comparison is not like-for-like. Fermi's figure is on total facility power and turnkey, and probably includes a fixed power charge. Peer figures are on critical IT MW. On the same basis, Fermi's premium may vanish.
    • Price is not the problem; credit is. Every peer deal has an investment-grade hyperscaler or a Google backstop (Google backstops $1.8B for TeraWulf and $1.4B for Cipher) [41,42]. Fermi's tenant has none yet.
    • Management declined to confirm the $700–900M EBITDA per GW when Stifel asked on the Q2 call [13].
    • One lease is one data point.
  • Verdict: Medium The headline price is good, but it is a turnkey gross figure with weaker credit behind it.

🐂 Bull #3A permitted 6 GW site outside ERCOT's large-load queue

  • The point: Matador has the permits and the inputs that new Texas sites are now fighting for, and it sits in SPP, outside ERCOT's new large-load rules.
  • Evidence:
    • TCEQ air permit for ~6 GW, final on 2026-02-25. A further ~5 GW has been applied for [3].
    • 99-year Texas Tech ground lease, 2.5 MGD of City of Amarillo water, ~4.6 miles of gas line laid, and an Energy Transfer supply arrangement [1,3,57].
    • ERCOT's large-load rules (NOGRR 282/NPRR 1308, approved 2026-07-09), a proposed $50,000/MW fee on large loads (PUCT Docket 58481), and the Governor's 2026-08-03 data-center audit all apply to ERCOT, not to SPP [1,50].
  • Why it matters: a buyer or tenant can skip a multi-year permitting and interconnection process. That is what makes the site valuable even to a strategic acquirer.
  • 😈 Devil's advocate:
    • SPP has its own new large-load process. FERC approved SPP's High Impact Large Load (HILL) rules on 2026-01-14: loads of 50 MW or more above 69 kV need enhanced studies and must accept curtailment [49]. Whether Fermi's 200 MW SPS service falls under HILL is not found.
    • Fermi's model of supplying power to tenants rests on the "incident of tenancy" exclusion in Texas Utilities Code §31.002(6), which covers power "not resold to or used by others" [48]. No authority found applies it to a data-center landlord charging a fixed power fee.
    • The 2027 Texas session has interim charges on data-center water and growth, and any statute could reach beyond ERCOT [50].
    • Amarillo considered a moratorium in May 2026, and critics dispute the claimed 80% water savings [57].
  • Verdict: Medium The permits are real. The legal basis for selling power to tenants is untested.

🐂 Bull #4Builders now run it, with named operators lined up

  • The point: the promoter era is over. A construction CEO and a professional operator stack should make Fermi bankable to lenders and guarantors.
  • Evidence:
    • CEO Lee McIntire (ex-Bechtel, CH2M Hill, TerraPower) since 2026-08-11. CFO Rob Masson permanent from 2026-07-20. Independent chair Marius Haas [1,18].
    • Named contractors and operators:
      • Primoris (SGT-800 balance of plant), TSK (F-class early works), Relevant Power Solutions (6B/TM2500) [10,11];
      • CBRE to operate the first data center (September) [22];
      • NAES to operate and maintain the turbine fleet (2026-09-28). McIntire: "Our first customer is signed, our turbines are secured, and our EPC partners are on board" [22].
    • Q2 net loss $25.8M versus $188.7M in Q1, which included $134M of share-based pay [1,10].
  • Why it matters: project lenders and guarantors underwrite the team and the contractors. These hires are preconditions for the debt that unlocks everything else.
  • 😈 Devil's advocate:
    • "EPC partners are on board" is not the same as signed EPC contracts. As of the 10-Q, neither turbine EPC was final [1,10].
    • The new team has no delivery record at Fermi: the CEO has been in the job nine weeks.
    • The new General Counsel, George Wentz, is CEO of MAD Energy. MAD Energy sold Fermi its first six SGT-800s, holds a net profits interest (2.5% of NOI on the first 1 GW, capped at $100M NPV), and is a co-defendant in the Firebird suit [1].
    • The material weakness persists, and remediation has no date [1].
  • Verdict: Medium It is the right profile and it is assembling the right names, but it has not converted that into a single signed EPC contract or an effective lease.

🐂 Bull #5The stock already prices failure: EV ~1.7x book PP&E

  • The point: at about $3.25, the market values the whole platform at not much more than the money already spent, with permits, the lease option and the turbines' scarcity value close to free.
  • Evidence:
    • Market cap ≈ 641.3M shares × $3.25 ≈ $2.08B [8,39].
    • EV ≈ $2.08B + ~$968M debt (gross $536.9M at 06-30 plus $431.25M converts) − ~$475M cash pro forma ≈ $2.58B, before Q3 burn (arithmetic on [1,6]).
    • PP&E at 06-30: ~$1.55B, all land and construction in progress [1]. EV is therefore about 1.7x book.
    • That is ~$0.43M per permitted MW (6 GW).
    • The stock is below every published target. The lowest is UBS at $6, and the average of 2026 targets is ~$11 (Mizuho $11, Stifel $17, Evercore $11, UBS $6, Cantor $8) [32,35,36,37].
  • Why it matters: if TensorWave closes with a guarantor and project debt follows, the gap to even the cautious targets is large.
  • 😈 Devil's advocate:
    • EV is above book, not below it. In a forced sale, construction-in-progress and turbines bought at peak prices could fetch less than cost.
    • Analyst targets are stale and mostly modelled before the August reset. Consensus revenue for FY2027 ranges from $0.66B to $2.05B across aggregators [33,34]. No target has been raised since the lease.
    • The ~$2.6B EV does not include the equity Fermi will need to raise before rent starts (Bear #2).
  • Verdict: Medium The stock is cheap relative to the plan, not relative to the assets or the funding need.

🐂 Bull #6Demand for deliverable AI power still outruns supply

  • The point: buyers of AI capacity will pay for any site that can deliver power soonest, so a site like Matador should fill eventually.
  • Evidence:
    • BNEF: capex at the 14 largest data-center operators is near $750B in 2026, and hyperscalers leased over $100B of neocloud capacity in the six months to March 2026 [52].
    • Global data-center vacancy was 4.9% in August 2026 (secondary) [60].
    • Applied Digital says expansion leases are being negotiated at "materially higher" rates [40].
    • Management on the Q2 call: other tenants are "very far along", across neoclouds and hyperscalers, all unnamed [12].
  • Why it matters: this would let Fermi sign more tenants and justify the 650 MW TensorWave option.
  • 😈 Devil's advocate:
    • Demand is for sites with a credible landlord. Fermi's first investment-grade tenant walked away in Dec 2025, at the peak of demand [3]. Business Insider reported that tenant was Amazon, citing Neugebauer; Fermi "categorically denies" that [30,31].
    • Big buyers are trimming. OpenAI/Oracle dropped the planned Abilene expansion in March 2026, citing grid delays, financing terms and shifting forecasts [55].
    • "Very far along" has been the message since November 2025 (Q3-25 call: first lease targeted for 2025-12-09) [17].
  • Verdict: Medium Demand is necessary, but it has not been enough at this site for a year.

🐂 Bull #7Hillcore could fund 2.6 GW of generation off balance sheet

  • The point: a third party would pay for, own and run the next big block of generation, so Fermi's capital can go into rent-earning buildings.
  • Evidence:
    • Framework signed 2026-08-11: build-own-operate-transfer of up to ~2.6 GW (combined-cycle gas plus solar and batteries) on a ~400-acre sublease [1,10].
    • Fermi takes at least 50% of tenant power on take-or-pay under a 20-year PPA. Its guarantee is capped at 12 months of fixed capacity charges. It has an option to buy after year 10 [1,10].
    • First ~350–360 MW within ~24 months of notice to proceed [10].
    • Definitive documents are due 90 days after signing (2026-11-09) [1].
  • Why it matters: at the CFO's $3–4M per MW benchmark for power, 2.6 GW would cost roughly $8–10B that Fermi would not have to raise [12].
  • 😈 Devil's advocate:
    • It is non-binding. Hillcore's financing capacity and turbine source were not found.
    • Take-or-pay means Fermi carries Hillcore's capacity charges if tenants don't show up. The 12-month cap limits but does not remove that.
    • It does nothing for the current tenant problem or the next 12 months of cash.
  • Verdict: Weak until definitive documents are signed.

🐂 Bull #8Convertible debt is long-dated and has no maintenance covenants

  • The point: the July raise bought time on terms that cannot trigger a default from weak results alone.
  • Evidence:
    • $431.25M of 5.00% notes due 2031-07-15. First coupon 2027-01-15. Senior unsecured, with no financial maintenance covenants (company statement) [6,7].
    • The conversion price is ~$9.52, so conversion dilution only bites in the success case (maximum 58.9M shares, 9.2% of basic) [6].
    • The Turbine Warehouse II (CSG/Beal) loan runs to about Dec 2028 [1].
  • Why it matters: the convertible holders cannot force a crisis before 2031 unless something else defaults first.
  • 😈 Devil's advocate:
    • The convertible is the least dangerous piece. The secured debt has hard dates: MUFG ($444.9M) matures 2027-08-10 and amortises from 2026-12-31 [5]. Keystone faces mandatory prepayment on 2026-12-31 if there is no approved customer agreement [1].
    • The MUFG cross-default threshold is only $5M [5], so trouble elsewhere spreads.
    • The capped call ($9.52–$14.64) used $34.5M of the proceeds and is worthless at $3.25 [6].
    • Holders can put the notes at par on a fundamental change [6], which complicates any takeover.
  • Verdict: Weak It removes one risk, but the secured lenders, not the convertible holders, set the timetable.

🐂 Bull #9Nuclear licensing is moving, giving distant option value

  • The point: Matador could host four AP1000 reactors in the 2030s. Fermi is the first applicant in the NRC's applicant-prepared environmental review pilot.
  • Evidence:
    • The NRC acknowledged receipt of Fermi's draft environmental impact statement in July 2026. Fermi says Matador "is the first project to complete these steps under the NRC's pilot program" [1,47].
    • Fermi is "in the pre-approval process" with DOE's Office of Energy Dominance Financing [1]. DOE offered up to $17.5B in conditional loans for up to 10 AP1000s on 2026-06-23. Westinghouse has seven potential partners, none named [46].
  • Why it matters: a licensed nuclear site is rare and could carry strategic value well beyond the gas phase.
  • 😈 Devil's advocate: Part 3 of the licence application was withdrawn in Dec 2025 and is targeted for re-tender by 2026-12-31. The first unit has slipped to 2033 [1]. Fermi is not shown among DOE's partners [46]. Each DOE project needs $500M of sponsor equity, which Fermi does not have [46].
  • Verdict: Weak Real, but 7+ years away and unfunded.

Bear case

🐻 Bear #1The guarantor is nine weeks late; TensorWave is worth a quarter of the lease

  • The point: the lease cannot close without an outside rent guarantor, and the tenant alone cannot carry $6.5B of rent. If closing fails, Fermi has no revenue path, no project debt and no answer to the year-end deadlines.
  • Evidence:
    • Closing conditions: guaranties, work letters, board approvals, an SLA, and "project-level financing sufficient to fund construction". Either side may terminate if they fail [1].
    • Timeline of the guarantor:
      • The original 2026-08-10 release said obligations were expected to be "guaranteed by one of the global leaders in AI" [20].
      • A "CORRECTION FROM SOURCE" the same day removed that line: "Unfinalized version was submitted" [21].
      • Q2 call (08-13): the backstop is investment-grade, covers "the full 15 years", and would be named "in the coming days" [12].
      • 2026-09-22: closing moved from 09-30 to 10-31 "to finalize that guaranty" [19].
      • No 8-K has been filed since 2026-08-31, and no guarantor has been named as of 2026-10-13 [25].
    • TensorWave raised $350M at a $1.55B valuation in June 2026, ~$493M in total, and discloses no revenue [26]. Its parent's guaranty of the tenant [1] is therefore worth little against ~$433M a year of rent.
    • The founder's claim that "the guarantee the Company first announced was rescinded within hours" [9] matches the corrected press release. It describes a withdrawn sentence, not a signed guarantee that was cancelled. No signed guarantee has ever been disclosed.
  • Why it matters: failure leaves the Texas Tech 200 MW condition unmet, triggers Keystone's prepayment and leaves the MUFG problem unsolved. It would be Fermi's second public tenant failure in under a year.
  • 🛡️ Counter-argument:
    • Both sides have publicly committed and extended once rather than walking away.
    • Fermi describes the guarantor as "a global leader in AI infrastructure" [1]. AMD Ventures co-led TensorWave's last round [26], so a strategic backer exists with a motive to keep a large AMD-only cloud alive.
    • Guaranty negotiations on a 15-year obligation are slow.
  • Severity: High It is binary, due in 18 days, and gates every other milestone.

🐻 Bear #2Cash covers under two quarters of the build plan

  • The point: Fermi needs billions before rent starts and has a few hundred million.
  • Evidence:
    • Cash at 06-30: $62.5M plus $29.2M restricted [1]. The convertibles added ~$382M after the capped call [6,12]. Pro-forma unrestricted cash is therefore ≈ $445M, before any Q3 flows (arithmetic).
    • Burn by quarter [1,2,3]:
      • Operating cash flow: −$25.9M (Q4-25), −$7.3M (Q1-26), −$48.7M (Q2-26).
      • Capex: ≈ −$480M (Q4-25), −$441M (Q1-26), −$185M (Q2-26).
      • Q1's operating figure was flattered by ~$22M of working capital. The CFO put underlying burn at ~$29M [14].
    • Plan: Phase 0+1 "could exceed $3 billion", with ~$2B in the next 12 months "subject to" TensorWave's conditions [1].
    • The 10-Q says resources are not sufficient against forecast disbursements, raising "substantial doubt", which management says its plans alleviate. Those plans include monetising equipment and "deferring or renegotiating" obligations. The 10-Q adds that renegotiation could lead counterparties to terminate [1].
    • Undrawn committed facilities: MUFG ~$55M (draw period ends 2026-11-10), Keystone ~$43M, CSG/Beal ~$150M (of which $22.9M is reserved for interest) [1,5]. The Yorkville note's availability window ended 2026-10-01 with nothing drawn at 06-30; later draws not found [1].
    • Runway (estimate): at ~$500M/quarter of capex plus ~$49M/quarter of operating burn, cash plus undrawn facilities cover roughly 1–2 quarters. With capex deferred, it is about 6–7 quarters. That falls to about 4 quarters after MUFG amortisation, a possible Keystone prepayment and the Texas Tech escrow.
  • Why it matters: even the success case needs sponsor equity alongside project debt. At ~$3.25, $1B of equity would mean about 300M new shares (+48%). The failure case is a rescue.
  • 🛡️ Counter-argument: capex is explicitly sequenced to signed tenants and financing [1]. Hillcore moves power capex off balance sheet. A guaranteed 15-year lease is financeable at project level, which caps the equity cheque.
  • Severity: High Dilution is likely in both the success and the failure case. Only its size differs.

🐻 Bear #3MUFG's 10% amortisation starts 12-31: 222 MW misses the 400 MW test

  • The point: a lender clock that report #1 partly misdated is now close. TensorWave alone does not satisfy it.
  • Evidence: MUFG Equipment Supply Loan Financing Agreement (closed 2026-02-10) [5]:
    • $444.9M drawn of $500M at Term SOFR + 4.0% (effective 12.2%), maturing 2027-08-10 [1].
    • "Minimum Principal Payment": $0 before 2026-11-10. After that, 10% of loans outstanding per quarter if no lease or offtake "for at least 400 MW" has been signed for Matador I by that date, or 5% if one has [5].
    • Quarterly payment dates are the last banking day of each quarter, so the first payment falls on 2026-12-31: ~$44.5M at 10% (arithmetic). Missing a payment is an immediate event of default with no grace period (§7.1(i)) [5].
    • §5.22: if no 400 MW lease is signed by 2026-11-10, the agent "shall be permitted to (but is not obligated to) start marketing the Collateral", but cannot sell or foreclose without an event of default [5].
    • Other tests: loan-to-value of 65% on delivered and 55% on undelivered equipment; the cross-default threshold is $5M; a judgment above $2M is an event of default [5].
    • TensorWave is 222 MW, and reaches 650 MW only if both expansion options are exercised [1]. Whether MUFG would count options or Hillcore as "offtake" is not found.
  • Why it matters: about $133M of principal is due across three quarters before a ~$311M balloon in Aug 2027 (arithmetic, if the balance is unchanged). That uses up most of the post-convertible cash unless the loan is refinanced, waived or the turbines are sold.
  • 🛡️ Counter-argument: "signed", not "effective", is the test, so a second lease signed before 11-10 would halve the rate. Lenders rarely enforce against a borrower in active financing talks, and a waiver is common. Fermi could also sell the F-class units, which are not needed for the TensorWave phase (F-class power is a 2028–29 objective [11]), and retire the loan.
  • Severity: High It is dated, contractual and near.

🐻 Bear #4Every dated milestone has slipped 6–12 months; nothing energised

  • The point: the company's timetable has been wrong at every checkpoint, and TensorWave's lease now puts money on the next one.
  • Evidence:
    • IPO prospectus (2025-10-01): "at least 1.1 GW of power by the end of 2026"; revenue "not expected until 2026" [4].
    • Q3-25 call (2025-11-11): "first gigawatt of power by the end of 2026"; first lease targeted for 2025-12-09 [17].
    • FY2025 letter (2026-03-30): "The initial Xcel grid connection is ready to go live this year" [16].
    • Q1-26 deck (2026-05-14): go-live dates of Q4-26 for 86 MW Xcel and 114 MW 6B, Q2-27 for the SGT-800s, Q3-27 for the TM2500s, and Q4-27 for the F-class [15].
    • Q2-26 (2026-08-13): these became "first 210 MW July 1, 2027", "640 MW nameplate by Q4 2027" and "about 200 MW of initial commercial power over the next six months". The F-class go-live is "contingent on award" [10,11].
    • No source confirms any SPS grid power is energised. The Q1 remark "We have also brought 86 megawatts of power from the Xcel to the site" [14] is not an energisation statement, and the Q2 release says work "continues toward 200 megawatts" [10].
    • Nuclear: first unit pushed from 2032 to 2033 [1].
    • Liquidity: in March the CFO said Fermi had liquidity for 12 months. The 10-Q in August carried going-concern language [1,3].
    • The one milestone delivered on time: F-class turbines arrived in July/September as scheduled [10,22].
  • Why it matters: TensorWave halls are due 2028-01-01 and 02-01. Late delivery brings escalating rent credits, and a phase that misses the outside date can be terminated with prepayments refunded [1]. The cap and the outside date are not public. The SGT-800s need ~15 months after EPC award [10], so an award after about April 2027 already threatens the January 2028 date.
  • 🛡️ Counter-argument: most misses happened under the founder. The new plan (640 MW for one tenant) is far smaller and more realistic. Physical work is real: the power-island excavation is done, the gas line is laid and the turbines are on site [10,22].
  • Severity: High The pattern is long, and the new team has not yet hit a dated target of its own.

🐻 Bear #5Three hard deadlines in eight weeks can cascade

  • The point: a handful of counterparties hold rights that fall due almost together. One miss can trigger others.
  • Evidence:
    • 2026-10-31: TensorWave closing [19].
    • 2026-11-10: MUFG 400 MW test and collateral-marketing right [5].
    • 2026-12-31:
      • Texas Tech may terminate the 99-year ground lease if no notice to proceed has been issued. That notice needs a ≥200 MW Phase-1 tenant sublease and Phase-1 financing. The $9M escrow and $2M prepaid rent are due by the same date, and the escrow was unfunded at 06-30 [1].
      • Keystone mandatory prepayment (~$77.3M drawn) if there is no approved customer agreement. Keystone also has a $20M minimum-liquidity covenant [1].
      • First MUFG amortisation payment [5].
    • TM2500 lease rent runs to 2045 with no termination for convenience; $35.97M already prepaid [1].
  • Why it matters: TensorWave closing with a guarantor solves Texas Tech and probably Keystone, but not MUFG (Bear #3). Failure hits all three within nine weeks.
  • 🛡️ Counter-argument: Texas Tech gets 1% of data-center value, 1% of power revenue and 25% of water revenue [1], so it gains more from extending than from terminating. The lenders face the same incentive with $1.5B already sunk. The 10-Q explicitly plans to "defer, scale, or renegotiate" [1].
  • Severity: High in Q4 2026, falling sharply if TensorWave closes.

🐻 Bear #6Disclosures keep running ahead of signed documents

  • The point: the gap between what Fermi announces and what it files makes every claim hard to rely on, and that is exactly what a guarantor or lender diligences.
  • Evidence:
    • The 08-10 lease release named a guarantor, then a correction deleted it the same day [20,21].
    • The chair's Q2 remarks called it "a binding agreement representing up to 650 megawatts". The 10-Q shows 222 MW committed, with the rest as tenant options [1,11].
    • "Our EPC partners are on board" (09-28) [22], while the 10-Q describes neither EPC as final [1].
    • The TensorWave lease, Fermi's single most material contract, has never been filed as an exhibit. There has been no 8-K since 2026-08-31; the 09-22 extension came by press release only [25].
    • The FY2025 letter: "We announce binding agreements when they are executed" [16]. The same letter disclosed the termination of the first tenant's $150M advance agreement [16].
    • The material weakness is still open: insufficient accounting staff, segregation of duties, informal processes [1].
  • Why it matters: investors cannot see the outside delivery date, the rent-credit cap, the prepayments or the guaranty size, which are the numbers that decide whether the lease is worth $6.5B. Credibility also affects the cost of every future raise.
  • 🛡️ Counter-argument: the lease may be filed once it is effective, which is common practice. The new General Counsel and CFO arrived in July, and controls work is under way [1].
  • Severity: Medium It does not destroy value directly, but it widens every discount.

🐻 Bear #7Federal subpoena, SEC request and a $415M claim overhang financing

  • The point: open criminal and civil inquiries plus large claims are a diligence red flag for the guarantor and project lenders Fermi needs right now.
  • Evidence [1,54]:
    • EDNY grand-jury subpoena dated 2026-07-30 for Project Matador records tied to "certain former members of management". SEC voluntary request dated 2026-08-03. Both are "broad in scope", with no allegation of wrongdoing.
    • Lupia v. Fermi (S.D.N.Y., filed 2026-01-05): Securities Act §11 and Exchange Act §10(b) class action, class period 2025-10-01 to 12-11. No loss estimate. Lead-plaintiff status not found.
    • Consultants' claim over 25,000 reclaimed Class B units (would have been ~11.25M shares), Texas Business Court, filed 2026-07-27: relief "exceeding $415 million". Fermi calls a loss "reasonably possible" but not probable, and has accrued nothing.
    • Firebird (≥$5.985M plus exemplary damages; motion to dismiss denied 2026-08-11) and Bayonne (≥~$8.5M).
    • Under the MUFG agreement, a judgment above $2M is an event of default [5].
  • Why it matters: $415M equals about $0.65 per share, or roughly 20% of today's market cap. The inquiries raise the price a guarantor will demand.
  • 🛡️ Counter-argument: the inquiries target former management. The company has won every early procedural round in the founder litigation [53]. The $415M figure is a pleading, not an assessed loss.
  • Severity: Medium It mainly slows dealmaking, with a tail risk if charges follow.

🐻 Bear #8A 22.5% founder at war with a board that entrenched itself

  • The point: the largest shareholder is suing the board, and the board has changed the rules to blunt him. Both sides add friction to any deal.
  • Evidence:
    • Neugebauer holds 144.1M shares (22.47%) [8]. He says he has "not sold a single share since the IPO" and does not want to return [9].
    • He was fired "for Cause" on 2026-04-30. The proxy lists alleged misrepresentations to the board, public communications in breach of duty, unauthorised meetings and abusive behaviour [8].
    • Litigation: suits in Texas Business Court were amended four times. The TRO was denied on 2026-09-16, and a Rule 91a motion to dismiss is set for 2026-11-19 [53].
    • Board changes: the bylaw supermajority was raised to 70% (05-13), and on 08-11 the director-election standard became a majority of all outstanding shares [8,9]. With 22.5% held by the founder, that standard makes contested elections hard for anyone to win.
    • New: his Rule 14a-8 proposal for an independent bank to review "extraordinary transactions" (submitted 09-10) does not appear anywhere in the 10-02 proxy, which lists only the election of Perry and Stein and EY's ratification [8,9]. Why it was omitted (no-action relief or otherwise) is not found.
    • Former CFO Everson left the board on 07-10, citing denied access to records and the finance committee's authority over financings [8].
  • Why it matters: the fight distracts a nine-week-old CEO, complicates any takeover (see the strategic-sale tailwind), and could escalate at the 10-30 annual meeting.
  • 🛡️ Counter-argument: the founder's special-meeting push drew minimal unaffiliated support, and he suspended it in July after ISS recommended against [59]. The courts have upheld the board's bylaws as allowed under Texas law [53].
  • Severity: Medium

🐻 Bear #9Peers sign hyperscalers or Google backstops; Fermi signs a startup

  • The point: in a market where landlords compete on credit as much as power, Fermi's competitors have stronger tenants and are already delivering.
  • Evidence:
    • Applied Digital: Polaris Forge 3 is a 15-year, 300 MW critical-IT lease with a high-investment-grade hyperscaler worth $7.5B, and five campuses total ~$36B of base-term rent [40].
    • TeraWulf and Cipher: Fluidstack leases backstopped by Google for $1.8B and $1.4B [41,42].
    • IREN energised Sweetwater 1 (1.4 GW) on 2026-05-01 [56].
    • Pacifico GW Ranch holds a 7.65 GW air permit, with Amazon reportedly the tenant (single source, unconfirmed) [61].
    • Fermi: one 222 MW lease, not effective, with a neocloud tenant valued at $1.55B [1,26].
  • Why it matters: the next hyperscaler choosing between sites will weigh Fermi's governance, liquidity and lack of track record against peers who have delivered.
  • 🛡️ Counter-argument: Fermi's permitted scale (6 GW, with 11–17 GW possible) is larger than most rivals'. Its SPP location avoids ERCOT congestion [1,50]. Demand is large enough for several winners [52].
  • Severity: Medium High. It explains why the guarantor is hard to land.

🐻 Bear #10Turnkey build puts ~$3B of construction risk on Fermi

  • The point: the first deal is turnkey "all the way up to the chips", so Fermi carries construction cost, schedule and completion risk before any rent.
  • Evidence:
    • CFO benchmarks of $10–12M/MW for a turnkey data hall plus $3–4M/MW for power imply ~$2.9–3.6B for 222 MW [12,13]. This matches the 10-Q's ">$3B" for Phase 0+1 [1].
    • Fermi has given a completion guaranty, owes escalating rent credits if late, and the tenant may terminate a late phase and recover prepayments [1].
    • The data-center contractor was "to be announced" as of Q2 [11]. CBRE is the operator, not the builder [22].
    • Section 232 tariffs of up to 50% on steel and aluminium derivatives and copper, and transformer lead times of ~128 weeks, raise cost risk (secondary) [63].
  • Why it matters: a 10% cost overrun on $3B is $300M, about 15% of today's market cap.
  • 🛡️ Counter-argument: the rent stack includes a fixed return on capital Fermi spends on buildings [3], which may pass some overruns to the tenant. Whether TensorWave's lease does this is not found.
  • Severity: Medium

🐻 Bear #11AI credit is tightening just as Fermi needs project debt

  • The point: lenders to AI infrastructure have become pickier about neocloud tenants, which is exactly Fermi's tenant type.
  • Evidence:
    • CNBC (2026-09-15): new data-center debt is pricing significantly wider as investors weigh an AI slowdown (snippet only) [51].
    • Neoclouds sign 15-year leases while their own customer contracts run hours to a few years, so lenders now demand investment-grade guarantees, collateral or chips (Yahoo Finance, snippet) [62].
    • OpenAI/Oracle dropped the Abilene expansion (March 2026) [55].
  • Why it matters: this is why the guarantor is a closing condition. It also means project debt for a neocloud-only campus will be expensive, raising the equity share.
  • 🛡️ Counter-argument: hyperscaler capex is still rising (~$750B in 2026) [52], and an investment-grade guarantor would put Fermi in the bankable category regardless of the cycle.
  • Severity: Medium

🐻 Bear #12Insiders sold, nobody bought, and shorts rose twentyfold

  • The point: no insider has bought on the 85% fall, and the short position keeps building.
  • Evidence:
    • No open-market purchase by any director or officer since the IPO [24].
    • Caddis (Griffin Perry) sold 11.0M shares on 03-30/31 at $5.02–5.54 (~$56M). Rick Perry sold 863,637 shares at $7.31 on 06-30 under pre-IPO Series C option exercises. Officers' April sales were sell-to-cover blocks [24].
    • Short interest went from 3.3M shares (2025-10-15) to 66.8M (2026-09-30): ~10.4% of shares, 3.7 days to cover [38].
    • Fuzzy Panda (short, 2026-04-20) alleges a "death spiral" Yorkville financing, construction halts and related-party links. These are allegations from a short seller and partly inconsistent [29].
  • Why it matters: it is a signal about how close holders read the odds. A crowded short also amplifies moves in both directions around 10-31.
  • 🛡️ Counter-argument: the largest holder (22.5%) says he has not sold [9]. Caddis says it still supports the board [24]. Sell-to-cover is mechanical. At 3.7 days to cover, the short is not extreme.
  • Severity: Low

Tailwinds NOT in the forecasts

Consensus is unreliable. Aggregators show FY2027 revenue anywhere from $0.66B (StockAnalysis, updated 08-14) to $2.05B, and FY2028 revenue of $1.24B with EPS of $1.47 (one analyst, Tickflow) [33,34]. Mizuho's model assumes ~1 GW of leased revenue through 2028 [59]. Interpretation: consensus already assumes TensorWave closes and at least one more tenant, so a TensorWave closing or a 650 MW expansion is mostly in the forecasts and is not listed here. The tailwinds below are things no model includes. Fermi is pre-profit, so impacts are in $/share of cash or per-share value, as stated; the FY2028 EPS baseline is $1.47 (single analyst, low reliability).

  • Turbine fleet monetised at or above loan value — 30% chance within 12 months

    • Impact on revenue: none in FY2028. The F-class units are not needed for the TensorWave phase [11].

    • Impact on EPS: about +$0.25 per share (+17%) vs consensus FY2028 EPS, through dilution avoided (per-share basis).

    • How we got the number (estimate): selling the F-class at ≥ the ~$445M MUFG balance retires that loan instead of refinancing it with equity; $445M ÷ $3.25 ≈ 137M shares avoided (+21%) → 1 − 1/1.21 ≈ 17% × $1.47 ≈ +$0.25.

    • Expected value: 0.30 × $0.25 ≈ +$0.08 per share.

    • What it is: a sale, sale-leaseback or JV of the three F-class units (up to 728 MW) or other turbines to a utility or developer, at prices that new turbines (~$600/kW heading into 2027 [45]) make plausible.

    • Why it's not priced in: the 10-Q lists "monetisation of equipment" only as a going-concern mitigant [1]. No analyst models it.

    • Trigger to watch: an 8-K on an equipment sale, or MUFG marketing the collateral after 2026-11-10 with Fermi's consent.

    • Why this probability: buyers exist in a sold-out market and the lender has the right to market from 11-10. But selling the flagship turbines would cut the campus's later capacity, and management may resist while it hopes for project debt.

  • Hillcore definitive and funded for the first ~360 MW block — 30% chance within 12 months

    • Impact on revenue: none directly in FY2028. It enables capacity for tenants beyond TensorWave.

    • Impact on EPS: about +$0.22 per share (+15%) vs consensus FY2028 EPS, through equity avoided.

    • How we got the number (estimate): 360 MW × $3.5M/MW ≈ $1.26B of power capex avoided; assume 30% would have been equity ≈ $378M ÷ $3.25 ≈ 116M shares (+18%) → 1 − 1/1.18 ≈ 15% × $1.47 ≈ +$0.22.

    • Expected value: 0.30 × $0.22 ≈ +$0.07 per share.

    • What it is: binding BOOT agreements plus Hillcore's own financing [1,10].

    • Why it's not priced in: it is a non-binding framework. Consensus does not split capex funding.

    • Trigger to watch: definitive documents around 2026-11-09, or a Hillcore turbine order or financing announcement.

    • Why this probability: Hillcore's take-or-pay protection depends on Fermi having tenants, so this likely moves only after TensorWave closes, which itself is uncertain.

  • Strategic sale or takeover of the company or the Matador site — 10% chance within 36 months

    • Impact on revenue: not applicable.

    • Impact on EPS: not quantifiable. There is no basis for a premium; the takeout value is the price, not an EPS change.

    • How we got the number (estimate): not quantifiable. No bids have been reported and no comparable asset sale was found.

    • Expected value: not quantifiable.

    • What it is: a hyperscaler, utility, infrastructure fund or energy company buys Fermi or the site.

    • Why it's not priced in: the price reflects standalone distress. The board says a sale now would "grossly undervalue" Fermi [7].

    • Trigger to watch: the board hiring a bank, the founder's review proposal resurfacing, or press reports of approaches.

    • Why this probability: the asset is rare. But a 22.5% hostile holder, federal inquiries, a fundamental-change put on the convertibles and the 70% bylaw all complicate any deal. Lower than report #1's 18% because the board has removed the proposal from the ballot.

  • NRC Part 3 re-tender plus a DOE conditional commitment for AP1000s — 10% chance within 36 months

    • Impact on revenue / EPS: not quantifiable; it affects the 2030s, not FY2028.

    • How we got the number (estimate): not quantifiable. No revenue before 2033 [1].

    • Expected value: not quantifiable.

    • What it is: Fermi gets DOE financing backing for nuclear and a complete licence application.

    • Why it's not priced in: no model includes nuclear.

    • Trigger to watch: the Part 3 filing by 2026-12-31; Fermi named in a DOE announcement.

    • Why this probability: each project needs $500M of sponsor equity [46], which Fermi lacks.

Headwinds NOT in the forecasts

Ranked by expected damage to per-share value (judgement, as the units differ). EBITDA per share assumes $0.8M EBITDA per MW per year, the midpoint of Stifel's earlier $700–900M per GW, which management declined to confirm [13]; 641.3M shares.

  • Highly dilutive equity raise or rescue financing — 65% chance within 18 months

    • Impact on revenue: none.

    • Impact on EPS: about −$0.50 per share (−34%) vs consensus FY2028 EPS of $1.47, per-share dilution.

    • How we got the number (estimate): $1.0B raised at ~$3.00 ≈ 333M new shares (+52%) → 1 − 1/1.52 ≈ 34% × $1.47 ≈ −$0.50.

    • Expected value: 0.65 × −$0.50 ≈ −$0.33 per share.

    • What it is: common or preferred equity, a PIPE, or rescue debt with warrants, to bridge to rent in 2028.

    • Why it's not priced in: analyst targets ($6–17 for 2026 notes) imply little further dilution. No shelf or ATM has been filed yet, so no model carries the share count [25].

    • Trigger to watch: an S-3 or ATM filing; a PIPE; Q3 cash (11-12) below ~$300M.

    • Why this probability: about $2B of 12-month capex against about $0.45B of cash. Even with deferral and Hillcore, sponsor equity behind project debt is very likely, and failure makes it certain. Slightly above report #1's 60% because the guarantor is still missing.

  • TensorWave lease not effective with an investment-grade guaranty by 2026-12-31 — 45% chance

    • Impact on revenue: −$433M (−35%) vs consensus FY2028 revenue of $1.24B.

    • Impact on EPS: about −$0.28 EBITDA per share in FY2028 (consensus EBITDA not found, so no % given).

    • How we got the number (estimate): $6.5B ÷ 15 years ≈ $433M/yr; 222 MW × $0.8M EBITDA/MW ≈ $178M ÷ 641M ≈ −$0.28.

    • Expected value: 0.45 × −$0.28 ≈ −$0.13 EBITDA per share.

    • What it is: termination, or repeated extensions without a guarantor.

    • Why it's not priced in: consensus models revenue from 2027 [33,34].

    • Trigger to watch: an 8-K on or before 10-31, either naming the guarantor and confirming effectiveness, or announcing another extension or termination.

    • Why this probability: the guarantor is nine weeks past "the coming days", and TensorWave cannot stand alone. But both sides keep extending, and a strategic backer (AMD) exists. Up from 35% because no progress has been disclosed since 09-22.

  • MUFG 10% amortisation from 2026-12-31 without a waiver — 80% chance

    • Impact on revenue: none.

    • Impact on EPS: −$0.28 of levered free cash flow per share in FY2027 (no EPS effect; principal).

    • How we got the number (estimate): 10% × ~$445M ≈ $44.5M per quarter × 4 ≈ $178M ÷ 641M ≈ −$0.28. The balloon of ~$311M is due 2027-08-10.

    • Expected value: 0.80 × −$0.28 ≈ −$0.22 FCF per share.

    • What it is: cash out of the door to MUFG, with collateral marketing allowed from 11-10.

    • Why it's not priced in: no published model shows the amortisation schedule.

    • Trigger to watch: a ≥400 MW lease signed before 11-10, a waiver or amendment 8-K, or the Q3 10-Q debt note.

    • Why this probability: no second tenant is in sight before 11-10, and TensorWave is 222 MW. A waiver is possible (hence not 95%).

  • TensorWave hall delivery slips at least one quarter past Feb 2028 — 55% chance (conditional on the lease closing)

    • Impact on revenue: about −$108M (−9%) vs consensus FY2028 revenue of $1.24B, plus rent credits (cap not public).

    • Impact on EPS: about −$0.07 EBITDA per share in FY2028.

    • How we got the number (estimate): $433M ÷ 4 ≈ $108M of rent lost; $178M ÷ 4 ≈ $44.5M EBITDA ÷ 641M ≈ −$0.07.

    • Expected value: 0.55 × −$0.07 ≈ −$0.04 EBITDA per share.

    • What it is: late power or late halls under the lease's delivery dates [1].

    • Why it's not priced in: management still quotes rent starting late Q3/early Q4 2027 [12], earlier than the lease's own 2028 dates.

    • Trigger to watch: signed final EPC contracts by Q1 2027; first commercial power by ~Feb 2027, as guided [10].

    • Why this probability: every earlier milestone slipped 6–12 months (Bear #4). The SGT-800s need ~15 months after an EPC award that has not happened.

  • A material loss on the $415M reclaimed-units claim — 8% chance within 36 months

    • Impact on revenue: none.

    • Impact on EPS: about −$0.39 per share one-off (cash or equity).

    • How we got the number (estimate): an assumed ~$250M settlement or judgment (midpoint of $100–415M) ÷ 641M ≈ −$0.39.

    • Expected value: 0.08 × −$0.39 ≈ −$0.03 per share.

    • What it is: the consultants win or settle their claim over the 25,000 Class B units [1].

    • Why it's not priced in: nothing has been accrued [1].

    • Trigger to watch: a ruling on dismissal motions; any accrual in the 10-Q.

    • Why this probability: early stage. Fermi calls a loss "reasonably possible" but not probable. Large pleaded claims usually settle far lower.

  • Federal inquiry escalates (charges, Wells notice or enforcement) — 15% chance within 36 months

    • Impact on revenue / EPS: not quantifiable. There is no basis for a penalty size; the main cost is frozen financing.

    • How we got the number (estimate): not quantifiable.

    • Expected value: not quantifiable.

    • What it is: the EDNY subpoena or SEC request leads to action [1,54].

    • Why it's not priced in: the inquiries are disclosed but not quantified.

    • Trigger to watch: 10-Q legal note updates; indictments of former executives.

    • Why this probability: a grand-jury subpoena plus a for-cause firing is a real signal, but most inquiries end without company-level charges.

  • Texas Tech lease renegotiated on materially worse terms, or terminated — 25% chance by end-2027

    • Impact on revenue / EPS: not quantifiable. The economic terms of any renegotiation are unknown.

    • How we got the number (estimate): not quantifiable.

    • Expected value: not quantifiable.

    • What it is: no notice to proceed by 2026-12-31, so Texas Tech "may terminate" [1].

    • Why it's not priced in: the deadline gets little public attention.

    • Trigger to watch: a ground-lease amendment 8-K; whether the $9M escrow is funded by year-end.

    • Why this probability: Texas Tech earns more by extending, but it can extract better terms. The probability rises if TensorWave fails.

Investment Risk Score: 80/100 (very high)

# Factor Weight Measured value Score Why
1 Liquidity & funding runway 15% ~$445M unrestricted pro forma (06-30 + converts net of capped call), before Q3; Q2 op CF −$48.7M, capex −$185M; plan ~$2B in 12 months → ~1–2 quarters; ~4–7 quarters with capex deferred; going-concern "substantial doubt" (alleviated by plans) 88 Under 12 months on the plan, with going-concern language and lender triggers on 11-10 and 12-31.
2 Leverage 10% Debt ~$968M (gross $536.9M secured/equipment + $431.25M converts); EBITDA negative; MUFG 12.2% effective, Keystone 14.1%, CSG/Beal 12% + $37M exit fee 85 Large, expensive debt with negative EBITDA.
3 Dilution (last 12 months + overhang) 10% Shares 614.0M (2025-11-11) → 641.3M (2026-08-31) = +4.4%; overhang: converts 45.3–58.9M (7–9%), unvested RSUs ~17M (2.7%), Yorkville up to 40M if drawn (6%); a large raise likely (Headwind #1) 72 Measured dilution is modest, but the overhang (~16–18%) plus a likely raise of up to +50% put it in the +10–25% band and above.
4 Profitability 10% Pre-revenue; Q2 net loss $25.8M, H1 $214.5M (incl. $118.7M SBC); no assets in service 92 Pre-revenue.
5 Revenue quality & concentration 10% Revenue $0; one signed but not effective lease (TensorWave 222 MW, 100% of contracted revenue); one site 92 No revenue yet, and future revenue rests on one tenant at one asset.
6 Execution & track record 10% Every dated power/tenant milestone since the IPO missed by 6–12+ months; material weakness open; two CFOs in 2026; "coming days" guarantor slipped 9 weeks 88 Serial misses plus a material weakness.
7 Competitive position & disruption 10% 1.5 GW landed turbines + 6 GW permit (scarce) vs peers with investment-grade tenants already delivering (APLD ~$36B base term, IREN 1.4 GW energised) 66 Strong assets, but losing the race for creditworthy tenants.
8 Cyclicality & commodity/macro sensitivity 10% 100% AI-capex dependent; neocloud tenant; AI debt pricing wider (Sept 2026); gas cost passed through 78 Survival depends on AI credit markets staying open in the next 6–12 months.
9 Regulatory, legal & geopolitical 5% EDNY subpoena + SEC request; $415M claim; class action; founder suits; untested Texas incident-of-tenancy exclusion; SPP HILL rules 72 Large pending matters, not existential on current evidence.
10 Valuation & expectations 10% Price ~$3.25 (−85% vs IPO); EV ~$2.58B ≈ 1.7x book PP&E, ~$11.6M per leased (not effective) MW; short interest 66.8M (10.4%) up from 3.3M; analyst targets stale ($6–37) 62 The price assumes little, but EV sits above book with a fragile story and a growing short.

Overall: weighted average = 0.15×88 + 0.10×85 + 0.10×72 + 0.10×92 + 0.10×92 + 0.10×88 + 0.10×66 + 0.10×78 + 0.05×72 + 0.10×62 = 13.2 + 8.5 + 7.2 + 9.2 + 9.2 + 8.8 + 6.6 + 7.8 + 3.6 + 6.2 = 80.3 → 80.

Change vs last: +3 (77 under the old 6-area method). What moved:

  • The 10-factor method scores profitability and revenue quality separately, and both are at the pre-revenue maximum.
  • New facts pushed liquidity and execution up: the MUFG amortisation starts 12-31 at the 10% rate (222 MW < 400 MW), the capped call cut net convertible proceeds to ~$382M, and the guarantor is still unnamed nine weeks after "the coming days".
  • Measured share dilution (+4.4% y/y) is lower than the old "financial health" bucket implied, which keeps factor 3 below the extremes of peers like SHAZ or USAR.

Ranking check: At 80, FRMI is now the riskiest holding in the ledger, above SHAZ and EU (both 73). SHAZ has some revenue and an NVIDIA contract, and EU produces and sells uranium. FRMI has no revenue, a lease that is not yet effective, going-concern language, and a lender amortisation trigger 11 weeks away.

Conclusion

  • What you are really betting on: that within about three months a creditworthy guarantor stands behind TensorWave, project debt follows, and Fermi raises whatever equity it still needs at a price well above today's. If that happens, the landed turbines and permits turn into a ~$433M-a-year lease at peer-beating rates. If not, the turbines and the site are the value, and much of it goes to the lenders.
  • The 2–3 things that will decide the outcome:
    1. Whether the TensorWave lease becomes effective with a named investment-grade guarantor by 10-31, or soon after.
    2. How Fermi handles MUFG (waiver, refinancing or turbine sale) and Texas Tech by 12-31.
    3. The size and price of the equity raise that funds the build.
  • Signposts:
    • Confirm bull:
      • An 8-K naming the guarantor and confirming the lease is effective.
      • Signed final EPC contracts with Primoris and TSK.
      • First SPS grid power or first turbine power, confirmed in a filing.
      • A second lease of ≥178 MW signed (taking the total to 400 MW) or an MUFG waiver.
    • Break bull:
      • Another extension without a named guarantor, or termination.
      • MUFG markets the collateral, or a payment default.
      • An equity raise below ~$3, or with warrants.
      • Texas Tech amends the lease on worse terms, or the escrow is not funded at year-end.
  • New since the last report:
    • The MUFG clock is more precise. Amortisation starts 2026-12-31 (not 11-10) at 10% of the balance per quarter, because TensorWave's 222 MW is below the 400 MW test. Missing a payment is an immediate default, and the loan matures 2027-08-10.
    • The "guarantee rescinded within hours" claim maps to a same-day press-release correction that deleted the guarantor sentence. No guarantee was ever signed.
    • Neugebauer's strategic-review proposal is missing from the 10-02 proxy. The Q3 report is set for 2026-11-12. NAES and CBRE were named operators in September, but there is still no 8-K since 08-31.

Open questions & data gaps

  • The TensorWave lease text: never filed. The outside delivery date, rent-credit cap, prepayments, security deposit and the guaranty's size, term and rating requirement are all unknown. This matters most: it decides what the lease is worth if it closes.
  • Cash at 2026-09-30 and any Yorkville or facility draws since July: only the Q3 10-Q (11-12) will show this. Runway estimates here ignore Q3 flows.
  • MUFG's reading of the 400 MW test (options, Hillcore) and Keystone's definition of an "approved customer agreement": not found. Any waiver would come by 8-K.
  • Why the 14a-8 proposal is absent from the proxy: no SEC no-action letter or company explanation found.
  • Q&A of the Q4-25, Q1-26 and Q2-26 calls: only prepared remarks, a partial third-party transcript and summaries were available.
  • Consensus: aggregators conflict (FY2027 revenue $0.66–2.05B) and most predate the August reset. Tailwind and headwind sizing uses a single-analyst FY2028 figure.
  • Secondary-market turbine prices for the SGT-800, SGT6-5000F, 6B and TM2500: not found, so Tailwind #1 sizing assumes a sale at least at the loan balance.
  • SPS 86 MW: no filing confirms energisation.
  • First tenant's identity: Business Insider reported Amazon citing the founder; Fermi denies it. The article could not be read in full.
Sources
  1. Fermi Inc. Form 10-Q, quarter ended 2026-06-30 — SEC EDGAR — 2026-08-14 — https://www.sec.gov/Archives/edgar/data/2071778/000207177826000051/frmi-20260630.htm
  2. Fermi Inc. Form 10-Q, quarter ended 2026-03-31 — SEC EDGAR — 2026-05-15 — https://www.sec.gov/Archives/edgar/data/2071778/000207177826000032/frmi-20260331.htm
  3. Fermi Inc. Form 10-K FY2025 — SEC EDGAR — 2026-03-30 — https://www.sec.gov/Archives/edgar/data/2071778/000207177826000010/frmi-20251231.htm
  4. Fermi Inc. IPO prospectus (424B4) — SEC EDGAR — 2025-10-01 — https://www.sec.gov/Archives/edgar/data/2071778/000121390025094424/ea0252333-11.htm
  5. MUFG Equipment Supply Loan Financing Agreement (Ex. 10.1 to 8-K) — SEC EDGAR — 2026-02-10 — https://www.sec.gov/Archives/edgar/data/2071778/000121390026014270/ea027627401ex10-1_fermi.htm
  6. 8-K, 5.00% convertible senior notes closing — SEC EDGAR — 2026-07-15 — https://www.sec.gov/Archives/edgar/data/2071778/000121390026078366/ea0298074-8k_fermi.htm
  7. 8-K and Ex. 99.1, offering memorandum excerpts — SEC EDGAR — 2026-07-09 — https://www.sec.gov/Archives/edgar/data/2071778/000121390026076788/ea0297535-8k_fermi.htm
  8. DEF 14A, 2026 annual meeting — SEC EDGAR — 2026-10-02 — https://www.sec.gov/Archives/edgar/data/2071778/000121390026106378/ea0307100-02.htm
  9. PX14A6G, Fermi Founder Parties exempt solicitation — SEC EDGAR — 2026-09-11 — https://www.sec.gov/Archives/edgar/data/2071778/000121390026099255/ea0305276-px14a6g_toby.htm
  10. Q2 2026 earnings release and investor presentation (Ex. 99.1/99.2) — SEC EDGAR — 2026-08-13 — https://www.sec.gov/Archives/edgar/data/2071778/000207177826000046/fermiq22026erex_991.htm
  11. 2Q26 prepared remarks — Fermi IR — 2026-08-13 — https://irp.cdn-website.com/4ee5afac/files/uploaded/Final+2Q26+Prepared+Remarks+8-13-26.pdf
  12. Q2 2026 earnings call transcript — Investing.com — 2026-08-13 — https://www.investing.com/news/transcripts/earnings-call-transcript-fermi-america-q2-2026-shares-slide-after-cash-raise-93CH-4858424
  13. Fermi Q2 2026 earnings call summary — Motley Fool — 2026-08-20 — https://www.fool.com/earnings/call-transcripts/2026/08/20/fermi-frmi-q2-2026-earnings-call-transcript/
  14. 1Q26 prepared remarks — Fermi IR — 2026-05-14 — https://irp.cdn-website.com/4ee5afac/files/uploaded/Fermi+1Q26+Prepared+Remarks+vF.pdf
  15. 1Q26 investor presentation (Ex. 99.2) — SEC EDGAR — 2026-05-14 — https://www.sec.gov/Archives/edgar/data/2071778/000207177826000027/fermiex99_2x1q2026invest.htm
  16. FY2025 shareholder letter (Ex. 99.1) — SEC EDGAR — 2026-03-30 — https://www.sec.gov/Archives/edgar/data/2071778/000207177826000007/frmiex991fy2025shletter.htm
  17. Q3 2025 earnings call transcript — Fermi IR — 2025-11-11 — https://irp.cdn-website.com/4ee5afac/files/uploaded/Fermi+America+Transcript.pdf
  18. 8-K, CEO appointment (McIntire) — SEC EDGAR — 2026-08-14 — https://www.sec.gov/Archives/edgar/data/2071778/000121390026089340/ea0302016-8k_fermi.htm
  19. Fermi and TensorWave Agree to Extend Closing Date to October 31, 2026 — ACCESS Newswire — 2026-09-22 — https://www.accessnewswire.com/newsroom/en/oil-gas-and-energy/fermi-and-tensorwave-agree-to-extend-closing-date-to-october-31-2026-1225550
  20. Fermi Announces Binding Lease Agreement with TensorWave (original version) — ACCESS Newswire via Webull — 2026-08-10 — https://www.webull.com/news/15380623964840960
  21. CORRECTION FROM SOURCE: Fermi Announces Binding Lease Agreement with TensorWave — newswire.com — 2026-08-10 — https://www.newswire.com/view/content/correction-from-source-fermi-announces-binding-lease-agreement-with-tensorwave
  22. Fermi Selects NAES to Operate and Maintain Its Natural Gas Turbine Fleet — ACCESS Newswire via Finviz — 2026-09-28 — https://finviz.com/news/395834/fermi-selects-naes-to-operate-and-maintain-its-natural-gas-turbine-fleet
  23. Fermi Announces Third Quarter 2026 Earnings Release and Call — Stock Titan (ACCESS Newswire) — 2026-10-12 — https://www.stocktitan.net/news/FRMI/fermi-announces-third-quarter-2026-earnings-release-and-call-1bll1rwfqla0.html
  24. Forms 4 and 13D/A (Caddis Holdings, R. Perry, officers) — SEC EDGAR — 2026-03 to 2026-09 — https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=2071778
  25. EDGAR submissions index for CIK 2071778 — SEC — retrieved 2026-10-13 — https://data.sec.gov/submissions/CIK0002071778.json
  26. TensorWave Raises $350 Million Series B at $1.55B Valuation — Business Wire — 2026-06-10 — https://www.businesswire.com/news/home/20260610650010/en/
  27. Exclusive: backstop for Fermi America's 222MW deal with TensorWave still being finalised — The Tech Capital — 2026-08-11 — https://thetechcapital.com/exclusive-backstop-for-fermi-americas-222mw-deal-with-tensorwave-still-being-finalised/ (headline only, paywalled)
  28. Fermi TensorWave lease lacks a guarantor — FinanceFeeds — 2026-09-25 — https://financefeeds.com/fermi-tensorwave-lease-guaranty-october-31/
  29. Fermi: fraudulent transfers, no tenant, AI hype — Fuzzy Panda Research (short seller) — 2026-04-20 — https://fuzzypandaresearch.com/fermi-fraudulent-transfers-no-tenant-ai-hype/
  30. Data center startup Fermi promised nuclear power for AI but couldn't sign a single client — Business Insider via DNYUZ — 2026-05-08 — https://dnyuz.com/2026/05/08/data-center-startup-fermi-promised-nuclear-power-for-ai-but-it-couldnt-sign-a-single-client/ (headline and summary only)
  31. The world's largest data center is running into trouble — Cleanview — 2026-04 — https://cleanview.co/reports/fermi
  32. FRMI analyst forecast and rating history — MarketBeat — 2026-10-12 — https://www.marketbeat.com/stocks/NASDAQ/FRMI/forecast/
  33. FRMI forecast — StockAnalysis — updated 2026-08-14 — https://stockanalysis.com/stocks/frmi/forecast/
  34. FRMI forecast — Tickflow (search summary only) and Tickergate (undated) — ~2026-08 — https://tickflow.io/stock/FRMI/forecast ; https://www.tickergate.com/stocks/frmi/stock-forecast
  35. Evercore ISI cuts Fermi rating; UBS downgrades to Neutral — Investing.com / MarketScreener — 2026-05-15 / 2026-05-05 — https://www.investing.com/news/analyst-ratings/evercore-isi-cuts-fermi-stock-rating-on-project-execution-concerns-93CH-4692156 ; https://www.marketscreener.com/news/ubs-downgrades-fermi-to-neutral-from-buy-adjusts-price-target-to-6-from-8-ce7f58dcd889fe2d
  36. Stifel maintains Buy rating on Fermi — Yahoo Finance — 2026-06-29 — https://finance.yahoo.com/markets/stocks/articles/stifel-maintains-buy-rating-fermi-111257137.html
  37. FRMI rips on $6.5B TensorWave deal (Mizuho comment) — Stocktwits — 2026-08-11 — https://stocktwits.com/news-articles/default/equity/frmi-stock-rips-6-5-b-tensorwave-deal-wall-street-praise-long-awaited-catalyst/cZoZ6fIRJQz
  38. FRMI short interest history (FINRA data) — curvedtrading.com — updated 2026-10-12 — https://curvedtrading.com/stocks/frmi/short-interest/
  39. FRMI quote and analysis — Yahoo Finance — 2026-10-12 — https://finance.yahoo.com/quote/FRMI/analysis/
  40. Applied Digital $7.5 billion hyperscaler lease (Polaris Forge 3) — Pulse 2.0 — 2026-05 — https://pulse2.com/applied-digital-7-5-billion-hyperscaler-lease-pushes-contracted-ai-capacity-above-1-gw/
  41. TeraWulf shares jump after Google secures 8% stake in $3.7B AI compute deal — The Block — 2025 — https://www.theblock.co/amp/post/366959/bitcoin-miner-terawulf-shares-jump-50-after-google-secures-8-stake-as-part-of-3-7-billion-10-year-ai-compute-deal
  42. Cipher Mining signs 168 MW 10-year AI hosting agreement with Fluidstack — Cipher IR — 2025 — https://investors.cipherdigital.com/news-releases/news-release-details/cipher-mining-signs-168-mw-10-year-ai-hosting-agreement/
  43. GE Vernova Q2 2026 results (8-K Ex. 99.1); Data centers drive record surge in GE Vernova orders — SEC EDGAR / Power Engineering — 2026-07-22 — https://www.sec.gov/Archives/edgar/data/0001996810/000199681026000147/gevpressrelease2q26.htm ; https://www.power-eng.com/gas/turbines/data-centers-drive-record-surge-in-ge-vernova-power-equipment-orders-as-turbine-slots-tighten-through-2030/
  44. Siemens Energy posts record backlog — Turbomachinery Magazine — 2026-08-06 — https://www.turbomachinerymag.com/view/siemens-energy-posts-record-backlog-joining-ge-vernova-and-baker-hughes-in-a-record-quarter-for-gas-turbines
  45. Gas turbine prices climb 195% as supply crunch reshapes power development (Wood Mackenzie) — Power Engineering — 2026-04-01 — https://www.power-eng.com/gas/turbines/gas-turbine-prices-climb-195-as-supply-crunch-reshapes-power-development/
  46. DOE announces $17.5B in conditional loans for AP1000 builds — American Nuclear Society — 2026-06-23 — https://www.ans.org/news/2026-06-23/article-8142/doe-announces-175b-in-conditional-loans-for-ap1000-builds/
  47. Project Matador joins EIS pilot program; NRC seeks public input — American Nuclear Society (citing Federal Register 2026-05487) — 2026-03-24 — https://www.ans.org/news/2026-03-24/article-7875/project-matador-joins-eis-pilot-program-nrc-seeks-public-input/
  48. Texas Utilities Code §31.002 — texas.public.law (statute mirror) — current — https://texas.public.law/statutes/tex._util._code_section_31.002
  49. FERC approves SPP's High Impact Large Load study process (Docket ER26-247) — Wright & Greenhill — 2026-01 — https://www.wrightlaw.com/2026/01/the-hills-have-ayes-ferc-unanimously-approves-spps-high-impact-large-load-study-process-and-high-impact-large-load-generation-assessment-process/
  50. Texas Legislature sets its sights on data centers and AI — Foley & Lardner — 2026-04 — https://www.foley.com/insights/publications/2026/04/texas-legislature-sets-its-sights-on-data-centers-and-ai-what-you-need-to-know/ ; SB 6 implementation — Perkins Coie — 2026 — https://perkinscoie.com/insights/update/sb-6-implementation-shaping-data-center-future-texas
  51. Wall Street weighs prospect of an AI slowdown on data center buildout — CNBC — 2026-09-15 — https://www.cnbc.com/2026/09/15/investors-ai-slowdown-data-center-buildout.html (snippet only)
  52. AI data center build advances at full speed: five things to know — BloombergNEF — 2026 — https://about.bnef.com/insights/data-centers/ai-data-center-build-advances-at-full-speed-five-things-to-know/
  53. Texas Business Court denies former CEO's request for emergency relief (Fermi release); Toby Neugebauer sues three Fermi directors (PR Newswire) — Finviz — 2026-09-17 / 2026-09-21 — https://finviz.com/news/393331/texas-business-court-denies-former-ceos-request-for-emergency-relief-against-fermis-board ; https://finviz.com/news/393863/toby-neugebauer-sues-three-fermi-directors-alleging-a-coordinated-effort-to-take-control-of-company-and-disenfranchise-fermi-shareholders-according-to-the-complaint
  54. Fermi says it received US subpoena tied to Project Matador — Bloomberg Government; Bisnow — 2026-08 — https://news.bgov.com/new-york-brief/fermi-says-it-received-us-subpoena-tied-to-project-matador-1 ; https://www.bisnow.com/news/national/data-center-capital-markets/fermi-project-matador-management-records-subpoena
  55. Oracle/OpenAI drop plans to expand flagship Abilene Stargate site — Data Center Dynamics — 2026-03 — https://www.datacenterdynamics.com/en/news/oracleopenai-drop-plans-to-expand-flagship-abilene-stargate-site-meta-in-talks-to-pick-up-crusoe-capacity-with-nvidias-help/
  56. IREN announces successful energization of Sweetwater 1 — IREN IR — 2026-05-01 — https://iren.gcs-web.com/news-releases/news-release-details/iren-announces-successful-energization-sweetwater-1
  57. Fermi signs $6.5 billion lease after months of turmoil (water, council vote) — ABC7 Amarillo — 2026-08 — https://abc7amarillo.com/news/local/fermi-america-signs-six-point-five-billion-dollar-lease-for-worlds-largest-data-center-after-months-of-turmoil-tensorwave-amarillo-texas-carson-county-water-city-council-greg-abbott-audir
  58. Fermi shares plunge after announcing departure of CEO — Bloomberg — 2026-04-18 — https://www.bloomberg.com/news/articles/2026-04-18/fermi-shares-plunge-after-announcing-departure-of-ceo
  59. Fermi stock falls on convertible note, dilution and governance — TIKR — 2026-08-04 — https://www.tikr.com/blog/fermi-stock-falls-convertible-note-dilution-governance-2026
  60. Global data centre capacity tops 25,100 MW as vacancy falls to 4.9% — Datacentres.com — 2026-08-17 — https://www.datacentres.com/news/global-data-centre-capacity-tops-25-100-mw-as-vacancy-falls-to-4-9-amid-280bn-hy-slot1-2026-08-17
  61. Developer behind Amazon's Pecos County data center approaches Brewster County officials — Texas Public Radio — 2026-09-18 — https://www.tpr.org/news/2026-09-18/developer-behind-amazons-pecos-county-data-center-approaches-brewster-county-officials
  62. Neoclouds tenants spark data center financing concerns — Yahoo Finance — 2026 — https://finance.yahoo.com/real-estate/articles/neocloud-tenants-spark-data-center-055013734.html (snippet only)
  63. How tariffs could derail the United States' $3 trillion AI buildout — CSIS — 2025–26 — https://www.csis.org/analysis/how-tariffs-could-derail-united-states-3-trillion-ai-buildout
All FRMI Bulls vs Bears reports
Disclaimer. AI-generated research and podcast (Claude by Anthropic, voices by Google Gemini) for educational purposes — not financial advice and not a recommendation to buy or sell. Willem (MR.INVEST) owns this stock. Check the company’s own filings before making any decision.
← All Bulls vs Bears reports