On November 12, Fermi Inc. — operating as Fermi America — reported third-quarter 2026 results and announced what looked like validation for its Project Matador campus: an undisclosed investment-grade hyperscaler has agreed to prepay about $200 million to advance an initial 600 megawatts of power capacity at the private-grid AI campus near Amarillo, Texas, beginning in 2027. The headline wrote itself. The filing told a more careful story.

The careful version is this: the money is real, the lease is not. The 600-megawatt arrangement rests on a nonbinding letter of intent dated November 10, not a signed long-term lease. The same release disclosed that Fermi’s first binding customer deal — the August 2026 lease with TensorWave — has been terminated, along with the AMD credit support agreement tied to it. And the same day’s quarterly filing disclosed that the executive in charge of building Fermi’s nuclear reactors has left, and the reactor schedule has slipped again. The headline says nuclear. The megawatts say gas and grid.

Three agreements, three different kinds of commitment

The announcement bundles three items that are not the same kind of promise. The first is a $200 million prepayment agreement: cash paid in advance toward developing the initial 600 megawatts. The second is a $15 million early construction and site development agreement. The third is a contemplated purchase of about $90 million of structural steel and building materials — still being finalized, not signed.

These should not be added together into a $305 million commitment. The $200 million and the $15 million are agreements; the $90 million is a negotiation. And none of the three is a lease. A prepayment buys development work and, in effect, an option on capacity. A lease commits a tenant to pay for capacity over a fixed term. Fermi says the parties continue to work toward a definitive long-term lease, and it has not given a signing date. Management said on the earnings call that the $200 million wire is due November 13 and that construction mobilization is targeted by December 1. None of that is a signature.

The refund terms have not been disclosed, which is the question that decides what the prepayment is actually worth if the lease never materializes. A nonrefundable prepayment is a meaningful transfer of risk to the customer. A refundable one is a deposit. Fermi has not said which it is, and the customer has not been named.

The tenant that left

The termination matters because it is the second act of a story Fermi told only three months ago. In August 2026, Fermi signed its first binding lease, with TensorWave, a neocloud operator that deploys AMD graphics processors: 222 megawatts of facility power, with first delivery expected in the second half of 2027, and a deal reported at about $6.5 billion over its term. AMD provided credit support. That lease has now been terminated, and the AMD agreement with it.

Fermi has not disclosed a reason. What can be said is what the sequence shows: the campus’s first anchor tenant arrived in August and was gone by November, and the replacement is a customer that has put down money but has not signed a lease. The 600 megawatts in the new arrangement is larger than the 222 megawatts in the old one, and the customer is, by Fermi’s description, larger too. But the old deal was binding and the new one is not, and that is a downgrade in legal status even as it is an upgrade in scale. The context is a pre-revenue company: Fermi lost $141.6 million in the quarter, or $0.22 per share, and held $315.5 million in cash and restricted cash at September 30 against net debt of $958.4 million. For a developer carrying nearly a billion dollars of net debt and no revenue, the difference between a signed lease and a letter of intent is not a technicality. It is the difference between contracted cash flow and a hope.

The reactors slip again

The nuclear half of the Matador story moved in the opposite direction from the commercial half. Fermi disclosed in its quarterly filing that Mesut Uzman, its chief nuclear construction officer, departed on October 12. Uzman was hired last year to lead the nuclear program on the strength of his work delivering reactors in China and the United Arab Emirates. Weeks after he left, Fermi’s chief operating officer signed a letter pushing back the nuclear licensing schedule.

The schedule itself has now slipped twice in a year. Fermi plans four Westinghouse AP1000 reactors at Matador. Its original plan called for the first unit in 2032. Its previous quarterly filing targeted commissioning one reactor a year from 2033 through 2036. The new filing moves that to 2034 through 2037. The licensing paperwork has slipped with it: Fermi told the NRC it would file its final siting chapter by December 31, 2026; an October 29 letter pushed that to October 31, 2027, ten months late.

Every revision moves the reactors further from the present, while the customer commitment Fermi just announced starts in 2027. A hyperscaler that needs power in 2027 is not waiting for a reactor that now commissions in 2034 at the earliest. The nuclear layer of Matador is a 2030s story. The 600 megawatts is a 2027 story.

What “permitted power” actually means

Fermi’s chief executive, Lee McIntire, has described the campus’s advantage as “permitted power,” and the phrase deserves unpacking. Project Matador holds an air permit from the Texas Commission on Environmental Quality covering about 6 gigawatts of on-site gas generation — 93 gas units approved in February 2026 — with an application for about 5 gigawatts more filed in March and still undecided. What Fermi holds, in concrete terms, is permission to burn gas at scale on its own land.

That permission is scarce in exactly the way this desk has been documenting all year. Gas turbines carry lead times measured in years, and the queue for them is crowded with developers, utilities, and power producers ordering at once. A campus that already holds turbines and a 6-gigawatt permit has something a competitor cannot build quickly. Fermi says it owns 2,000 megawatts of turbine capacity, including three Siemens SGT6-5000F gas turbines — rated up to 780 megawatts in simple-cycle mode — that arrived at the Port of Houston in July. Six Siemens SGT-800 turbines are on order for 2028 delivery. The company cites 220,000 MMBtu per day of firm gas supply and five major pipelines within 20 miles of the site.

The grid contributes the rest of the near-term picture. Fermi’s investor presentation counts 200 megawatts from Xcel Energy’s Southwestern Public Service Company toward its 2.2 gigawatts of current capacity, delivered over a 115-kilovolt line under an electric service agreement signed in December 2025. The private grid is the headline; the utility connection is the bridge that lets the first megawatts arrive before the turbines do.

The 2.2-gigawatt arithmetic

Fermi describes 2.2 gigawatts as its current capacity, and the composition of that number is worth reading closely. It includes the 200 megawatts from Xcel and the six SGT-800 turbines on order for 2028 — capacity that is contracted or ordered, not capacity that is operating. As of September 30, none of Fermi’s own equipment had been placed in service. The chief commercial officer, Anna Bofa, targeted initial power availability for the 600-megawatt development in late second quarter or early third quarter of 2027.

The long-term plan is larger still: about 17 gigawatts at full build-out, on roughly 8,400 acres in Carson County, about 12 miles northeast of Amarillo. Fermi’s own presentation says anything beyond about 6 gigawatts depends on further permits, financing, interconnection capacity, and land. The 17-gigawatt figure is an ambition with conditions attached, not a permitted project. The 6-gigawatt permit is the permitted project. The 600 megawatts is the piece with a customer attached, and that customer has not signed a lease.

The December 31 clock

There is a deadline running underneath all of this, and it arrives in six weeks. Fermi’s quarterly filing ties two obligations to December 31, 2026. Under the Keystone Master Loan Agreement, a facility requires mandatory prepayment if an approved customer agreement has not been received by that date. Separately, the Texas Tech University System, Fermi’s landlord, must issue a notice to proceed by December 31, conditioned in part on a lease with a Phase 1 tenant of at least 200 megawatts — and it may terminate the land lease if it does not.

An amendment to Fermi’s MUFG equipment loan eliminated the covenant requiring a lease or offtake arrangement for 400 megawatts by November 10, 2026. The amendment fees totaled $25 million.

The Texas permit pause, and why it may not reach Matador

One more regulatory wrinkle cuts in Fermi’s favor. On September 21, Governor Greg Abbott directed the Texas Commission on Environmental Quality to pause issuance of all permits related to data center projects until the Electric Reliability Council of Texas completes its review, with the ERCOT audit expected in mid-December. But the pause does not directly address permits already issued, and Fermi’s 6-gigawatt air permit was issued in February. Whether it touches Fermi’s pending application for additional capacity has not been confirmed.

There is also a jurisdictional subtlety. Matador’s utility service comes from Xcel’s Southwestern Public Service Company, not from ERCOT. The campus sits in the Panhandle, outside the ERCOT footprint that the governor’s order and the Senate Bill 6 large-load rules — a 75-megawatt threshold for new interconnection requirements — were written for. That is an advantage for Fermi, and a reminder of how much of this story turns on which regulator’s map you are standing on.

Why a hyperscaler prepays without signing

The natural question is why a company large enough to be called investment-grade would wire $200 million to a pre-revenue developer without a signed lease. The answer is probably the same arithmetic that has driven this desk’s coverage all year: time. Connecting a large data center to the grid takes three to seven years when major upgrades are needed. A campus with a 6-gigawatt air permit in hand, turbines on order, firm gas supply, and a 200-megawatt utility connection can plausibly deliver power in late 2027. For a hyperscaler racing to deploy AI infrastructure, a year of waiting is worth more than the legal comfort of a signed lease.

There is also the turbine market to consider. GE Vernova’s chief executive has said the company’s gas turbine capacity is mostly sold out through 2030; Siemens Energy’s chief has described delivery dates stretching to 2029 and 2030. A developer that already holds turbines and permits is selling scarcity.

What the $200 million buys, and what it does not

So what does the prepayment buy? In the near term, mobilization: construction crews on site by December 1, steel ordered, turbines installed — the physical work that turns a permitted site into a powered one. It buys Fermi time with its lenders and its landlord, and it buys the hyperscaler a reserved position at a campus whose power is permitted and whose turbines are already spoken for. It is the bring-your-own-power story in its most advanced form — a private utility built to lease to tenants who cannot wait for the public grid — and the model only works if the tenants sign.

What it does not buy is a tenant, a reactor schedule, or a nuclear construction chief. It does not buy a lease, a rent roll, or revenue. And it does not resolve the December 31 deadlines: a nonbinding letter of intent is not, on its face, the approved customer agreement the Keystone facility requires, nor the Phase 1 lease Texas Tech’s notice to proceed is conditioned on. Whether the prepayment agreements satisfy those tests is a question for Fermi’s lenders and landlord, not for a press release.

What to watch

Four things will settle whether this announcement was a turning point or a bridge payment. The first is the wire itself: whether the $200 million arrives on November 13 as management said, and on what terms — refundable or not. The second is the lease: whether a definitive long-term lease is signed before the December 31 deadlines with the Keystone lenders and Texas Tech. The third is the December 31 clock itself: a waiver, an extension, a notice to proceed, or a mandatory prepayment will each tell a different story about how much room Fermi actually has.

The fourth is the power. Fermi targets initial availability in late second quarter or early third quarter of 2027. Whether the turbines, the gas supply, and the Xcel connection add up to 600 megawatts of delivered power on that schedule is verifiable once the date arrives. The reactors, meanwhile, will be watched on a different clock — the NRC docket, where the siting chapter is now due October 31, 2027, and the first commissioning is penciled in for 2034. The headline said nuclear. The megawatts say gas and grid. The next filing will say which one the customer actually paid for.