Bloomberg reported on May 1, 2026 that Fermi, the nuclear-powered AI data center startup with a Texas project, has not signed a single client while its former CEO contests the company's direction, offering the clearest evidence yet that the AI infrastructure gold rush is producing winners and casualties in unequal measure.
There is a specific kind of startup failure that is harder to diagnose than the obvious kind. A company that runs out of money, misses its product deadlines, or loses its market to a faster competitor is easy to understand in retrospect. Fermi represents something more instructive: a company that identified a real problem, positioned itself against genuine demand, raised what appears to be sufficient capital to pursue the opportunity, and still cannot get the buyers it needs to sign on the line. That failure mode is worth examining in detail, because the conditions that produced it are not unique to Fermi.
The facts reported by Bloomberg are stark. Fermi has been pitching hyperscale AI operators on a Texas data center site backed by nuclear power capacity. The market those operators represent is actively spending on exactly this kind of infrastructure. And Fermi has nothing to show for its sales effort in the form of a signed agreement. No client, no named pilot, no preliminary commitment of any kind. Simultaneously, the company's former CEO is now publicly contesting its future, which transforms what was a pure commercial problem into a governance crisis visible to every procurement team Fermi might approach going forward.
The AI power agreement market has produced real transactions over the past eighteen months. Microsoft's deal with Constellation Energy to restart Three Mile Island's Unit 1 reactor is the most cited example, but Google, Amazon, and others have signed agreements with nuclear developers, renewable energy providers, and independent power producers at significant scale. Studying what those agreements have in common reveals the criteria Fermi cannot currently meet.
Every successful deal involved a power source that either already existed or had a licensed, permitted, near-term path to operation. Constellation was not offering Microsoft a future nuclear plant built from scratch by a startup. It was offering the restart of a reactor with decades of operational history, managed by a company with deep NRC relationships and a proven nuclear operations workforce. The risk profile of that commitment, while still substantial, was bounded in ways that a startup's pre-development nuclear project simply cannot match.
Second, every successful deal involved an organization whose leadership and strategic direction were not in question. Multi-billion dollar infrastructure commitments require a counterparty that procurement teams, legal departments, and board-level executives at the buyer are prepared to stand behind internally. Recommending a vendor whose former CEO is fighting publicly over the company's direction is a career risk for everyone in that chain, and procurement professionals understand that clearly even when they do not say it explicitly.
Third, the deals that closed involved sellers with either existing customer relationships at the enterprise level or demonstrable sales execution experience in long-cycle infrastructure markets. Closing a hyperscaler data center agreement is not a startup sales motion. It involves months of technical evaluation, legal review, regulatory risk assessment, and executive alignment across organizations with complex internal governance structures of their own. The companies that have navigated that process successfully were not doing it for the first time.
The harder question for investors watching this unfold
Fermi raised capital. That capital came from investors who evaluated the opportunity and concluded it was worth backing. The thesis those investors accepted, that nuclear power access plus AI demand plus Texas land equals commercial traction, was not unreasonable on its face. It was incomplete in ways that have become apparent only as the company has tried to convert thesis into revenue.
The incompleteness is instructive for evaluating other companies in the AI infrastructure space that have raised on similar logic. The questions that Fermi's situation suggests every infrastructure investment should answer before a term sheet is signed are practical ones: Has the company demonstrated the ability to navigate the specific sales cycle required to close its target customers? Does the organization have leadership stability and strategic alignment that will survive the multi-year development timeline between fundraise and revenue? Is the underlying asset, whether power, land, or infrastructure, deliverable on a timeline that matches buyer needs rather than seller optimism?
Fermi's Texas project may eventually find a path forward. Leadership disputes can be resolved, sales processes can be restarted, and a first client commitment can change the commercial trajectory of a company quickly if it arrives with the right credentials attached. But none of that is guaranteed, and the current state of the company, no clients, contested leadership, and a story now defined by its difficulties rather than its promise, is a harder position to recover from than it might appear from the outside.
The market's attention tends to follow success stories in AI infrastructure, the announced deals, the capacity commitments, the valuation milestones. Fermi is a reminder that for every announced deal there are companies that never got there, and that the reasons they did not are worth understanding as carefully as the reasons the winners did. The AI power buildout is real. The demand is genuine. The customers are spending. Not every company positioned against that spending will capture any of it, and the criteria that separate the ones that do from the ones that do not are considerably more demanding than the fundraising environment of the past two years has made them appear.
Also read: Fermi's nuclear data center had the right pitch at the right time and still cannot find a customer • The World Is Going Digital Faster and Some People Are Deliberately Going the Other Way • MoonPay launches a stablecoin Mastercard that lets AI agents spend crypto on behalf of their users
