Fermi Crisis: A Case Study of a Stalling AI Power Stock

Author: Ada, Deep Tide TechFlow

On April 20th, Fermi’s stock price settled at $5.40.

Six months ago, this figure was about $37, and it IPO’d just last October.

A company less than 12 months old, with zero revenue, no tenants, and no marketable products, yet raised $785 million on NASDAQ, with a market cap once reaching $12.5 billion.

However, the CEO and CFO both resigned on the same day, construction halted, insiders sold $68 million worth of stock, short-selling institutions issued accusations of fraud, and class-action lawsuits have been filed.

This marks the first major collapse of the AI power narrative.

“World’s First” in the Texas Wilderness

Fermi’s story began in early 2025.

Former U.S. Energy Secretary Rick Perry and private equity mogul Toby Neugebauer co-founded the company. The core bet was called “Project Matador”: building the world’s largest AI data center park on 5,800 acres outside Amarillo, Texas, initially powered by natural gas, with plans to add four nuclear reactors in the future.

The plan was to provide 11 GW of power capacity and develop about 18 million square feet of data center facilities. The label “world’s largest” was repeatedly emphasized.

The power hunger of AI is real, nuclear energy is green, and Trump signed an executive order to expand U.S. nuclear capacity from 100 GW to 400 GW. All the trends aligned.

The market believed it. On October 1st last year, Fermi went public at $21 per share, opening with a jump to $25, fully oversubscribed. The next day, it hit a high of $37, 76% above the IPO price. Within days, this company with no revenue surpassed a $10 billion market cap.

At that time, everyone was buying into AI power stocks. No customers needed, no revenue required—just a PPT and a vision that others couldn’t see.

The Real Crisis

The first crack appeared in December last year.

Fermi’s only anchor tenant terminated its contract, and the market widely believed this tenant was Amazon. The tenant had promised to prepay up to $150 million in construction funds but paid nothing.

Short-selling firm Fuzzy Panda uncovered the reason. Fermi had promised to raise $5 to $5.5 billion to ensure project execution, but the funds never materialized. The tenant couldn’t wait and simply left.

According to lease terms between Fermi and Texas Tech University, without a signed tenant, Fermi was not even allowed to start construction. This created a vicious cycle: no tenants, no financing; no financing, no construction; no construction, no tenants.

Construction stopped, and workers took to social media saying, “We’ve all been laid off.”

Then came recent major news: CEO Neugebauer and CFO Miles Everson resigned simultaneously. The company framed it as a “Fermi 2.0” strategic transformation. But the stock fell another 22%. Since last year’s IPO, investors who bought FRMI stock have suffered losses of up to 78%.

Insiders had already started selling. When the lock-up period expired on March 30, co-founder Rick Perry’s son Griffin Perry immediately sold 11 million shares, cashing out $56.3 million. The COO, CFO, and Chief Development Officer followed, with insiders selling over $68 million worth of stock.

Fuzzy Panda revealed that before the lock-up, Griffin Perry attempted to sell 30 million shares in a block trade.

This isn’t Neugebauer’s first company to face a crisis.

In 2022, his “anti-woke” bank GloriFi collapsed after burning through investments from conservative financiers like Peter Thiel, Ken Griffin, and Vivek Ramaswamy. Bankruptcy trustees accused Neugebauer in court documents of “securities fraud,” “extreme self-dealing,” and “fraudulent transfers.”

Fuzzy Panda’s report also pointed out that many current Fermi executives were partners with Neugebauer during GloriFi’s time. Chief Site Development Officer Charlie Hamilton was described as Neugebauer’s “longtime friend” in bankruptcy filings. CFO Miles Everson was also accused of participating in unfair transactions involving potential conflicts of interest.

The bankruptcy court ruled that several of Neugebauer’s transactions constituted fraudulent transfers. Despite being previously accused of fraud at his last company, he managed to raise $785 million on NASDAQ with the next. The IPO prospectus acknowledged these lawsuits could distract management, yet investors still bought in. What does this tell us? During bubble periods, people ignore risk disclosures and only care about how sexy the story is.

A Microcosm of the Bubble

Fermi is not an isolated case. It’s a microcosm.

According to Sightline Climate data, as of April 2026, about 140 large data center projects are planned to go live in the U.S. this year, but only one-third are actually under construction. The rest are delayed or canceled.

The bottleneck lies in electrical components.

Transformers, switchgear, and batteries are essential for building data centers. Before 2020, delivery times for high-capacity transformers were 24 to 30 months. Now, wait times can stretch up to five years. For data centers with deployment cycles under 18 months, this is structurally unacceptable. Any delay in component delivery can halt the entire project.

Deeper issues stem from intergenerational mismatches. The U.S. power grid was not designed to handle the loads required by AI. Data centers can be built in three years, but power generation takes longer. Solar or wind power projects take three to six years, gas turbines about six years, and nuclear power over ten years. As The Network World magazine pointed out, small-scale data centers can sometimes overcome this mismatch. But with AI’s scale now so large—power in the hundreds of megawatts per facility—this has become an insurmountable bottleneck.

OpenAI’s flagship project Stargate, which claims to have invested $500 billion, has made no substantial progress as of April.

Partners are fighting over site ownership and system control. The 800 MW expansion of the Texas flagship has been canceled. Stargate projects in the UK and Norway have been paused, and three core executives responsible for Stargate have left for Meta.

Meanwhile, Alphabet, Amazon, Meta, and Microsoft are expected to invest over $650 billion in AI capacity expansion by 2026. Among them, Alphabet alone plans to spend $175–185 billion, equivalent to burning $500 million daily. Yet, the infrastructure needed to support this grand vision cannot develop at the industry’s required pace.

The last time the U.S. experienced a similar energy infrastructure boom was in the late 1990s. The internet bubble and deregulation of the power market sparked a wave of natural gas plant construction, with about $100 billion invested. After the bubble burst, many plants sat idle.

This time, the scale is an order of magnitude larger. U.S. utilities alone have announced plans to spend $1.4 trillion—27% higher than last year’s forecast. Tech companies’ investments in energy-related infrastructure are already twice the entire annual U.S. power industry investment.

But from Q3 to Q4, new data center deals have dropped over 40%. Analysts believe capital expenditure by supercomputing companies may be halved this year.

Money is shrinking, but the story continues. That’s the danger.

Built In’s report summarized: When suppliers heavily invest in startups, and those startups turn around and spend the money on their own products, real demand and manufactured illusions get mixed. When a company’s customer is also its investor, and revenue grows faster than actual usage, it’s a sign a bubble is forming.

When the Bubble Bursts

There are three tiers of players in this food chain.

The first tier is the real winners. Companies that already own operational nuclear plants, like Constellation Energy, don’t need to build anything new. They just transfer existing plant contracts to data centers and profit from AI power. Meta signed a 20-year, 1.1 GW nuclear supply contract with Constellation. Microsoft spent $1.6 billion to restart the Three Mile Island nuclear plant. These are tangible transactions.

The second tier includes various small modular reactor (SMR) startups, like Oklo. Their stock prices are soaring, but no reactors have been built. Historically, U.S. nuclear projects are known for delays and cost overruns, with few completing on time and within budget over the past decades. But investors don’t care.

The third tier is companies like Fermi—no reactors, no gas plants, no tenants. They’re at the bottom of the food chain, selling stories, not electricity. When the story collapses, there’s nothing left.

Fermi’s collapse won’t be an isolated event.

When an industry’s actual delivery capability lags far behind its PPT promises, a collapse is only a matter of time.

In the U.S., only 6.3 GW of the planned data center capacity for 2027 is actually under construction, out of a total announced 21.5 GW. A 15 GW gap exists on paper, backed by hundreds of billions of dollars and countless unfulfilled promises.

Who will be the next Fermi? Nobody knows. But in this sector, $500 billion is looking for power, high-capacity transformers are waiting for delivery, and many startups without even basic grid access are assuring investors that everything is under control.

The last energy infrastructure bubble burst, some power plants remained. This time, many projects might not even dig a foundation.

And the 5,800 acres of land in Texas where Fermi sat will, along with all the unfulfilled grand narratives, slowly be buried by time.