AI Goes Nuclear: Why Software Companies Became Nuclear's Customer, and Where the Value Sits

Three things are true right now, and they do not seem to fit together.

First, Microsoft has signed a 20-year contract to restart Three Mile Island, with commercial operation targeted for the second half of 2027. Second, five test reactors went critical in the United States in nine weeks this summer, the first new American reactors in a generation. Third, nuclear developer stocks fell hard. Oklo is down 79% from its high, and NuScale is down 76% over a year.

The largest software companies in the world are buying nuclear power on 20-year contracts. America is building reactors again. And the market is selling. Those cannot all describe the same industry.

This piece examines five ideas that resolve the contradiction, the risks the market is right to price, and how we are investing behind the shift.

Both obvious readings are wrong.

The old consensus was not stupid. For about two decades, “nuclear is dead” was the safest call in energy, and it was correct. The United States completed two reactors in roughly 30 years. Vogtle Units 3 and 4 came in about seven years late and $17 billion over budget.

Nobody changed that answer. Somebody changed the question. What changed is the customer.

The new customer is a software company that needs firm, around-the-clock, carbon-free power at gigawatt scale, and will sign a 20-year contract to get it. Nuclear has never had that buyer. It has had regulated utilities optimizing for the cheapest possible electron, and those utilities looked at a bespoke, decade-long megaproject and reasonably said no.

A hyperscaler is not optimizing for the cheapest electron. It is securing a scarce input before a competitor does. That is a different calculation, and it explains much of what follows.

Three forces sit behind all of it.

AI’s appetite for compute and power: This is the demand shock, and the one most people have heard about. National security as industrial policy: Fuel, supply chain, and now the Army as an actual customer. It gets less attention and may prove more durable, because it does not depend on the price of electricity. Government as investor: Loans, equity stakes, and milestone-gated programs. This is why a category with almost no revenue is getting built anyway.

Those forces show up in five specific ideas.

1. AI ended 20 years of flat demand in about 18 months.

For about 20 years, American electricity demand went nowhere. LED bulbs and better appliances, along with heavy industry moving offshore, canceled out growth almost exactly. Utilities planned on flat demand and built accordingly.

That is over. The analysts who forecast national power demand for a living now expect 166 gigawatts of growth by 2030, up from a forecast of 24 gigawatts four years ago (Grid Strategies). They were not wrong about the physics. They were wrong about who would show up and ask for power.

Those buyers went straight to nuclear.

Nuclear fits because these buyers need firm, always-on, carbon-free power in one location, they hold public commitments they do not want to walk back, and they can pay. They are among the most creditworthy counterparties on earth, signing 20-year contracts. That is the balance sheet nuclear has always needed and never had.

2. Five reactors went critical, and the engine turned over.

Five reactors went critical in nine weeks this summer: Antares on June 4, Valar on June 18, Deployable Energy on June 30, Aalo on July 4, and Oklo Isotopes on August 5.

That is remarkable, and narrower than it sounds. These were zero-power test articles, and none is licensed by the Nuclear Regulatory Commission (NRC). Idaho National Laboratory’s own analogy fits: the difference between an engine turning over and a car driving down the highway. The engine turned over five times in nine weeks.

One exception matters. Valar’s Ward 250 produced electricity on July 1 and ran an NVIDIA chip with it. It is small, but it is the only demonstrated fission-to-electricity event in the group, and it captures the thesis in miniature: a reactor as a product, powering compute. Valar and Aalo are both AV portfolio companies.

The milestone that may matter more in five years is less dramatic. TerraPower received an NRC construction permit on March 9, the first commercial reactor construction approval in nearly a decade and the first for a non-light-water design in more than 40 years.

3. The NRC used to get paid more for being slower.

The NRC was built after Three Mile Island to make sure nothing bad ever happened. It succeeded, partly by making sure almost nothing happened at all. It is also funded by hourly fees on applicants, so a slower review was a more profitable review. Nothing in its mandate required it to care whether the industry it regulated survived.

Congress rewrote the job description. The ADVANCE Act added enabling deployment to the mission, and executive orders put hard deadlines and fee caps behind it. The results are visible.

4. The Army became nuclear’s first customer.

In August, the Army stood up the Janus Program: up to $2.2 billion across 2027 to 2031 for five microreactor developers at five installations.

Unlike a hyperscaler memorandum of understanding, this is appropriated federal money with milestone gates and a hard date behind it.

The Army’s program manager, Jeff Waksman, was unusually candid about the goal. He described success as helping multiple nuclear companies build reliable, affordable microreactors they can sell to buyers beyond the military. That is industrial policy stated out loud. The program falls short if these companies sell only to the Army.

The pattern is familiar. GPS had a military first customer. So did jet engines, semiconductors, and the internet. In each case, the military paid for the expensive, unreliable early version, and the commercial market inherited the cheaper, better one.

The hardest unit to build is the second one. The first has a research budget behind it. The second has to be paid for by someone who wants power. A customer who is insensitive to price and tolerant of first-of-a-kind risk is what gets a manufacturing line to unit two.

5. We gave away enrichment, and Russia picked it up.

A reactor is only as deployable as its fuel. Most new designs need HALEU, uranium enriched to between 5% and 20%, rather than the under 5% that today’s fleet uses.

The Department of Energy estimates demand at 50 metric tons of HALEU per year by 2035. The announced commercial build-out is 12 tons a year, with first new capacity in 2029. That is a fourfold gap, arriving six years late.

We got here by giving up commercial enrichment on the theory that global markets would always supply it. The U.S. share of global enrichment fell from 64% in 1985 to under 1% in 2024. Russia’s share rose to 43%.

Then comes the hard date. Every DOE waiver that lets American utilities keep buying Russian enrichment terminates on Jan. 1, 2028. Not phases out. Terminates. The market is already repricing it. The long-term uranium price now sits above spot, $96.50 against $89.68 a pound. When long-term goes above spot, utilities have stopped shopping and started locking in supply.

The market is skeptical, and it is pricing the right gap.

The public market has delivered a verdict, and the data shows what it is paying for.

The market is not saying nuclear is dead. It is paying for fuel and for existing megawatts, and it is not paying for reactor designs with no revenue. That is close to what we would have said ourselves.

It also resolves the puzzle at the top of this piece. Physics risk got retired this summer. What did not get retired is the ability to turn a test article into a licensed, financed, built, connected plant that sells electricity. The public market is pricing that gap honestly.

Two risks do not show up on the stock chart.

Our conclusion is conviction on the change in customer and discipline on which layer you own. The companies that concern us most are the ones whose value depends on a single design winning. The ones that concern us least make something everyone needs regardless of which design wins.

Assembly-line nuclear is the frame.

The old model was a cathedral: bespoke design, built on site, on decade timelines, with no learning curve. Every one was different, and the workforce dispersed in between. Cathedrals get more expensive each time you build one.

The new model is a product: factory-made, shippable, and repeatable, with a cost curve that improves as volume rises. Manufacturing gets cheaper with repetition. Construction does not.

If that is right, the value does not sit with whoever owns the power plant. It migrates to whoever owns the factory, and to the fuel and supply chain that feed it. Nobody has yet proven a nuclear learning curve. That is the open question in this category.

Fusion is a call option, not a thesis.

The money in fusion is real. The industry raised $4.48 billion in the 12 months to July, up 69%, bringing cumulative private funding to $14.24 billion across 56 companies.

The milestones are more limited. No company has produced net electricity, and no private device has exceeded scientific breakeven. Commonwealth Fusion’s SPARC is about 80% complete and has not achieved first plasma. Its Google power contract is contingent on reaching breakeven, no earlier than 2027. Helion achieved measurable D-T fusion in February and has a contract to deliver 50 megawatts to Microsoft in 2028. Pacific Fusion, an AV portfolio company, broke ground on a $1 billion Albuquerque campus targeting net facility gain by 2030, a harder metric because it counts the whole building’s energy rather than only the energy on target.

2025 and 2026 were a financing and permitting breakthrough for fusion. They were not a physics breakthrough. We treat it as a call option inside a portfolio, not a thesis to build a portfolio around.

Why this is a private-markets story.

The case for investing privately is not that public investors are wrong. It is about where in a company’s life the value gets created, and in this category those moments have names and dates.

They are the round before TerraPower’s construction permit, the round before a Janus award, and the round before a fuel-supply offtake gets signed. Each is a step change in a company’s risk profile, and each happens years before anything is public.

X-energy is the cleanest illustration. Private rounds compounded for years while the DOE task order and NRC filings were happening. The public stock arrived only after most of that milestone value had been captured.

The next set of milestones is already on the calendar: 12 metric tons a year of new HALEU capacity starting in 2029, upcoming NRC operating license decisions, and Janus milestone gates running through 2031. These are dated events, and you can underwrite against them.

The constraint is access, not capital. These rounds are oversubscribed and largely invitation-only.

We hold four groups rather than one large position because we can describe the direction of this shift with real confidence, and we cannot say which single reactor design wins. The portfolio spreads across the parts of the thesis that pay off on different timelines and for different reasons.

The open question is the learning curve.

A test article becoming a licensed, financed, built plant is a long road, and nuclear has not yet shown that cost falls with repetition. Regulatory speed could reverse. Demand forecasts have moved by more than half in a month. And the public market is already discounting every design that has not yet earned revenue.

That is why we favor the layers that matter regardless of which design wins, and why we underwrite against dated milestones rather than announcements.

The customer changed first. The market is still catching up.

Nuclear spent two decades without a buyer who needed what it makes and could pay for it. It now has one. The technology is progressing on a timeline that matters, and the physics risk has largely been retired. The regulatory and manufacturing questions remain.

The U.S. Strategic Tech Fund’s view is straightforward. Back the companies built for the new customer, stay disciplined about which layer you own, and pay attention to access. Whether nuclear can become a product rather than a cathedral is the question worth watching.

This communication is from Alumni Ventures, a for-profit venture capital company that is not affiliated with or endorsed by any school. It is not personalized advice, and AV only provides advice to its client funds. This communication is neither an offer to sell, nor a solicitation of an offer to purchase, any security. Such offers are made only pursuant to the formal offering documents for the fund(s) concerned, and describe significant risks and other material information that should be carefully considered before investing. For additional information, please see here. Achievement of investment objectives, including any amount of investment return, cannot be guaranteed. Co-investors are shown for illustrative purposes only, do not reflect all organizations with which AV co-invests, and do not necessarily indicate future co-investors. Example portfolio companies shown are not available to future investors, except potentially in the case of follow-on investments. Venture capital investing involves substantial risk, including risk of loss of all capital invested. Diversification cannot prevent investment loss; it is a strategy to mitigate investment risk. This communication includes forward-looking statements, generally consisting of any statement pertaining to any issue other than historical fact, including without limitation predictions, financial projections, the anticipated results of the execution of any plan or strategy, the expectation or belief of the speaker, or other events or circumstances to exist in the future. Forward-looking statements are not representations of actual fact, depend on certain assumptions that may not be realized, and are not guaranteed to occur. Any forward-looking statements included in this communication speak only as of the date of the communication. AV and its affiliates disclaim any obligation to update, amend, or alter such forward-looking statements, whether due to subsequent events, new information, or otherwise.