Silicon Valley has a new obsession, and it involves split atoms. For years, the tech elite focused on software, apps, and crypto. Now, the grandest ambition in tech circles is building nuclear reactors.
Walk into a specific factory in Austin, Texas, and you will find Matt Loszak trying to reinvent an industry. The 35-year-old founder of Aalo Atomics is assembling components for the Aalo-X, a tiny reactor designed to sit right next to massive artificial intelligence data centers. His stated goal sounds like something out of an iPhone assembly plan, shipping hundreds or thousands of these reactors every year.
It is a striking scene. Young tech founders, backed by billions in venture capital, are adopting the mantle of the nuclear bro. They talk about energy abundance with the same casual certainty that previous founders used when talking about disrupting local taxi fleets. They want to move fast. They want to scale.
They are running straight into a wall of regulatory reality, supply chain bottlenecks, and physics.
Nuclear engineering is not software. You cannot push a buggy beta version to production and patch it over the weekend. When software crashes, users refresh their browsers. When a nuclear project fails, companies go bankrupt and communities face generational hazards. The collision between Silicon Valley bravado and the rigid realities of atomic energy is creating the most expensive experiment in modern industrial history.
The Insatiable Hunger of the AI Beast
We need to talk about why this is happening right now. Artificial intelligence is an energy vampire. A simple Google search takes a negligible amount of electricity, but generating an AI response using large language models requires exponentially more power. Every time a user asks an AI to write an essay or generate an image, server farms hum with heat.
The numbers are terrifying for grid operators. Data from Goldman Sachs Research indicates that US data center power demand sat at roughly 31 gigawatts in 2025. By the end of this year, that number is climbing to 41 gigawatts. By 2027, data center power demand will nearly double again to 66 gigawatts.
To put that in perspective, a single advanced data center campus can consume as much electricity as hundreds of thousands of traditional homes. The massive tech hyperscalers like Microsoft, Meta, Amazon, and Alphabet are building data centers faster than utility companies can string up power lines.
The traditional energy grid cannot keep up. Solar and wind are clean, but they are intermittent. The sun sets, and the wind stops blowing. AI data centers must run 24 hours a day, 365 days a year, with zero downtime. If the power drops for even a fraction of a second, billions of dollars in computational work evaporates. Tech companies tried turning back to natural gas, but that ruins their public climate commitments.
That leaves one option that offers continuous, carbon-free baseload power. Nuclear energy.
The Hyperscaler Checkbooks Are Open
The shift from theoretical interest to cold hard cash happened quickly. Over the past 18 months, early-stage nuclear companies raised more than $7 billion to design and build small modular reactors. The big tech giants are signing massive agreements to secure their energy sovereignty.
Take a look at Microsoft. They signed a 20-year power contract with Constellation Energy to revive the shuttered Unit 1 reactor at Three Mile Island. Yes, that Three Mile Island, the site of America’s most famous nuclear meltdown in 1979, though that happened at Unit 2. Microsoft is paying an estimated $16 billion over two years to buy 100 percent of the revived plant's output starting in 2027. The facility is being renamed the Crane Clean Energy Center.
Other tech giants are placing bets across the board.
- Meta has backed a multi-gigawatt commitment combining agreements with Vistra, Oklo, and TerraPower.
- Amazon partnered with X-energy to develop up to a dozen small modular reactors.
- Google signed a fleet deal with Kairos Power to buy electricity from a series of small reactors.
The tech companies are not just buying power from old utilities anymore. They are acting as venture capitalists for unproven nuclear startups.
The Move Fast and Break Things Problem
Traditional nuclear engineers are watching this influx of tech capital with a mix of gratitude and absolute terror. The tech industry thrives on an ethos of rapid iteration. If code has a bug, you deploy a fix. If a prototype rocket explodes on a launchpad in South Texas, you celebrate the data gathered and build the next one.
Apply that logic to nuclear material, and you get a disaster.
The traditional sector is defined by extreme caution, heavy regulatory oversight, and redundant safety systems. The US Nuclear Regulatory Commission is notoriously slow, often taking up to a decade and hundreds of millions of dollars just to review and approve a single reactor design.
Tech entrepreneurs view this bureaucratic process as an existential threat to American innovation. They point out that China and Russia are moving far ahead in building small modular reactors. In response, the political climate has shifted. The White House has set aggressive timelines to revive atomic energy, eager to mark industrial milestones for America's global standing.
Yet, engineering veterans point out a fundamental flaw. You cannot scale a nuclear factory the way you scale a gigafactory for electric cars. The components require specialized metallurgical standards, precise welding, and specialized concrete that only a handful of suppliers globally can provide. A single microscopic flaw in a reactor vessel can kill a project permanently.
The Pre Revenue Reality of Nuclear Startups
Wall Street has caught the nuclear bug, driving up the stock prices of any company with the word nuclear or reactor in its pitch deck. Publicly listed small modular reactor developers like Oklo and NuScale Power have become retail investor favorites.
If you look closely at the financial filings, the gap between expectations and reality becomes painfully clear.
Oklo, which is backed by OpenAI founder Sam Altman, posted zero revenue for the full year of 2025. Instead, it registered an operating loss of $139.3 million. NuScale Power generated a modest $31.5 million in revenue in 2025, down from the previous year, while recording a net loss of $355.8 million.
These startups are pre-revenue or near-pre-revenue. They are essentially research and development labs trading at massive valuations based on power purchase agreements for reactors that do not yet exist. They are selling promises of future electrons to power AI models that are being built today.
Even the suppliers of raw materials are feeling the crunch. Cameco, one of the largest uranium producers globally, has locked up long-term contracts to deliver millions of pounds of uranium through 2030, but the supply chain remains incredibly tight. The specialized fuel needed for many of these new reactor designs, high-assay low-enriched uranium, was historically dominated by Russian state suppliers. Building a domestic supply chain for this specialized fuel will take years.
The Physical Constraints of the Real World
Building a software product requires laptops, servers, and talent. Building a nuclear ecosystem requires heavy manufacturing infrastructure that the United States largely abandoned decades ago.
Consider the supply bottleneck for components like superconducting magnets used in fusion research, or specialized valves for fission reactors. Companies like Fujikura in Japan are spending millions to triple their production capacity for superconducting materials, expecting a massive boom by 2027. Yet, the entire global supply chain relies on a tiny handful of manufacturers in Asia and Europe.
If a startup lacks a reliable supply of reactor-grade forgings, its timeline slips. When timelines slip in the hardware world, burn rates destroy capital quickly. Unlike software startups that can lay off half their staff and survive on a skeleton crew, a nuclear startup must keep paying highly specialized nuclear physicists, regulatory lawyers, and quality control engineers just to stay compliant with federal law.
Practical Next Steps for Navigating the Clean Energy Boom
If you are trying to understand where this energy shift is going, you need to look past the flashy press releases of tech founders standing in front of shiny reactor mockups. Focus on the structural realities of the energy grid instead.
Watch the layer one energy providers
The companies actually generating revenue today are the traditional utility giants with existing nuclear fleets. Enterprises like Constellation Energy and Vistra Corp are the immediate winners of the AI power boom. They own the operating licenses, the cooling towers, and the grid connections. They can sell existing power at a premium to tech buyers while the startups spend years fighting for regulatory approval.
Track the physical supply chain
Keep an eye on the industrial manufacturing base. The companies making the specialized components, components like reactor vessels, fuel assemblies, and specialized testing equipment, hold the real leverage. Without businesses like BWX Technologies or GE Vernova, no startup reactor will ever go critical.
Monitor the local grid operator queues
Securing a promise of power from a nuclear startup is meaningless if the local grid operator cannot grant an interconnection agreement. Look at the regional transmission organizations like PJM or ERCOT. Their backlogs for connecting new power sources to the grid are years long. A reactor built in a Texas factory cannot help a data center if it takes five years just to get permission to plug it into the transmission lines.
The tech industry is discovering that the physical world does not follow Moore’s Law. Doubling computing power every two years is a matter of etching smaller transistors onto silicon. Doubling nuclear capacity is a matter of concrete, steel, regulatory clearance, and public trust. The nuclear bros have the cash, the motivation, and the political backing to try an atomic revival. Now they have to prove they can build it without breaking the world around them.
