A startup founded just three years ago has convinced Silicon Valley's top investors that nuclear reactors can be manufactured like smartphones, not built like megaprojects. Valar Atomics closed a $1 billion Series B funding round led by Sequoia Capital, marking a turning point in how the technology industry plans to power artificial intelligence infrastructure. The company's ambition is straightforward but historically difficult: prove that nuclear energy can scale through manufacturing efficiency rather than massive construction projects.

The funding surge reflects an urgent reality facing hyperscalers like Amazon, Microsoft, Google, and Meta. Data centers consume electricity at industrial scales, and the power grid cannot keep pace with demand. Goldman Sachs projects U.S. data center power demand will double from 31 gigawatts in 2025 to 66 gigawatts by 2027, yet only about 50 to 60 percent of data center capacity scheduled for 2027 is expected to come online on time because power delivery, not equipment, has become the binding constraint.

Valar's approach differs from traditional nuclear development. The Hawthorne, California-based company is pursuing high-temperature, gas-cooled reactors that use helium as a coolant. Rather than building one massive plant over a decade, Valar wants to manufacture standardized reactor units repeatedly, clustering hundreds of them at single sites called "gigasites." Each new reactor would generate operational data feeding improvements into the next production run, resembling a manufacturing learning curve more than traditional nuclear economics.

Why Are Tech Giants Turning to Nuclear Power?

The shift reflects a fundamental change in how hyperscalers view energy infrastructure. Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart Pennsylvania's Three Mile Island Unit 1. Google has backed advanced nuclear developer Kairos Power. Amazon has invested in small modular reactor projects. Meta has pursued agreements to secure nuclear generation for its data centers. These are not speculative bets; they are contractual commitments driven by immediate supply constraints.

Xcel Energy, a major utility serving the Upper Midwest and Southwest, reported a "high probability portfolio" for data centers exceeding 20 gigawatts. The company estimates that every 1 gigawatt of data center load requires $5 billion to $6 billion in generation investment, often involving wind, solar, and storage alongside other sources. Xcel expects to secure 1 gigawatt of new data center load by the end of 2026 and a total of 4 gigawatts of additional load by year-end 2027.

The infrastructure under construction across all four major U.S. hyperscalers is not AI-exclusive. Land acquisition, data center shells, power infrastructure, and networking capacity serve both AI and non-AI workloads simultaneously. When Amazon raised full-year capital expenditure guidance to approximately $220 billion, that figure funds the underlying physical plant from which all AWS services are delivered. The four companies are now projected to spend more than $700 billion in 2026 alone, a 77 percent increase from 2025's roughly $410 billion.

How Are Hyperscalers Securing Power for AI Infrastructure?

• Long-term Contracts: Amazon disclosed that its AI business has already scaled to a $25 billion-plus annual revenue run rate growing at triple-digit rates, with demand already booked for 2028 described as "striking" by management, signaling that spending cycles are already partially committed in customer contracts.
• Custom Nuclear Partnerships: Valar demonstrated its Ward 250 reactor generating electricity that powered an NVIDIA Blackwell system, and announced a collaboration with NVIDIA on a waterless 30-megawatt AI factory paired with Valar's waterless reactor technology, particularly attractive in water-constrained regions.
• Utility-Scale Renewable Plus Nuclear: Xcel Energy is investing in approximately 13 gigawatts of new renewable generation and battery storage through the mid-2030s while simultaneously supporting data center growth through large load tariffs designed to ensure new industrial customers contribute to grid fixed costs.
• Sequencing Long-Life and Short-Life Assets: Hyperscalers are committing early to long-lived assets like land, data center shells, and power infrastructure while deferring final decisions on short-lived assets like chips until a few months before deployment, when demand signals are clearer.

Valar's $1 billion Series B represents more than just capital; it signals investor confidence that the nuclear manufacturing model can work at scale. Sequoia partner Shaun Maguire joined Valar's board as part of the investment. Additional investors included Apandion, Atreides Management, Conviction, Dream Ventures, HOF Capital, Point72, Riot Ventures, Snowpoint Ventures, and Valor Equity Partners. Separately, Valar closed a $200 million credit facility led by Erebor Bank and J.P. Morgan, bringing total newly announced equity and credit financing to $1.2 billion.

The company has moved unusually quickly for a nuclear startup. Valar completed its non-nuclear Ward Zero prototype, achieved cold criticality of its NOVA core at Los Alamos National Laboratory, and secured selection for Department of Energy programs focused on advanced reactors and nuclear fuel. On June 18, Valar said Ward 250 achieved self-sustaining criticality, which the company described as the first time a private company had taken a nuclear reactor critical outside a national laboratory.

What Challenges Remain for Nuclear Manufacturing at Scale?

Valar's ambition to eventually manufacture tens, then hundreds, then thousands of reactors annually remains far from proven. The company must navigate regulatory approvals, build manufacturing capacity, develop fuel infrastructure, and demonstrate that economics work at commercial scale. It plans to vertically integrate much of the process, spanning reactor production, deployment, long-term operations, and nuclear fuel manufacturing at facilities located alongside its reactors rather than relying entirely on outside suppliers.

That approach resembles a manufacturing learning curve more than traditional nuclear economics, but it faces real bottlenecks. Supply chains, fuel availability, regulatory approvals, and construction costs can become major obstacles. Valar's bet is that controlling more of those pieces can shorten deployment times and push costs lower as production volumes increase.

Wall Street's confidence in hyperscaler capital spending remains strong despite the power constraint. Morgan Stanley published a research note concluding that consensus 2027 cloud capital expenditure estimates are still too low by roughly $200 billion. The firm's own aggregate 2027 estimate is $1.4 trillion, 17 percent above consensus of $1.2 trillion. The objection is not optimism but arithmetic: consensus growth of 29 percent for 2027 cloud capex implies non-AI cloud infrastructure growing at just 7 percent year-over-year, a rate the firm argues is implausible given the visible committed infrastructure pipeline.

All four hyperscalers remain firmly supply-constrained. External cloud demand from enterprise customers and internal AI workloads are running ahead of available compute capacity simultaneously, with no near-term relief expected. Microsoft stated explicitly that demand exceeds available supply. Alphabet's Chief Financial Officer stated that cloud revenue would have been higher if the infrastructure had been in place to serve it. That is not a 2027 demand uncertainty; it is a 2027 revenue certainty contingent on infrastructure delivery.

The power grid problem has produced a structural shift in how hyperscalers approach energy: they are becoming de facto power companies. This transformation, accelerated by Valar's funding and similar nuclear partnerships, suggests that AI infrastructure will reshape not just technology but the entire energy sector for the next decade.