Nuclear startup Valar Atomics announced Monday it closed a $1 billion Series B led by Sequoia Capital, valuing the three-year-old El Segundo, California company at [$6 billion according to Bloomberg](https://www.bloomberg.com/news/articles/2026-08-03/sequoia-leads-1-billion-funding-round-for-nuclear-startup-valar) — triple the $2 billion it was worth just months earlier. The announcement followed a June milestone that no US nuclear company had ever achieved: a privately built reactor generated electricity and sent it to an Nvidia AI chip, on the first attempt, outside a national laboratory.
The $1 billion in equity — accompanied by a $200 million credit facility from Erebor Bank, J.P. Morgan, Crescent Cove, and Hercules Capital — gives Valar fuel for the next phase of its ambition: moving from one demonstration reactor in the Utah desert to a manufactured fleet of machines that can actually power an AI data center at commercial scale. That transition is where every nuclear startup in American history has stalled. Sequoia, whose partner Shaun Maguire will join Valar’s board, is betting this one will not.
The full investor roster, disclosed in Valar’s official Series B announcement, includes Apandion Capital, Atreides Management, Conviction, Dream Ventures, HOF Capital, Point72 Ventures, Riot Ventures, Snowpoint Ventures, and Valor Equity Partners alongside Sequoia.
What the Ward 250 Actually Did — and What It Didn’t
The milestone that made Monday’s raise possible unfolded in two steps, separated by two weeks. On June 18, 2026, at approximately 4:30 p.m. MDT, Valar’s Ward 250 high-temperature gas reactor achieved self-sustaining nuclear criticality at the Utah San Rafael Energy Research Center in Orangeville. The DOE confirmed the milestone, marking it as the second advanced reactor criticality under President Trump’s Executive Order 14301, which directed at least three reactors to reach criticality by Independence Day — and the first DOE-authorized advanced reactor ever built and operated outside the national laboratory system.
On July 1, a live demonstration followed. Isaiah Taylor, Valar’s 27-year-old founder and CEO, invited investors, state officials, and reporters to the Orangeville site, where a team member plugged an Nvidia DGX Spark — an AI-optimized desktop computer built on Nvidia’s Blackwell architecture — into a circuit fed by electricity from Ward 250. The reactor ran the Spark and hosted a website, and at that moment became the first US nuclear reactor to supply power to an AI chip.
What Taylor told the crowd: Ward 250 was generating roughly 100 kilowatts of thermal energy, running at 37 percent of its intended output at the time of the demonstration. The heat flowed through a pressurized helium loop to a thermoelectric generator — a device that produces electricity directly from a temperature gradient, using the Seebeck effect, without a steam turbine. Every kilowatt the reactor produced was generated and measured by Valar itself; no independent auditor verified the 37 percent figure or the 100 kilowatt claim on the day.
That caveat matters for context. The Ward 250’s design target for commercial deployment is up to 5 megawatts electric — roughly 50 times the thermal power it was producing during the demonstration. The Nvidia DGX Spark is an AI workstation that draws a few hundred watts of power, not a data center GPU rack that draws 50 to 100 kilowatts. The demonstration proved the reactor can produce electricity and deliver it to a real load. It proved neither that the system is ready for commercial deployment nor that the thermoelectric conversion architecture used at the test site would translate to the efficient power generation a commercial data center requires.
For commercial scale, Valar will need to move from the thermoelectric generator used in the Ward 250 test to a closed Brayton cycle — a helium gas turbine — or a steam generator, both of which achieve heat-to-electricity efficiencies of 35-40 percent versus a thermoelectric’s 5 to 10 percent. That engineering step, from test architecture to commercial architecture, is a significant one, and is part of what the $1 billion is designed to fund.
How the Ward 250 Works — and Why Helium Matters
The Ward 250 belongs to a class of reactors called high-temperature gas-cooled reactors (HTGRs). Where conventional nuclear plants circulate pressurized water through their cores to carry heat away from the fuel, Ward 250 circulates helium — an inert gas that cannot activate or flash to steam in a loss-of-coolant accident. Because helium boils at -269°C (-452°F) at atmospheric pressure, the reactor loop can operate near atmospheric pressure — eliminating the high-pressure vessel that makes conventional pressurized-water reactors mechanically complex and costly to license.
The fuel is TRISO — tri-structural isotropic — particles, each about the size of a poppy seed. Inside each particle sits a uranium kernel in ceramic layers, with a silicon carbide shell acting as an individual containment vault, retaining fission products at temperatures exceeding 1,600°C (2,912°F) — well above any credible reactor accident temperature. A TRISO-fueled reactor cannot melt down in the way water-cooled reactors can, because the fuel’s containment is distributed through millions of individual ceramic shells rather than concentrated in a single cladding.
The hot helium that exits the Ward 250 core was converted to electricity at the July 1 demonstration via a thermoelectric generator — simpler and more reliable than a steam cycle, but far less efficient. The commercial path for the Ward 250’s successors involves replacing the thermoelectric generator with a Brayton-cycle turbine that drives a generator directly with the hot pressurized helium, achieving efficiencies competitive with conventional power plants.
The entire Ward 250 reactor — at less than 100 kilowatts thermal — fits inside a package small enough to transport in a C-17 military transport aircraft; components made the trip from March Air Reserve Base in California to Utah aboard three C-17s in February 2026. The commercial reactor design targets up to 5 megawatts electric. Valar’s 30-megawatt feasibility study with Nvidia would require roughly six Ward-class reactors operating in parallel — a far cry from the single 100-kilowatt test unit currently running in Orangeville, but precisely the scale of multiplication the $1 billion is meant to fund.
Why Behind-the-Meter Nuclear Has Become the Race
The structural reason Valar attracted a $1 billion round at a $6 billion valuation is not primarily the Ward 250’s physics. It is the data center power bottleneck that has emerged at the core of the AI industry.
U.S. grid interconnection queues have grown to more than 2,600 gigawatts of pending projects, with average wait times of four to five years in many regions. The International Energy Agency has confirmed that the SMR pipeline for data centers grew from 25 gigawatts at the end of 2024 to 45 gigawatts by April 2026 — an 80 percent expansion in 18 months. AI-focused data centers grew power consumption 50 percent in 2025 alone, according to the same IEA analysis.
That growth is driving a structural mismatch: AI racks now require 50 to 100 kilowatts of power per rack, compared with 5 to 10 kilowatts for traditional server racks, and the U.S. grid was not built for that density. Microsoft, Amazon, Google, and Meta have all signed direct nuclear power purchase agreements and invested in SMR startups seeking to bypass public utilities entirely.
Valar’s pitch to data center operators is specific: rather than waiting years for grid-connected nuclear, a Ward-class reactor can be manufactured in a factory, shipped to a data center site, and operated as behind-the-meter power — drawing no electrons from the public grid, eliminating the interconnection wait, and sidestepping the utility relationship entirely. The reactors are also waterless — helium cooling in a closed loop means no evaporative cooling tower, no water intake from local aquifers, and no exposure to the community water-use backlash that has stalled or cancelled conventional data center projects in water-stressed Western states.
Utah’s governor signed new data center development rules in May 2026, following community opposition to projects’ water and power demands. A nuclear-powered AI facility consuming neither grid electricity nor local water addresses both objections at once — a feature that makes the Orangeville site’s regulatory and community context unusually favorable for Valar’s purposes.
The Nvidia partnership, formalized on July 1 alongside the reactor demonstration, is a feasibility study for a 30-megawatt closed-loop AI facility in Utah. Levi Patterson, Nvidia’s director for energy, science, and AI infrastructure policy, appeared on stage with Taylor at the demonstration. Nvidia’s financial interest in the partnership’s success is direct: the company reported $75.2 billion in data center revenue in its most recent quarter, a 92 percent year-over-year increase.
Why $6 Billion Without a Commercial License
The hardest question a $6 billion valuation asks of Valar Atomics is not whether the Ward 250 works. It is whether the company can cross the regulatory gap between where it is now — operating under a Department of Energy Reactor Pilot Program authorization — and where it needs to be to sell a single commercial kilowatt-hour: licensed by the Nuclear Regulatory Commission.
These are not the same thing. The DOE Reactor Pilot Program is an expedited authorization designed for research and demonstration, covering a specific test reactor at a specific site under a Documented Safety Analysis. Valar received preliminary DSA approval in February 2026 and final approval on April 23. It is not a commercial license. NRC Chairman Ho Nieh said explicitly in a Deseret News editorial board meeting that companies like Valar must still pursue commercial licensing through the NRC after the DOE pilot phase: “A company like Valar can take that design and move it into commercial licensing through the NRC,” Nieh said.
The NRC licensing path for a novel non-light-water reactor design like the Ward 250’s HTGR is uncharted at commercial speed. NuScale’s VOYGR — the only SMR the NRC has fully licensed, and a conventional pressurized-water design with far more prior licensing history — required more than a decade of NRC review. The Trump administration’s executive orders directed the NRC to target 18-month review timelines for new designs, but that goal has not yet been tested against a commercial HTGR application.
Complicating matters further, Valar has been simultaneously litigating against the NRC. In April 2025, the company joined states including Texas, Utah, Louisiana, Arizona, and Florida in a lawsuit challenging NRC licensing authority over certain small reactors — a claim that nuclear engineer and TerraPower researcher Nick Touran publicly challenged. Touran also disputed Valar’s claim, made as part of the same legal argument, that Ward 250 spent fuel was safe enough to hold by hand — calculating the radiation dose as lethal within milliseconds of contact.
The litigation was still unresolved as of Monday’s announcement, with TechCrunch reporting repeated pauses in the case in a way that suggests a settlement may be in progress.
Sacra’s independent analysis of Valar’s commercial risks identified what it called “regulatory fragility”: the company’s commercialization timeline “depends on an alternative regulatory framework remaining politically durable and legally intact” over the years required to move from test reactor to multi-unit commercial deployment. The fuel supply chain is a parallel risk: Valar’s commercial HTGR will require TRISO-coated fuel at industrial scale, from a supply chain that is “only beginning to receive commercial licenses and start construction in the U.S.”
Natixis Corporate and Investment Banking, in a June 2026 report on the SMR sector, concluded that the industry “was still far from commercial reality and was entering a phase where engineering ambition must confront regulatory complexity, financing realities and industrial execution.” Academic analysis has reinforced that picture: a 2023 Monte Carlo study published in the journal Energy found that none of 19 surveyed SMR designs were economically competitive without subsidy, and that HTGR designs specifically had median levelized costs of electricity between $116 and $137 per megawatt-hour before any subsidy credit.
None of that economic history proves Valar cannot succeed — it proves that no one has yet succeeded at what Valar is attempting on this timeline. The $1 billion is the financial expression of a thesis that factory iteration, software-driven design improvement, and an aggressive regulatory posture can compress what has historically taken decades into a window the AI industry’s energy timeline can accommodate.
The Investor Roster and the Bet It Represents
Sequoia leads the round, with Shaun Maguire joining Valar’s board. Maguire is understood to have moved quickly after Valar’s Nvidia partnership became public in late June, recognizing — per sources familiar with the deal — that nuclear partnerships with hyperscalers were establishing a strategic moat in energy infrastructure before most AI investors understood what they were looking at.
Atreides Management, Point72 Ventures, Snowpoint Ventures, Apandion Capital, Conviction, Dream Ventures, HOF Capital, Riot Ventures, and Valor Equity Partners all joined the Series B. Earlier investors include Anduril founder Palmer Luckey and Palantir CTO Shyam Sankar — a roster that places Valar at the intersection of defense technology, AI infrastructure, and frontier energy in a way that few other companies can claim.
The $6 billion valuation is not without risk. Valar is one of four companies to have achieved a self-sustaining nuclear chain reaction in 2026 — Antares (June 4, using a different reactor technology), Valar (June 18), and Aalo Atomics (July 4, sodium-cooled) are three of them — and all are raising large rounds simultaneously. None has yet deployed a commercially licensed reactor. Moving from a sub-megawatt test demonstration to a commercially licensed, multi-unit data center power source involves regulatory approvals, fuel supply chains, and manufacturing scale-up that no nuclear startup has completed at speed. At a $6 billion valuation from a company that is three years old, the margin for the kind of delays that have historically plagued nuclear construction is narrow.
Valar’s own founder has built the iterative thesis into the company’s communication. “It took two years to complete the NOVA core. It took seven months to take Ward 250 critical,” Taylor wrote in the official fundraising announcement. “With each reactor built, the tick rate will become smaller until Valar is producing tens, then hundreds, then thousands of reactors per year.”
What Comes Next
The $1 billion gives Valar the capital to pursue several tracks in parallel: continuing Ward 250 power operations and data collection in Utah, beginning design work on larger commercial reactor variants, advancing the Nvidia feasibility study, and — critically — preparing the regulatory case for NRC commercial licensing.
What it cannot buy directly is the licensing itself. That process, for a novel HTGR design, will be measured in years. The DOE Reactor Pilot Program that authorized Ward 250 was explicitly designed as a stepping stone — a way to generate operational data at small scale that can then be submitted to the NRC as the foundation of a commercial license application. As of the $1 billion announcement, Valar has not publicly filed that application.
The company is also building its fuel supply in-house: the Valarin Fuel Fabrication Facility, co-located with Ward 250 at the Orangeville site, is designed to manufacture TRISO fuel on-site using established German HOBEG technology. Vertical integration of the fuel supply — producing its own TRISO fuel rather than depending on third-party suppliers whose capacity is committed to other reactor designs — addresses one of the most concrete risks in Sacra’s analysis.
Whether that vertical integration advantage, combined with $1 billion and a Sequoia board seat, is enough to produce commercially licensed nuclear power on the AI industry’s timeline is the question the next few years of Valar’s operation will answer. The Ward 250 reactor, still running in the Utah desert as of Monday, is generating the data that will either prove the model or reveal its limits.
Frequently Asked QuestionsWhat is the Ward 250, and how does it generate electricity for AI?
Ward 250 is Valar Atomics’ test reactor at the Utah San Rafael Energy Research Center in Orangeville, Utah. It is a high-temperature gas-cooled reactor (HTGR) fueled by TRISO-coated uranium particles and cooled by pressurized helium. Helium flowing through the reactor core picks up heat from the TRISO fuel and carries it to a thermoelectric generator, which converts the temperature gradient directly into electricity using the Seebeck effect. On July 1, 2026, that electricity powered an Nvidia DGX Spark desktop AI computer — the first time any US nuclear reactor supplied power to an AI chip. The reactor is a 100-kilowatt-thermal test unit; the commercial design targets up to 5 megawatts electric. The thermoelectric conversion used at the test site will be replaced by a more efficient helium Brayton-cycle turbine for commercial deployment.
Why does the gap between Valar’s current DOE authorization and an NRC commercial license matter for investors?
Ward 250 operates under a Department of Energy Reactor Pilot Program authorization — an expedited research approval designed to generate test data, not to authorize commercial power sales. Selling electricity commercially in the United States requires a separate Nuclear Regulatory Commission license. NRC Chairman Ho Nieh has confirmed publicly that companies like Valar must still pursue commercial licensing through the NRC after the DOE pilot phase. For a novel non-light-water reactor design — which the Ward 250’s HTGR is — NRC licensing is an uncharted process with no guaranteed timeline. NuScale’s conventional pressurized-water SMR required more than a decade of NRC review. Valar is simultaneously suing the NRC over its licensing framework, adding legal uncertainty to the commercial path. The $1 billion raised Monday funds the attempt to bridge that gap at a pace the AI industry’s energy timeline demands. No nuclear startup has ever succeeded at crossing it this quickly.
What does “factory nuclear” mean, and why hasn’t it worked before?
The nuclear industry’s historical cost and schedule failures trace largely to bespoke, one-of-a-kind construction: each plant was designed specifically for its site, built from scratch on location, and never repeated at scale. Valar’s thesis is that treating nuclear reactors the way SpaceX treats rockets — factory-built, iteratively improved across a serial production run, with each unit generating data that improves the next — can change that cost curve. Academic research confirms that SMR factory economics do not materialize until a manufacturer produces 40 to 70 units of the same design. Valar has produced one. The $1 billion is intended to fund the path toward the threshold where factory production actually becomes cost-competitive — and the question the market is currently pricing at $6 billion is whether the company can reach that scale before the regulatory, fuel supply, and financing constraints that have stopped prior nuclear startups do the same to this one.
Can AI data centers realistically run on nuclear power, given how AI workloads vary?
AI training and inference workloads impose rapid, large-scale power fluctuations on any generator they’re connected to — GPU clusters alternate between high-draw compute phases and lower-draw communication phases in cycles as short as seconds. Conventional nuclear reactors are designed for constant baseload output, not rapid ramp-up and ramp-down. Whether nuclear microreactors can maintain stable power delivery to a data center running real GPU workloads is the central engineering question of the nuclear-AI integration effort. Valar and Nvidia announced a feasibility study for a 30-megawatt closed-loop facility in Utah. Separately, Aalo Atomics and Crusoe are building toward what they describe as the first nuclear-powered AI Factory data center at Idaho National Laboratory in 2027 — a test that will produce the first real dataset on whether nuclear power can buffer GPU variability at data center scale.