The microreactor milestone is real, but shifting AI models, China’s challenge, and deployment delays could complicate America’s nuclear ambitions.

It was impressive that four microreactor start-up companies beat the clock and achieved reactor criticality in compliance with the Trump administration’s arbitrary July 4 Reactor Pilot Program (RPP) deadline. But these reactors are years away from deployment, and the excessive swaggering associated with this modest achievement is out of step with the quavering nuclear energy realities ricocheting off a rapidly evolving artificial intelligence (AI) system.  

US Microreactor Achievements

Of the four reactors that achieved zero power, only Antares has an identified market application beyond this demonstration. Its Mark-1 reactor has been selected as a pilot technology for powering Joint Base San Antonio, Texas.

The Antares Mark-0 reactor, which is the zero-power test reactor and prototype for the commercially oriented R1 Mark1, was the first to meet the criticality deadline on June 4 at Idaho National Laboratory (INL). It uses High Assay Low-Enriched (HALEU) TRistructural isotropic (TRISO) fuel and is sodium-cooled. The projected commercial reactor can provide 100 kilowatts to 1 megawatt (MW) electric power. Beyond its work with the Department of Defense (DOD), Antares also has been working with the National Aeronautics and Space Administration (NASA) on potentially powering a Moon base. 

Valar Atomics’ Ward-250 achieved criticality on June 18 in Utah and claimed it produced 10 kilowatts (kW) of thermal output. It had previously announced a criticality test at Los Alamos National Laboratory in November 2025. The Ward-250 is a TRISO-fueled high-temperature gas reactor (HTGR) that uses helium as a coolant. After the test, Valar announced a partnership with Nvidia to explore powering AI data centers and stated that its evolving commercial unit could be “deployed by the hundreds at ‘gigasites’” that could support manufacturing as well as energy production.

Deployable Energy, a year-old company, met the criticality mark on June 30 at Idaho National Laboratory (INL). The company’s founder reportedly delivered the reactor core in the bed of a Ford F-150 truck. The reactor is part of the Nuclear Energy Launch Pad initiative at INL, which is designed to carry the RPP into the future. The reactor is a water-moderated, helium-cooled design using 4.9 percent enriched fuel. It is considered a “nuclear battery” and would max out at 1 MW. It is designed to fit in a 20-foot shipping container and be dropped at any location.

Aalo Atomics Aalo-X Critical Test Reactor sneaked in under the criticality deadline at midnight on July 4 at INL. The test reactor included a full-scale core and was designed to demonstrate key components that will be used in the Aalo-X, a low-enriched uranium (LEU) fueled, sodium-cooled 10 MW reactor. The goal is to deploy groups of these reactors in 50 MW Aalo Pods to support commercial data centers. Aalo recently formed a strategic partnership with Crusoe, a vertically integrated AI infrastructure provider, to  “validate nuclear power’s effectiveness with AI workloads” at INL.

The Mutating AI Data Center Model 

With the microreactor milestone behind them, the Trump administration and reactor developers now face an evolving landscape for nuclear energy. AI data centers have been viewed as the top driver of America’s energy growth and ideal for pairing with nuclear energy. Projections indicate these centers could add 100 gigawatts (GW) to the US grid by 2030. 

But the data center opportunity is rapidly morphing, and opposition is growing; the now-dominant American AI model is under pressure at home and abroad. That raises questions about what role new nuclear power ultimately will play as the AI arena evolves.

The assumption has been that the hyperscalers, including Apple, Amazon Web Services, Microsoft Azure, and Google Cloud Platform, will build massive data centers as fast as they can to support US AI  frontier modelcomputing growth. One assessment identified the data center build time as roughly two to three years. 

But the frontier model concept and its computing campuses are under pressure.

Alex Karp, CEO of Palantir, has been on a recent crusade criticizing the American AI frontier model as a trap that requires businesses to provide their intellectual property (IP) to AI’s leading companies like Anthropic and OpenAI. Karp claims that these models’ architecture creates an “addiction” among users and allows the system provider to absorb proprietary corporate data that can be used to compete against their current customers in the future. 

He argues that companies should build their AI systems on models they own, training them on their proprietary content and keeping their data under their control. He also advocates for an “application layer” his company has developed with NVIDIA that can protect an enterprise’s IP. If Karp’s concept catches on, and Meta seems to have taken note, it could impact the mega data center build-out, reducing the number required and downsizing their power requirements. That could be a challenge for some leading nuclear reactor companies betting on the data center boom, or it could favor dial-an-output modular reactors if the costs are competitive.

China’s Dual AI Challenge

A version of the Karp concept seems to have already been adopted by China, which, because it is already closing the quality gap with the best US AI models, now poses a double threat to US dominance. 

China is giving away fast, cheap open-source AI models. These free Chinese AI platforms are popular in Africa and Latin America but are also being used by US and European companies, including Siemens, DoorDash, and Airbnb. 

Of course, one goal of China is to allow companies to easily build systems on their platform in order to serve geopolitical objectives and create dependencies it can exploit in the future. The US National Institute of Standards and Technology (NIST) evaluated China’s DeepSeek AI model and found that its AI offerings advance Chinese Communist Party narratives. It concludes that China’s DeepSeek “remains a leading open weight model developer and has contributed to a rapid increase in adoption of PRC [People’s Republic of China] models globally.”  It cautions that the expanding global use of these models “May pose a threat to application developers, to consumers, and to US national security.” 

This approach by AI is similar to China’s Belt and Road Initiative (BRI), which has helped the country capture hundreds of infrastructure projects across over 140 countries, resulting in deep relationships and financial dependencies, particularly with emerging economy nations. And while its energy projects so far have heavily tilted toward fossil fuels, BRI’s relationships and financial model enhance China’s position as a future provider of new nuclear energy in the Global South, posing a threat to the future US export of small reactors and creating worry for the US political, commercial, and national security establishment.

Lagging Nuclear Reactor Technology and Scaling Challenges

Even if the US data center boom were to play out as planned, there are questions about when and whether new US nuclear power will be ready to power it. 

So far, the hyperscalers have modestly invested in small modular reactors (SMRs), but they are unlikely to be available in time for a near-term data center boom. If the buildout is over by the early 2030s, most SMRs will still be on the sidelines.

As a hedge, these companies have banked on the resurgence of mothballed nuclear power plants like Three Mile Island in Pennsylvania and Palisades in Michigan. Both of those projects are estimated to deliver new power before the end of this decade for a combined 1.6 GW. What the hyperscalers have not done is back construction of new large reactors despite the fact that a single Westinghouse AP-100 can provide over one GW of power. 

The US government wants 10 new AP-1000s under construction by 2030, but its $80 billion strategic partnership with Westinghouse and its commitment to provide $17.5 billion in loans for long lead items haven’t attracted the necessary additional investment and commitment required from the private sector. 

This reticence reflects a cost-overrun risk aversion from domestic electric utilities and Wall Street. In response, the administration is repurposing federal lands to create data center-nuclear power campuses. And it has turned to allies such as South Korea and Japan to jumpstart the process with foreign cash derived from bilateral trade and tariff deals. 

But that avenue also doesn’t seem to be working. The Trump administration has been frustrated with the lack of progress on the nuclear energy component of the $350 billion South Korean commitment to invest in the United States, and with Japan, it has found that its involvement in a $40 billion nuclear power project is running aground over nuclear liability concerns.

With less than 30 months left in Trump’s term, and the prospect of a hostile House of Representatives sucking the air out of his agenda, there is an urgency in rethinking the Trump administration’s nuclear deployment policy. The dependence on a rapidly evolving AI and data center market leaves questions about whether the country will achieve its long-term core nuclear deployment objectives. And over-relying on a boost from the micro reactor miracle offers meager dividends when deployment is paramount.

About the Author: Kenneth Luongo

Kenneth N. Luongo is a recognized innovator, entrepreneur, and leader in global nuclear energy and transnational security policy. He is the president and founder of the Partnership for Global Security (PGS). He has been a TEDx presenter, written over 100 articles, including in The New York Times and Foreign Affairs, and engaged extensively with global media, governments, and audiences around the world on nuclear energy and transnational security challenges and responses. He was formerly a senior advisor to the secretary of energy and a professional staff member on Capitol Hill.

The post The Microreactor Miracle Meets a Morphing AI-Nuclear Future appeared first on The National Interest.