What's at the Other End of Stargate?

On January 21, President Trump announced an initial $100 billion investment for the Stargate project with plans to invest $500 billion over the next four years by OpenAI, Oracle, and Softbank into building new data centers and other infrastructure to power artificial intelligence. 

Stargate is breaking ground on new data centers – will this be helpful?

Called the “Stargate Project,” the investment has been compared to the Manhattan Project and the Apollo Program primarily because of its enormous size. However, as Politico reporter Derek Robertson points out, these comparisons aren’t especially apt because of the lack of a clear goal: the Manhattan Project involved building a nuclear weapon. Similarly, the Apollo Program involved sending humans to the Moon and returning them safely to Earth. So what, exactly, is the Stargate Project supposed to accomplish?

The second Trump administration has been keen on ensuring American “dominance” in AI – President Trump’s executive order on January 23 repeatedly stresses the need for America to “solidify its position as the leader in AI.” 

The Center for AI Policy (CAIP) supports these goals of ensuring American dominance in AI, but we are skeptical that building new data centers is the key.

US-based tech companies regularly donate a colossal amount of intellectual property to rivals like China by openly publishing their software or by maintaining such lax cybersecurity practices that Chinese hackers are able to reverse engineer their products. For example, China’s new “DeepSeek” AI reasoning model is said to outperform OpenAI’s o1 on several benchmarks. DeepSeek was “based on existing open source architectures like Qwen and Llama,” but the problem isn’t just open source software. Even when a company’s AI assets are theoretically closed-source, the FBI warns that Chinese agents are often able and willing to steal them.

If America builds new data centers without tightening our cybersecurity and beefing up our counterintelligence efforts, then we won’t pull any further ahead of China, because China will still be able to acquire US technology within a few months of when we invent it. The Economist reports that China's DeepSeek was trained using 2,000 second-rate chips—versus 16,000 first-class chips for Meta’s model. Onshoring the data centers is of limited use if we’re not going to adequately protect the AI models generated by those data centers. 

To surpass our rivals, we need to develop strategies specifically aimed at widening the gap between US and Chinese capabilities—not just strategies that will send both countries racing recklessly forward in tandem. For example, CAIP’s 2025 Action Plan calls for programs that bring together cybersecurity researchers and AI companies to discuss cybersecurity best practices and amend the Cyber Incident Reporting for Critical Infrastructure Act (CIRCIA) to ensure that AI companies report all major cybersecurity incidents, enhancing visibility into national security threats.

American AI leadership needs to be aimed at promoting American values

The push to accelerate the construction of new data centers is particularly troubling because, as Sam Ashworth-Hayes, business columnist for The Telegraph, points out, the implicit goal of the AI being developed at these data centers is to “make humanity obsolete.” This is not hyperbole.

Sam Altman’s stated goal is to develop superintelligence, i.e., software that can radically outperform humans at any and all tasks. If Altman is successful, then how will people earn a living?

If the AIs can thoroughly outcompete humans, then we will be at the AIs’ mercy. Why should we trust that these systems will treat humans kindly when they have been trained on a haphazard collection of Internet data that includes a great deal of cruelty and apathy? Why should we trust that these AIs will honor American values?

As Ashworth-Hayes notes in his column: “If it all works out, we’ll be rich. If it doesn’t, we could end up anywhere on the scale between ‘dystopian science fiction’ and ‘dead.’”

The new Trump AI Executive Order offers a constructive path forward

One encouraging note from the second Trump administration’s first opening days is Section 2 of the new AI Executive Order, which states that “It is the policy of the United States to sustain and enhance America’s global AI dominance in order to promote human flourishing, economic competitiveness, and national security.” (emphasis added)

These are good reasons to seek dominance – but they will not occur spontaneously; we must work toward these goals. AI that will protect our national security is worthy but not enough. We need to take action to prevent AI-powered weapons of mass destruction, such as by reintroducing and passing legislation introduced by Senator Romney (R-UT) last Congress, which outlined a framework to handle chemical, biological, radiological, and nuclear (CBRN) threats. If we want AI that will promote human flourishing, we need to invest in aligning AI with human values – as the first Trump administration did through the National Science Foundation.

The new Trump AI executive order calls for an AI Action Plan to be published within 180 days. That’s a good idea and a good timetable. CAIP calls on Trump’s technology leadership team – including Michael Kratsios, David Sacks, and Sriram Krishnan – to ensure that the new AI Action Plan includes concrete, substantive policies to secure American AI. 

If we are going to invest $500 billion to enhance American AI capabilities, then let’s make damn sure that those capabilities remain under Americans’ control.