The concept that every country should develop its own AI capability — known as “sovereign AI” — is picking up steam both among industry giants like Nvidia and inside the halls of D.C. power. This week, POLITICO and its guests got a close look at different strands of that idea when two people thinking about America’s AI future opened up at our AI & Tech Summit. The top science official at the White House, Arati Prabhakar, said the government had to provide a counterweight to the vast computing infrastructure and capacity currently concentrated among the industry giants. “What’s happening today is an extraordinary moment, where a huge surge in a powerful technology is coming largely from a few large tech companies,” said Prabhakar, director of the Office of Science and Technology Policy. “There's a lot of good in what's happening, but by itself, it's not going to get us where we need to go.” Unlike some other countries, whose AI strategies may be driven by fears of losing out amid a shortage of computing resources, the U.S. government wants to fund public resources in part to hit policy goals like promoting a more diverse AI ecosystem, said Pablo Chavez, an adjunct senior fellow at the Center for a New American Security think tank. The government’s focus on AI-specialized supercomputing infrastructure is one concrete example of the broader notion of sovereign AI, he said. Both are motivated by a desire to maintain national control over the development of this fast-moving technology and the ethical challenges it’s bound to face. Prabakhar pushed to fully fund the National AI Research Resource, which launched as a pilot in January to provide U.S. researchers and educators with free access to advanced computing, datasets, powerful AI models and user support. Last week, the House Science Committee greenlit a bill to formally authorize and fund NAIRR for $2.6 billion. But with the legislative calendar rapidly ticking down, Congress has little time to pass it (it could get tacked onto must-pass legislation if it’s popular enough). Former President Donald Trump’s chief technology officer, Michael Kratsios, made his own appeal to sovereign AI, for a different reason. “Each country wants to have some sort of control over our own destiny on AI,” said Kratsios, now the managing director for Scale AI, later on Tuesday afternoon. “The most sort of like base, easiest thing that they can do is try to create a model that is fine-tuned to the language, culture, tradition or specifics of that country.” But, as he pointed out, there are countless interpretations of what sovereign AI actually means in practice. For example, governments around the world like the Netherlands have directly and indirectly financed the development of large language models to protect their national languages or other cultural principles. Countries have also tried to position themselves as AI hubs to attract talent and startups. Outside of government, sovereign AI promoters like Nvidia have a vested interest in their chips and products being the ones to power that infrastructure. But some policy experts also warn that the U.S. government must develop in-house capabilities, because relying on corporate expertise might undermine its role in setting and enforcing AI standards. There’s also a risk — according to Kratsios — that forfeiting U.S. leadership on fine-tuning AI models could leave an opening for an adversary to step in. “The Chinese are very open to being the ones to do that, and they would love to be the ones who then build the compute stack which they run on as well,” he said Tuesday. “We have to be much more vigilant in the U.S. to kind of push back on that.” The idea of sovereign AI is not an all-American concept. In fact, it has roots in the digital sovereignty movement and China’s playbook — although the U.S is not seeking to be entirely independent from the global tech ecosystem, and has followed a more collaborative strategy it calls digital solidarity. Even in all this, the U.S. is largely moving in step with allies. U.S. national labs have had supercomputers for decades. But now, in response to AI, “countries are starting to either retrofit their supercomputing infrastructure, or they're building completely brand new supercomputing infrastructure,” said Chavez. For example, Brazil recently announced that it wants to build an AI specialized supercomputer that would be among the top five in the world. “NAIRR is an example of a robust government effort around creating an AI ecosystem, and I'd say that there’s a lot of countries that are essentially trying to, in one way or another, replicate that model,” Chavez said.
| 
