Compute becomes lifeblood, constraint of AI boom

Computing power has become the lifeblood — and a key limiting factor — of the race to develop AI, as the push to integrate the technology into daily life clashes with the finite supply of one of its most crucial inputs.

Often referred to simply as “compute,” the processing power that forms the foundation of AI is limited by physical infrastructure: the chips, servers and data centers that take up space in the real world.

As AI companies grapple with the political and financial realities of obtaining and expanding access to compute, increased AI usage and the rise of agentic AI are raising the stakes and making that processing power even more valuable.

“Modern AI systems don’t work without compute. You need compute for almost everything, and more compute is almost always better,” Erich Grunewald, a senior researcher on the compute policy team at the Institute for AI Policy and Strategy, told The Hill.

Compute has long been at the heart of the AI boom. For years, AI leaders have been pushing to rapidly build more data centers to support the vast computing needs of their technology.

Tech companies have poured billions of dollars into expanding AI infrastructure, investing more than $400 billion last year. Total AI capital expenditures could reach $800 billion this year and rise above $1 trillion in 2027, according to Bank of America estimates.

This endeavor initially enjoyed wide bipartisan appeal, with President Trump and former President Biden alike signing orders to fast-track data center construction.

But public opinion has rapidly shifted on data centers, complicating the politics surrounding the buildout. As Americans have become increasingly concerned about the impacts on their electricity bills and local environment, numerous projects have been blocked or delayed.

Supply chains are also a key factor constraining compute, noted Janet Egan, senior fellow and deputy director of the Technology and National Security Program at the Center for a New American Security.

“People on the inside of AI have been talking about compute........

© The Hill