We Need More Compute to Power Our Future

The demand for AI has skyrocketed over the past few years. This is a positive step for lasting prosperity for American families and communities. But we are at risk of drifting away from that path.

Compute, which powers AI tools and our digital economy, isn’t keeping up with demand. This is not theoretical. As the Wall Street Journal reports, frontier AI companies like Anthropic have faced frequent outages more recently with their tools like Claude and Claude Code. Angel Au-Yeung and Robbie Whelan point out how Anthropic has metered computing supply to users in peak hours as a fix. However, that is not a sustainable solution. Platforms like Anthropic serve millions of users, meaning those limits are hit quickly.

That is a loss for consumers who rely on these models to build products and tools that improve their lives and others’. It’s also bad for AI companies who have to allocate resources towards handling constant outages instead of working to improve their models and build new ones.

Rationing usage will limit features that customers can take advantage of. Degraded performance would be inevitable. With agentic AI on the rise and enterprise adoption climbing, AI will become a crucial player to maximize productivity. When models must be constantly evaluated to ensure quality and protect users, these compute constraints are alarming.

The lack of compute is also causing increases in the prices of GPUs, which is critical to accelerating development in data centers. In the article, J.J. Kardwell, the Chief Executive Officer of Vultr (a cloud infrastructure firm), stated how “the question is, why don’t we just deploy more gear? The lead times are too lengthy. Data center buildouts take years, the power that’s available through 2026 is already all spoken for.” GPUs already have long manufacturing lead times, and data centers take years to build. It doesn’t help that our permitting process is mired in steep governmental regulations. At this rate, energy won’t be our only bottleneck—compute will join the list.

If we lack data centers, chips and other parts of the stack, we are forfeiting the ability to usher in a stronger economy and our chance to lead in the AI race against our adversaries. It is imperative that we invest in and build more compute capacity.

Increasing compute capacity will enable more GPU clusters, data centers, and faster deployment timelines. Infrastructure is highly capital-intensive. When compute is scarce, costs rise—discouraging companies from scaling and limiting their ability to reach more people. Our global dominance depends on the ability to go all-in on compute.

‍Jay Burstein is a fellow with Build American AI.