The Trillion-Dollar AI Capex Gap: Why Too Much Hardware Could Be Nvidia’s Trap and Microsoft’s Opportunity

The problem with AI valuations is that infrastructure spending is running far ahead of the revenue available to support it. Goldman Sachs estimates roughly $7.6 trillion in cumulative AI capital spending from 2026 through 2031. OpenAI and Anthropic, meanwhile, were generating combined annualized revenue of more than $105 billion by August 2026, which is impressive growth, but still a small base relative to the buildout.

To avoid cataclysmic infrastructure write-downs on hardware with brief 3-to-5-year lifecycles, the industry must scale its annual recurring revenue past $1 trillion by 2030.

Too Much Capacity, Too Soon

Skeptics think the industry is building hardware capacity faster than customers can use it profitably. Under the bear case, the revenue gap eventually reaches hardware suppliers. If enterprise demand fails to fill new capacity, hyperscalers will slow purchases, demand lower prices and move routine workloads to internal chips. Short hardware lifecycles would make even a temporary glut expensive.

Nvidia Corporation (NASDAQ:NVDA) carries the greatest exposure because its 74.9% quarterly gross margin depends on customers competing for scarce, high-end GPUs. Google’s TPUs, Amazon’s Trainium and Microsoft’s Maia can absorb predictable inference workloads, reducing Nvidia purchases and increasing…