Frontier Models Drift Toward Commodity Infrastructure as $1 Trillion Data Center Buildout Eases Supply Crunch
Updated
Updated · ben-evans.com · Jul 9
Frontier Models Drift Toward Commodity Infrastructure as $1 Trillion Data Center Buildout Eases Supply Crunch
1 articles · Updated · ben-evans.com · Jul 9
Summary
$1 trillion or more of coming data-center capex, faster inference efficiency and shifting token use point to easing AI supply constraints, undermining frontier models' current pricing power.
40%-50% reported inference gross margins look fragile because training costs still exceed revenue and buyers' ROI remains unclear beyond software development, the main use case driving this year's crunch.
A durable premium would require a clear moat—fewer frontier rivals, network effects, differentiated capabilities or stronger value capture at the model layer—but the report says none is visible yet.
Over the next 12 months to five years, token prices could swing sharply as supply, demand and capacity reset, with the likelier outcome resembling low-margin infrastructure and value accruing to software built on top.
As AI model costs plummet, who will truly profit: the creators, app developers, or the hardware giants building the infrastructure?
Will 'good enough' AI commoditize the market, or will advanced 'agentic' systems create a new, high-value frontier?
Could government intervention create a protected AI oligopoly, reversing the trend toward open competition and lower prices?
AI’s Trillion-Dollar Power Play: Infrastructure, Energy, and Strategic Risks to 2030
Overview
The report highlights a massive global buildout of AI infrastructure, driven by soaring investments and rapid data center expansion to support advanced AI technologies. This surge is causing electricity demand to nearly double, with data centers consuming more power than some of the world’s largest economies. However, the growth faces immediate bottlenecks, including power supply limitations and hardware shortages, which threaten to delay or cancel many projects. As AI models split into high-cost frontier systems and affordable commodity options, companies must adopt new strategies for resilient, sustainable power and carefully navigate evolving risks, value shifts, and intense global competition through 2030 and beyond.