Evans Sees $1 Trillion AI Build-Out Pushing Frontier Models Into Low-Margin Commodity Infrastructure
Updated
Updated · ben-evans.com · Jul 9
Evans Sees $1 Trillion AI Build-Out Pushing Frontier Models Into Low-Margin Commodity Infrastructure
1 articles · Updated · ben-evans.com · Jul 9
Summary
$1 trillion or more in coming data-center capex should ease today’s AI supply crunch, and Benedict Evans argues that points frontier models toward commodity-like, low-margin economics.
40% to 50% reported inference gross margins look fragile because training costs still exceed revenue, efficiency is improving fast, and buyers’ willingness to pay remains unclear outside a narrow software-development use case.
12 months to five years from now, token pricing will hinge on unknowns including new demand, capacity additions, model differentiation and whether any lab gains durable pricing power through network effects, regulation or export controls.
Evans says the default outcome still resembles mobile data or cloud infrastructure more than a winner-take-all software monopoly: huge spending and strategic importance, but most value captured by companies building on top.
As foundation models become cheap commodities, who will ultimately capture the trillions in value from the AI revolution?
Trillion-dollar investments fuel the AI boom, but can the industry solve its fundamental flaws before the bubble bursts?
AI's voracious energy appetite is straining global power grids. Can efficiency gains possibly keep pace with its explosive growth?
Unprecedented AI Infrastructure Investment: $1.4 Trillion, Energy Strain, and Systemic Financial Risks
Overview
Driven by the escalating demands of advanced artificial intelligence, the years 2025 to 2027 are seeing an unprecedented surge in capital expenditure for AI infrastructure. AI companies are set to invest over $500 billion in 2026 alone, with U.S. tech giants’ total spending projected to reach nearly $1.4 trillion. This massive investment is fueled by the exponential growth in computational power needs, as envisioned by leaders like Nvidia’s CEO, who predicts a shift from billions of human users to billions of AI agents. Robust infrastructure is now essential to power the next generation of AI capabilities, highlighting the critical role of both hardware and energy resources.