Frontier labs don’t use most AI compute (yet)

2026-01-28 · Source: Epoch AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Advanced, extended

Summary

An analysis of global AI compute usage reveals that leading frontier labs, including OpenAI, Anthropic, xAI, Google DeepMind, and Meta Superintelligence Labs, collectively utilized less than half of the world's operational AI compute supply by the end of 2025. Despite OpenAI initiating the AI boom, its research, training, and inference accounted for only 10% to 15% of the global 20 million Nvidia H100-equivalent (H100e) units. Anthropic and OpenAI are, however, rapidly increasing their compute capacity, growing approximately four times annually compared to the industry's three-fold growth. This trajectory suggests these top labs could absorb 80% of global compute within five years, potentially slowing overall AI capability growth unless the already nearly \$1 trillion annual AI capital expenditure dramatically accelerates.

Key takeaway

For Directors of AI/ML or investors planning AI infrastructure or investment strategy, the rapid compute growth of frontier labs like OpenAI and Anthropic signals impending market consolidation. Your long-term AI strategy must account for this shift, as their accelerated compute acquisition, exemplified by Anthropic's \$15 billion xAI deal, will tighten supply and increase costs. Prepare for a future where scaling capabilities may be constrained by overall compute production, demanding significant capital expenditure and potentially slowing the pace of AI progress.

Key insights

Frontier AI labs currently use less than half of global compute, but their rapid growth could lead to market consolidation and unsustainable capital expenditure.

Principles

• Compute growth at top labs outpaces global supply.
• AI capital expenditure is nearing \$1 trillion annually.
• Consolidation of compute by leading labs is accelerating.

In practice

• Monitor top labs' compute acquisition for market shifts.
• Evaluate AI capex trends for investment decisions.
• Assess compute supply chain constraints.

Topics

• AI Compute
• Frontier AI
• Large Language Models
• OpenAI
• Anthropic
• AI Infrastructure
• Capital Expenditure

Code references

• epoch-research/ai-chip-counts

Best for: Director of AI/ML, CTO, Investor

Related on AIssential

• The Sequence Radar #865: Last Week in AI: Last Week in AI: Karpathy, Google, Colossus, and the Coming IPO Wave — The AI frontier is rapidly evolving into a vertically integrated capital structure, where financial capacity and compute supply dictate market leadership.
• Why the rise of open source AI isn’t hurting Anthropic … yet — Frontier AI models drive discovery, while open-source models increasingly own production in a two-tiered AI economy.
• Krishna Rao - Anthropic's CFO on Compute, Scaling to $30B ARR, and the Returns to Frontier Intelligence - [Invest Like the Best, EP.472] — Anthropic's exponential growth stems from efficient, fungible compute allocation and continuous frontier model development, yielding high enterprise returns.
• it's all bad now... — The current US AI regulation risks creating a two-tiered access system, hindering progress, and fostering dangerous blind spots.
• AI to ROI News & Analysis: May 22, 2026 — The AI market is rapidly evolving with major players making significant financial and product moves, while regulatory efforts face resistance.
• 🔮 Exponential View #574: Inside Anthropic’s rocket ship; AI pluralism; love commoditized, context-maxxing & Voltaire++ — Frontier AI intelligence drives exponential enterprise value through enhanced capabilities and compute efficiency.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Epoch AI.