Epoch AI finds AI data center scale doubles every 7 months, driven by Google, Microsoft, and Amazon investments. This accelerates beyond the earlier 12-month cycle, raising training cost projections to $10 billion by 2028.
Epoch AI reports AI data center scale doubles every 7 months. The finding underscores a structural shift in infrastructure economics outpacing Moore's Law.
Key facts
- AI data center scale doubles every 7 months.
- Google committed $11B/year to SpaceX compute in June 2026.
- Google booked Intel to package 3 million TPUs by 2028.
- Training costs for frontier models exceed $500 million per run.
- Data centers now consume 500 MW or more.
Epoch AI reports that the scale of AI data centers is doubling every 7 months According to Epoch AI. This growth rate exceeds the earlier 12-month doubling cycle seen through 2023, reflecting an acceleration in compute investment for training frontier models.
Key Takeaways
- Epoch AI finds AI data center scale doubles every 7 months, driven by Google, Microsoft, and Amazon investments.
- This accelerates beyond the earlier 12-month cycle, raising training cost projections to $10 billion by 2028.
What's Driving the Acceleration
Google alone has committed $11 billion per year to SpaceX compute as of June 2026, and booked Intel to package 3 million TPUs by 2028 [per recent history]. Microsoft and Amazon are similarly scaling their GPU clusters, with Microsoft's Azure investing in clusters of over 100,000 H100-equivalent GPUs. The trend is not just about more chips—it's about the density of interconnects and power infrastructure, with data centers now consuming 500 MW or more.
Implications for AI Economics
Epoch AI's finding implies that training costs for frontier models, which already exceed $500 million per run, could reach $10 billion by 2028 if the trend holds. This favors incumbents like Google and Microsoft with deep capital access, while startups face a rising barrier to entry. The 7-month doubling also suggests that inference demand may follow a similar trajectory, as agentic systems and real-time applications proliferate.
Counterpoint and Context
Hacker News commentary notes the finding but offers no counterarguments—the community appears to accept the trend [Hacker News]. However, the metric may conflate planned capacity with actual utilization; Epoch AI's methodology likely relies on announced buildouts, which could overstate real growth if utilization rates lag. Still, the capital commitments from Google, Microsoft, and Amazon suggest that the trend is real.
What to watch: The Q3 2026 earnings calls from Google, Microsoft, and Amazon for capital expenditure guidance. If combined AI infrastructure capex surpasses $100 billion annually, the 7-month doubling will be validated. Conversely, any slowdown in orders from Nvidia or AMD would signal a pullback.
Source: news.google.com
Originally published on gentic.news
Top comments (0)