Five Hyperscalers Now Own over Two-Thirds of Global AI Compute
Five hyperscalers—Google, Microsoft, Meta, Amazon, and Oracle—now command roughly two‑thirds of the world’s AI compute capacity, up from about 60% at the start of 2024. The share increase reflects continued investment in custom silicon and massive data‑center expansions. AI labs such as OpenAI and Anthropic depend almost entirely on these providers for the GPU and TPU power needed to train large models. The figures come from Epoch’s new AI Chip Owners datahub, which tracks hardware ownership across the industry.
Google Controls the Most AI Computing Power, Driven by Its Custom TPUs
Google now controls about 25% of all AI compute sold since 2022, with roughly 75% of that capacity delivered by its custom‑designed Tensor Processing Units (TPUs). This makes Google the single largest owner of AI‑focused hardware, while other hyperscalers continue...
The Epoch Brief - March 2026
Epoch AI’s March 2026 brief highlights three new Data Insights, including a 4.1× annual rise in AI chip memory bandwidth now at 70 million TB/s, and reveals that advanced packaging and high‑bandwidth memory, not logic dies, constrained chip production in 2025. A...
Total AI Chip Memory Bandwidth Has Grown 4.1x per Year, Now Reaching 70 Million TB/S
AI chip memory bandwidth has accelerated to 70 million terabytes per second, representing a 4.1× annual growth since 2022. This capacity dwarfs global internet traffic by a factor of roughly 300,000, highlighting the massive data movement required for modern inference. The...
What Do Frontier AI Companies' Job Postings Reveal About Their Plans?
Analysis of job postings at leading AI labs reveals a sharp shift toward go‑to‑market roles. At OpenAI and Anthropic, sales‑related positions have risen to roughly 30% of all openings, while research hires have fallen below 10%. The postings also expose...
Final Training Runs Account for a Minority of R&D Compute Spending
The analysis shows that final training runs represent only a minority of AI R&D compute spending. Across OpenAI, MiniMax and Z.ai, final runs account for 9.6%, 22.6% and 12.3% of total compute respectively. OpenAI’s 2024 R&D compute bill was about...
First AI Solution on FrontierMath: Open Problems
A team led by Kevin Barreto and Liam Price coaxed GPT‑5.4 Pro into solving a Ramsey‑hypergraph conjecture that has been open since a 2019 paper by Will Brian and Paul Larson. The solution marks the first AI‑generated answer on the...
Microsoft’s Recent $68 Billion in Physical Assets Additions Were Driven by AI-Related Purchases
Microsoft reported a $68 billion increase in property, plant and equipment during the second half of 2025, nearly matching the total addition recorded in the previous full fiscal year. The bulk of the spend—57%—went to IT hardware such as GPUs and...
