State of AI Compute Index Refreshed, Shows Shifting GPU Land

State of AI Compute Index Refreshed, Shows Shifting GPU Landscape

@nathanbenaich· July 1, 2026 View original

Summary

The State of AI Compute Index has been updated to July 2026, revealing key trends in AI compute usage, researcher preferences, and cluster builds. The report details NVIDIA's product cycle, the rebound in chip citations, the status of challenger chips, and the significant deployment of Hopper GPUs.

The State of AI Compute Index has been updated, providing a comprehensive overview of the AI hardware landscape through July 2026. The report highlights a significant shift in NVIDIA's product cycle, with Hopper (H100/H200) GPU deployments more than doubling and becoming the dominant installed base, while the older A100 fleet is declining. Blackwell deployments are still nascent but growing rapidly. Despite challengers like AMD and Huawei making inroads, NVIDIA continues to dominate chip citations in AI papers, accounting for approximately 91% of mentions. However, the demand side shows major players like OpenAI diversifying their compute contracts, with a majority of their disclosed book now being non-NVIDIA. The Grace-Blackwell pipeline indicates massive future deployments, though only a small fraction is currently operational. The index also tracks startup silicon, noting Groq as the most-cited startup chip before its acquisition by NVIDIA. Overall, the data suggests a dynamic market with NVIDIA maintaining a strong lead, but with increasing diversification in compute procurement and significant future build-outs across various vendors.

Why it matters

This data is crucial for strategic planning, investment decisions, and understanding the competitive landscape in AI infrastructure, impacting hardware procurement, cloud strategy, and research directions.

How to implement this in your domain

1Review current compute infrastructure against industry trends to identify potential bottlenecks or opportunities.
2Evaluate diversification strategies for AI compute procurement, considering non-NVIDIA options for future projects.
3Forecast future compute needs based on projected AI model sizes and research directions.
4Engage with hardware vendors to understand their roadmap and secure access to next-generation GPUs.

Who benefits

TechCloud ComputingVenture CapitalResearch & DevelopmentSemiconductors

Key takeaways

NVIDIA's Hopper GPUs are now the dominant installed base, with Blackwell rapidly emerging.
NVIDIA maintains a strong lead in chip citations within AI research papers.
Major AI players like OpenAI are diversifying their compute contracts beyond a single vendor.
The Grace-Blackwell pipeline indicates massive future compute infrastructure build-outs.

Original post by @nathanbenaich

"we refreshed the @stateofai compute index with zeta alpha, now current to july 2026. 8 charts on where ai compute actually is, what researchers use in their work and the sizes of clusters being built now for a highlight thread! stateof(dot)ai/compute inside nvidia it is a product…"

View on X