Chips in AI research

By Nathan Benaich (Air Street Capital)

Which GPUs researchers use, how adoption is changing, and which chips specialize in LLM research.

Questions in this topic

Which GPU is most widely used in AI research?How dominant is NVIDIA in AI research?Are H100 and H200 replacing A100 in AI research?Do researchers use consumer GPUs such as RTX 4090?Which AI chip startups are most cited in research?Which chips are associated with LLM research?

Questions and answers

2026 estimates · inputs through September 1

Which GPU is most widely used in AI research?

NVIDIA’s A100 is the most cited GPU in the index’s 2026 NVIDIA-chip series, with 15,327 citations. H100/H200 follows with 9,823, then RTX 4090 with 6,557.

The 2026 estimates combine counts through September 1 with projections for the second half of the year. These figures measure mentions of hardware used in papers, not sales, installed GPUs, or which chip performs best.

NVIDIA usage chart · Zeta Alpha and methodology

Read this answer

2026 estimates · inputs through September 1

Are H100 and H200 replacing A100 in AI research?

H100/H200 citations rise from 4,653 in 2025 to a projected 9,823 in 2026. A100 citations remain higher, at 15,266 and a projected 15,327, respectively.

Hopper is gaining visibility in published research, but these figures do not establish that researchers have replaced their A100 fleets. The 2026 values include second-half projections; the series also groups H100 and H200 together.

NVIDIA usage chart · Zeta Alpha and methodology

Read this answer

2026 estimates · inputs through September 1

Do researchers use consumer GPUs such as RTX 4090?

Yes. RTX 4090 appears 6,557 times and RTX 3090 4,521 times in the index’s projected 2026 NVIDIA-chip citations. Consumer GPUs remain visible alongside data-center accelerators in published AI research.

The 2026 estimates combine counts through September 1 with projections for the second half of the year. A citation does not reveal the number of GPUs used, workload size, or whether a particular card would suit your project. This is an adoption signal, not a hardware recommendation.

NVIDIA usage chart · Zeta Alpha and methodology

Read this answer

2026 estimates · inputs through September 1

Which AI chip startups are most cited in research?

Groq and Cerebras lead the six companies tracked in the startup chart, with projected 2026 citation counts of 264 and 242, respectively.

The 2026 estimates combine counts through September 1 with projections for the second half of the year. The comparison covers these six tracked companies, not every NVIDIA competitor, and measures research-paper citations rather than commercial adoption.

Startup usage chart · Zeta Alpha and methodology

Read this answer

Topic sample · January 1 to June 1, 2025

Which chips are associated with LLM research?

In the topic sample, LLM research is overrepresented among papers using AMD MI300, AMD MI250, Huawei Ascend, and NVIDIA H100/H200. Their LLM shares exceed the overall paper sample’s baseline by 42.5, 34.2, 33.9, and 27.8 percentage points, respectively.

This measures specialization, not which chip appears in the most LLM papers or runs them fastest. The analysis covers 6,356 papers, uses model-assigned topic labels, and requires at least three papers per chip-topic pair. It is a 2025 sample, separate from the 2026 citation estimates.

Research-topic chart · Original analysis

Read this answer

Charts and sources

Citation estimates use inputs through September 1, 2026; fleet counts have an October 1, 2026 cutoff. The research-topic sample covers January 1 to June 1, 2025. See each answer for its period and limitations.

All data sources