# Epoch AI reports rising AI chip performance per dollar [Canonical reading](https://escapemeridian.com/readings/epoch-ai-reports-rising-ai-chip-performance-per-dollar-ecbd8fd4) Epoch AI published this estimate on August 13, 2026. It estimates that the performance per dollar of AI chips purchased each quarter rose by an average of about 49% per year, measured in constant 2025 dollars. The measure counts bit-operations per second for each dollar spent, averaged across the chips purchased in a quarter; it is not a claim that every chip became equally cost-effective. Epoch AI attributes most of the gains to newer chip generations delivering more performance, even as chip prices rose. Its comparison of NVIDIA’s GB300 with the P100 illustrates the difference: the newer chip costs more but, in Epoch AI’s estimate, delivers far more performance. The practical point is that higher chip prices do not necessarily mean less performance for each dollar spent. The estimate is an average across purchases, not the highest available performance per dollar. Epoch AI says most AI hardware spending goes to Nvidia’s GPUs, whose relatively high margins mean lower price-performance than some custom ASICs. The retrieved text refers readers to an assumptions and limitations section but does not include its details, so the evidence here does not show how those assumptions affect the estimate. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Epoch AI · Article Original source date: Aug 13, 2026 (2026-08-13T00:00:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-06T06:03:26.199Z. First observed: 2026-10-18T03:01:37.114Z. Retrieved: 2026-10-24T03:09:38.819Z. - [The performance per dollar of AI chips purchased each quarter has grown by an average of 49% per year](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Limits Text as retrieved on October 24, 2026; originally published August 13, 2026. The retrieved text refers to an assumptions and limitations section but does not include its contents, so its effects on the estimate cannot be assessed from this evidence. ## Open question How would the estimate change under the assumptions listed in Epoch AI’s full analysis? ## Source-stated Epoch AI estimates average annual growth of about 49% in performance per dollar for AI chips purchased each quarter. > Fitting an exponential trend gives an average growth rate of about 49% per year (90% CI: 36% to 66% per year). - [Supporting source](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Source-stated Epoch AI attributes most of the gains to newer chip generations becoming more powerful, with performance rising faster than prices. > The gains are largely due to chips getting exponentially more powerful with each new generation. While recent chips have become more expensive, they have grown more powerful at an even faster rate. - [Supporting source](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Source-stated Epoch AI says Nvidia’s GPUs account for most AI hardware spending and have lower price-performance than some custom ASICs because of their relatively high margins. > Most AI hardware spending currently goes towards Nvidia’s GPUs, which command a relatively high margin and thus have a lower price-perf than some custom ASICs. - [Supporting source](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Source-stated Epoch AI’s estimate describes a quarterly purchase average, which is lower than the highest performance per dollar available among chips. > While chips are available with performance per dollar as high as 3.6 × 1012 bit-operations per second per dollar (Google’s TPU v6e), the average across each quarter’s purchases is lower. - [Supporting source](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Source and contributor context Epoch AI · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Graphics processing unit A chip that performs many calculations at once; the main hardware used to train and run AI models. - [Definition reference](https://en.wikipedia.org/wiki/Graphics_processing_unit) ## Epoch AI Epoch AI studies trends in AI capability, training resources and the infrastructure behind them. - [Explore this context](/entities/epoch-ai) - [Reference](https://epoch.ai/) - [Reference](https://epoch.ai/data) ## NVIDIA NVIDIA develops processors, networking and software used for AI, graphics and scientific computing. - [Explore this context](/entities/nvidia) - [Reference](https://blogs.nvidia.com/) ## Editorial provenance Version 1. Publication ID: fc3ce42f-95f3-aa66-6404-8782cdc42f43. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-06T05:58:57.562Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 36d702d6f66a029c51983c9b4865bdae4f57c62b605e7c4cacd816932bd0d479. Source text hash: 467fda697e928492d98d1d55241e962fdb26a6d08c64c936d7568622c8086b82. ## Revision and source-check history ### Version 1 — 2026-11-06T06:03:26.199Z Epoch AI published this estimate on August 13, 2026. It estimates that the performance per dollar of AI chips purchased each quarter rose by an average of about 49% per year, measured in constant 2025 dollars. The measure counts bit-operations per second for each dollar spent, averaged across the chips purchased in a quarter; it is not a claim that every chip became equally cost-effective. Epoch AI attributes most of the gains to newer chip generations delivering more performance, even as chip prices rose. Its comparison of NVIDIA’s GB300 with the P100 illustrates the difference: the newer chip costs more but, in Epoch AI’s estimate, delivers far more performance. The practical point is that higher chip prices do not necessarily mean less performance for each dollar spent. The estimate is an average across purchases, not the highest available performance per dollar. Epoch AI says most AI hardware spending goes to Nvidia’s GPUs, whose relatively high margins mean lower price-performance than some custom ASICs. The retrieved text refers readers to an assumptions and limitations section but does not include its details, so the evidence here does not show how those assumptions affect the estimate. ## Original source Epoch AI · Article Original source date: Aug 13, 2026 (2026-08-13T00:00:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-11-06T06:03:26.199Z. First observed: 2026-10-18T03:01:37.114Z. Retrieved: 2026-10-24T03:09:38.819Z. - [The performance per dollar of AI chips purchased each quarter has grown by an average of 49% per year](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Limits Text as retrieved on October 24, 2026; originally published August 13, 2026. The retrieved text refers to an assumptions and limitations section but does not include its contents, so its effects on the estimate cannot be assessed from this evidence. ## Open question How would the estimate change under the assumptions listed in Epoch AI’s full analysis? ## Source-stated Epoch AI estimates average annual growth of about 49% in performance per dollar for AI chips purchased each quarter. > Fitting an exponential trend gives an average growth rate of about 49% per year (90% CI: 36% to 66% per year). - [Supporting source](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Source-stated Epoch AI attributes most of the gains to newer chip generations becoming more powerful, with performance rising faster than prices. > The gains are largely due to chips getting exponentially more powerful with each new generation. While recent chips have become more expensive, they have grown more powerful at an even faster rate. - [Supporting source](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Source-stated Epoch AI says Nvidia’s GPUs account for most AI hardware spending and have lower price-performance than some custom ASICs because of their relatively high margins. > Most AI hardware spending currently goes towards Nvidia’s GPUs, which command a relatively high margin and thus have a lower price-perf than some custom ASICs. - [Supporting source](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Source-stated Epoch AI’s estimate describes a quarterly purchase average, which is lower than the highest performance per dollar available among chips. > While chips are available with performance per dollar as high as 3.6 × 1012 bit-operations per second per dollar (Google’s TPU v6e), the average across each quarter’s purchases is lower. - [Supporting source](https://epoch.ai/data-insights/chip-performance-per-dollar) ## Source and contributor context Epoch AI · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Graphics processing unit A chip that performs many calculations at once; the main hardware used to train and run AI models. - [Definition reference](https://en.wikipedia.org/wiki/Graphics_processing_unit) ## Epoch AI Epoch AI studies trends in AI capability, training resources and the infrastructure behind them. - [Explore this context](/entities/epoch-ai) - [Reference](https://epoch.ai/) - [Reference](https://epoch.ai/data) ## NVIDIA NVIDIA develops processors, networking and software used for AI, graphics and scientific computing. - [Explore this context](/entities/nvidia) - [Reference](https://blogs.nvidia.com/) ## Editorial provenance Version 1. Publication ID: fc3ce42f-95f3-aa66-6404-8782cdc42f43. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-06T05:58:57.562Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 36d702d6f66a029c51983c9b4865bdae4f57c62b605e7c4cacd816932bd0d479. Source text hash: 467fda697e928492d98d1d55241e962fdb26a6d08c64c936d7568622c8086b82. ### passed — 2026-11-06T05:58:57.562Z — anthropic/claude-opus-5.5 Publication: fc3ce42f-95f3-aa66-6404-8782cdc42f43 attribution — pass: The page is a single Epoch AI data insight with its own analysis. The reading consistently attributes the estimate, the explanation of its causes and the Nvidia margin point to Epoch AI. > We estimate the aggregate performance per dollar of the AI chips purchased in each quarter. summary — pass: The reading leads with the actor and the finding: about 49% average annual growth in performance per dollar. It explains that the metric is a quarterly purchase average rather than the best chip available, and it conveys the cause Epoch AI gives. GPU is defined on the page. 'Custom ASICs' is not explained, but it appears only in a secondary comparison and does not block understanding of the central point. The question follows from the stated gap about the assumptions section. > Fitting an exponential trend gives an average growth rate of about 49% per year (90% CI: 36% to 66% per year). date — pass: The reading states August 13, 2026, which matches the UTC day of originalPublished and the CSV update date on the page. It contains no hindsight framing and relies on no later material. The related-insight listing is dated earlier and is not used. The evidence is a publisher page, so no archive disclosure is needed. > CSV, Updated Aug. 13, 2026 support — pass: Each claim matches its excerpt. The GB300-versus-P100 comparison and the statement that higher prices do not necessarily mean less performance per dollar follow directly from the source. The reading correctly notes that the details of the assumptions and limitations section are absent from the retrieved text. No unsupported benefits or costs are implied. > While recent chips have become more expensive, they have grown more powerful at an even faster rate. media — pass: No media is selected, and no verified media candidates were supplied, so leaving the media field empty is appropriate. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/ecbd8fd4433e5cf471f2259b8e9c47a4869091babefead57f62f32079aa4674b/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/epoch-ai-reports-rising-ai-chip-performance-per-dollar-ecbd8fd4.json)