The computing power of Google’s TPU chips sold in the second quarter of 2026 was about 74% higher than in the first quarter, against about 17% for Nvidia’s, by our arithmetic from Epoch AI’s estimates published on 29 October. Epoch puts Google’s additions at 1.61 million H100-equivalents, up from 0.92 million, and Nvidia’s at 3.95 million, up from 3.39 million.
The research group’s headline finding is about the whole market. It says “Quarterly additions of computing capacity have grown about 2.5ร per year since 2024”, and that this is “down from roughly 6ร in 2023”.
A vertical news Short narrated from the headlines and opening lines of this piece and three related pieces. It goes public at 12:00 UTC on 30 October.
Business Pill · Vertical Integration
A short explainer on vertical integration: owning more of the stack you depend on. It teaches the general idea only and says nothing about any company in this piece.
The key insight: By our arithmetic from Epoch AI’s estimates, the computing power of Google’s TPUs sold in the second quarter of 2026 rose about 74% from the first quarter, about 4.4 times Nvidia’s 17% rise, lifting Google to about 25.4% of the compute added in the quarter.
Epoch AI estimates the computing capacity of AI chips sold each quarter by six designers: Nvidia, Google, AMD, Amazon, Huawei and Cambricon. It measures that capacity in H100-equivalents, a unit based on Nvidia’s H100 chip, using “financial reports, company disclosures, and more”, according to its post on X.
Epoch says “Quarterly compute capacity additions rose from 0.8 million H100e in Q1 2024 to 6.3 million in Q2 2026”. Its fitted trend is “2.5ร per year (90% CI: 2.3ร to 2.7ร), a doubling time of 9.1 months”.
By our arithmetic, a 6x yearly pace, as Epoch puts 2023, would mean a doubling time of about 4.6 months. Epoch says the 2023 surge followed the release of ChatGPT and GPT-4, “led by Nvidiaโs new H100 GPU”.
Computing capacity of AI chips sold per quarter by designer, in million H100-equivalents, first and second quarter of 2026, Epoch AI median estimates as of 19 October 2026; percentage changes are our arithmetic. Source: Epoch AI (CC-BY).
What Was Posted on X
The computing power of AI chips sold each quarter has grown 2.5ร per year since the start of 2024. Thatโs down from roughly 6ร in 2023, led by Nvidiaโs then-new H100 GPU.
Some evidence suggests growth slowed further. But quarterly sales are noisy, and our data ends just as Nvidia begins shipping its next-generation chips, a move that has historically spiked sales. This could be a lull between generations.
This is from our AI chip sales dataset, which uses financial reports, company disclosures, and more to estimate compute, power usage, and spending for a wide variety of chips: https://epoch.ai/data/ai-chip-sales
Epoch AI says the computing power of AI chips sold each quarter has grown 2.5x per year since the start of 2024, down from roughly 6x in 2023, and that recent data could reflect a lull between chip generations.
Google Gained Share of New Compute
Epoch’s table gives the second-quarter totals by designer. By our arithmetic, Google’s TPUs were about 25.4% of the computing power added in the quarter, up from 18.5% in the first quarter and 14.7% in the first quarter of 2024. Nvidia’s share was about 62.5%, against 68.0% in the first quarter and 69.2% two years earlier.
Nvidia still added far more. Its 3.95 million H100-equivalents were about 2.5 times Google’s 1.61 million in the second quarter, by our arithmetic. Amazon’s additions rose about 27% in the quarter, AMD’s about 14% and Huawei’s about 1%.
Epoch says year-over-year growth in quarterly additions “fell from 2.8ร in Q2 2025 to 2.2ร in Q2 2026”. It adds that a regression “shows a downward trend, from about 32% to 19% per quarter, but it is not statistically significant (p = 0.26)”.
The research group offers a reason to wait: the slowdown “may also reflect a lull between generations”. It says that “While Nvidia has recently started shipping its next-generation Rubin chips, it had not yet recognized any Rubin sales as of its latest fiscal quarter ending in July”.
As we read Epoch’s table, the market-wide 2.5x yearly pace hides a change in mix: Nvidia’s share of new compute was about 62.5% in the second quarter, against 69.2% in the first quarter of 2024, by our arithmetic.
By our arithmetic, total additions in the second quarter of 2026 were about 7.6 times those of the first quarter of 2024, using Epoch’s totals of 6.33 million and 0.83 million H100-equivalents.
As we read Epoch’s caveats, one quarter does not set a trend: it says its main estimate rests on ten quarters of data and that recent figures may reflect a lull before Nvidia’s Rubin chips.
Epoch AI, data insight, 29 October 2026
“Quarterly additions of computing capacity have grown about 2.5ร per year since 2024”
Three Implications
FOR NVIDIA Epoch says Nvidia had not yet recognized any Rubin sales as of its latest fiscal quarter ending in July, so its next-generation chips are not yet in these figures.
FOR GOOGLE Epoch’s estimates put Google’s TPU additions at 1.61 million H100-equivalents in the second quarter, up from 0.92 million in the first.
FOR CHIP SUPPLIERS Epoch says year-over-year growth in quarterly additions fell from 2.8x in the second quarter of 2025 to 2.2x a year later, and that a regression of quarterly growth rates shows a decline from about 32% to 19% per quarter that is not statistically significant.
The analysis puts it this way: “Three competitive categories within one layer: generalist substrate (NVIDIA), hyperscaler custom (TPU, Trainium, MTIA), merchant custom challengers (Cerebras, Groq, SambaNova).”
As we read it, Epoch’s table measures two of those categories side by side: Nvidia as the generalist supplier, and Google and Amazon as hyperscalers designing their own chips. The analysis does not discuss Epoch’s October 2026 estimates.
What Is Not Established
Epoch says its data “tracks chips when they are sold, not when they come online”, and that it omits smaller designers such as “Intel, Meta, Microsoft, Cerebras, and Groq”. It also says its main estimate “rests on only ten quarters of data”. The quarter-on-quarter changes, shares and ratios here are our arithmetic from Epoch’s table, not Epoch’s own statements.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Epoch AI estimates that new AI computing capacity sold each quarter has grown about 2.5 times a year since 2024, a doubling time of 9.1 months, down from about 6 times in 2023. By our arithmetic from its table, Google’s TPU additions rose about 74% in the second quarter of 2026, against about 17% for Nvidia’s, lifting Google to about 25.4% of new compute.
A note on sourcing. The chip-sales figures, growth rates and caveats come from Epoch AI’s data insight of 29 October 2026 and its AI Chip Sales hub (median estimates as of 19 October 2026, CC-BY), and from Epoch’s post on X. The quarter-on-quarter changes, shares and ratios are our arithmetic from Epoch’s table. We did not contact Epoch AI, Google or Nvidia.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
