---
type: "post"
id: "new-ai-computing-capacity-is-doubling-every-9-d4c7c1"
slug: "new-ai-computing-capacity-is-doubling-every-9-d4c7c1"
title: "New AI computing capacity is doubling every 9 months."
description: "Epoch AI estimates that quarterly additions of AI-chip computing capacity grew 2.5× per year from Q1 2024 through Q2 2026, equivalent to a doubling about every 9.1 months."
retrieval_nugget: "Epoch AI estimates that quarterly additions of AI-chip computing capacity grew 2.5× per year from Q1 2024 through Q2 2026, equivalent to a doubling about every 9.1 months. The estimate gives builders a concrete infrastructure trend, but it measures chips sold rather than compute already operating in data centers."
published_at: "2026-11-11"
updated_at: "2026-11-11"
record_date: "2026-11-11"
date_kind: "published_at"
topics: ["google"]
entities: ["Epoch AI","Nvidia","Google","AMD","Amazon","Huawei","Cambricon"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "Epoch AI reports that global AI computing capacity additions are doubling every nine months, with quarterly growth running about 2.5× per year since 2024, down from roughly 6× in 2023."
prior_refs: []
source_url: "https://epoch.ai/data-insights/ai-computing-capacity-additions-doubling"
source_urls: ["https://epoch.ai/data-insights/ai-computing-capacity-additions-doubling"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:new-ai-computing-capacity-is-doubling-every-9-d4c7c1"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/new-ai-computing-capacity-is-doubling-every-9-d4c7c1/","markdown":"https://newruntime.com/posts/new-ai-computing-capacity-is-doubling-every-9-d4c7c1.md","json":"https://newruntime.com/posts/new-ai-computing-capacity-is-doubling-every-9-d4c7c1.json"}
markdown_url: "https://newruntime.com/posts/new-ai-computing-capacity-is-doubling-every-9-d4c7c1.md"
---

# New AI computing capacity is doubling every 9 months.

## Retrieval answer

Epoch AI estimates that quarterly additions of AI-chip computing capacity grew 2.5× per year from Q1 2024 through Q2 2026, equivalent to a doubling about every 9.1 months. The estimate gives builders a concrete infrastructure trend, but it measures chips sold rather than compute already operating in data centers.

## Summary
Epoch AI estimates that quarterly additions of AI-chip computing capacity grew 2.5× per year from Q1 2024 through Q2 2026, equivalent to a doubling about every 9.1 months. The estimate gives builders a concrete infrastructure trend, but it measures chips sold rather than compute already operating in data centers.

## What happened
Epoch AI’s October 29, 2026 Data Insight, by Isabel Juniewicz and Josh You, estimates the computing capacity of AI chips sold worldwide each quarter. The headline finding is that additions have grown about 2.5× per year since 2024, down from roughly 6× in 2023. Epoch says the earlier surge followed the initial releases of ChatGPT and GPT-4 and was led by Nvidia’s H100 GPU.

The analysis covers Nvidia, Google, AMD, Amazon, Huawei and Cambricon. It expresses capacity in H100-equivalents (H100e), using quarterly chip-sales estimates from Epoch’s AI Chip Sales dataset. That dataset estimates units sold by chip model from company financial disclosures, analyst estimates and media reports. Epoch used the dataset’s median estimates as of October 19, 2026. The reported quarterly total rose from 827,914 H100e in Q1 2024 to 6,327,416 in Q2 2026. The series includes 4,980,154 H100e in Q1 2026.

To estimate the rate, Epoch fit a log-linear regression to quarterly capacity additions across ten quarters, Q1 2024 through Q2 2026. The fitted annual growth rate is 2.5×, with a 90% confidence interval of 2.3× to 2.7×; the corresponding doubling time is 9.1 months, with an interval of 8.4 to 10.0 months. Epoch analyzes additions rather than cumulative capacity because a surge followed by a slowdown can make the two growth rates diverge. It says fitting cumulative capacity across 2024–2026 would overstate the current pace.

Epoch observes that year-over-year growth in quarterly additions fell from 2.8× in Q2 2025 to 2.2× in Q2 2026. A regression of quarterly growth rates indicates a downward trend, from about 32% to 19% per quarter, but it is not statistically significant (p = 0.26). The analysis notes that quarter-to-quarter growth is noisy and can reflect chip-generation transitions, including Nvidia’s Blackwell rollout in late 2024 and early 2025. Epoch says Nvidia had begun shipping Rubin chips but had not recognized Rubin sales as of its latest fiscal quarter ending in July.

## What the archive adds
New Runtime’s earlier coverage introduced Epoch AI’s Inference Performance Index, which compared chips by modeled concurrent inference capacity rather than raw FLOP/s alone ([November 10, 2026 note](https://newruntime.com/posts/epoch-ai-launched-the-inference-performance-index-scoring-1bb9b1/)). This is a new fact on the broader story of measuring AI compute: the new analysis tracks the pace of global capacity additions, not per-chip inference performance.

## Trend signal
This supports the infrastructure context behind New Runtime’s earlier coverage of compute, supply chains and AI infrastructure: scaling model capacity depends on more than model releases. The result is not direct evidence for a specific agent-economics pattern, so this does not yet fit a pattern New Runtime tracks.

## Why now
Epoch’s comparison makes the change in pace visible: the estimated growth rate for quarterly additions is far below the 2023 surge, while still implying a rapid doubling cadence. It also cautions that a generational lull can resemble a slowdown, and that sales do not map immediately to operational data-center capacity.

## Why it matters
For teams building AI products, capacity growth is relevant context for availability, competition among compute buyers and the economics of serving models. A faster flow of hardware sales can eventually affect the supply behind model APIs, cloud capacity and self-hosted deployments. It may also shape the competitive environment in which providers decide where to place new capacity and which workloads receive priority.

But the headline should not be read as a measure of usable capacity arriving every nine months. Chips are counted at sale, and Epoch says they can take several months to become operational. Nor does a global growth estimate tell an individual team what its provider can allocate, at what price, under which quota, or in which region. The series is therefore most useful as a directional supply-side indicator.

Teams should still base architecture and procurement decisions on provider-specific capacity, latency, price, quota and regional availability. Epoch’s distinction between new quarterly additions and cumulative installed stock matters as well: using cumulative capacity to describe a current flow can make a growth trend appear faster than the current additions support. The practical planning question remains whether a team can secure and afford the capacity it needs for a particular workload.

## Limits and uncertainty
The estimate rests on ten quarters and median sales estimates for six chip designers. Epoch omits smaller designers including Intel, Meta, Microsoft, Cerebras and Groq, as well as chips not intended for data-center use. Epoch believes those omitted volumes are minor, but the analysis does not independently establish their exact effect on the trend.

Its confidence intervals describe fit around the median estimates, not uncertainty in the underlying sales data. Sales are not the same as chips coming online, and new generations ramp in waves. The apparent post-2024 slowdown is not statistically significant; Epoch also lacks pre-2024 sales estimates for the full set of six designers. The figures consequently describe an estimated global hardware-sales trajectory, not deployed capacity, utilization, power availability, or customer access.

## Reader takeaway
Epoch’s best estimate is that newly sold AI-chip capacity has doubled about every nine months since early 2024, slower than the 2023 surge. Builders can use it as broad infrastructure context, but should not mistake sold capacity for operational compute or a guarantee about their own access.

## Open question
Will the next chip-generation ramps restore faster quarterly additions, or will the post-2023 pace persist?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Google](https://newruntime.com/topics/google.md): Explore the google topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)