Market Size and Investment

The LLM market itself is still measured in the low tens of billions, but the money flowing around it, from enterprise spending to corporate investment, is an order of magnitude larger.

1. 01

   $7.77 billion was the estimated size of the global large language model market in 2025, projected to grow to $10.57 billion in 2026, per Precedence Research.

   Source: Precedence Research, 2025

2. 02

   $149.89 billion is the forecast size of the LLM market by 2035, implying a compound annual growth rate of about 34 percent.

   Source: Precedence Research (GlobeNewswire), 2026

3. 03

   $581.7 billion in global corporate AI investment flowed in 2025, up 130 percent from the prior year, per the Stanford AI Index.

   Source: Stanford AI Index, 2025

4. 04

   $285.9 billion in private AI investment came from the United States in 2025, about 23 times China's $12.4 billion.

   Source: Stanford AI Index, 2025

5. 05

   $37 billion was spent by enterprises on generative AI in 2025, 3.2 times the $11.5 billion spent in 2024, according to Menlo Ventures.

   Source: Menlo Ventures, 2025

6. 06

   $12.5 billion of 2025 enterprise generative AI spending went to foundation model APIs, the largest single infrastructure category.

   Source: Menlo Ventures, 2025

Model Landscape: Open and Closed

The United States still produces the most notable models, but Chinese open-weight models now dominate downloads. In paid enterprise workloads, closed frontier models are winning share back from open source.

1. 07

   50 notable AI models came from the United States in 2025, versus 30 from China and just 2 from Europe, per the Stanford AI Index.

   Source: IEEE Spectrum (Stanford AI Index), 2025

2. 08

   More than 2 million public models were hosted on Hugging Face in 2025, alongside over 500,000 public datasets and 13 million users.

   Source: Hugging Face, 2025

3. 09

   41 percent of Hugging Face downloads over the past year went to Chinese models, and China passed the United States in monthly downloads in 2025.

   Source: Hugging Face, 2025

4. 10

   1 billion downloads of Meta's Llama models were recorded by March 2025, up from 650 million in December 2024.

   Source: Meta, 2025

5. 11

   11 percent of enterprise LLM usage ran on open-source models in 2025, down from 19 percent in 2024, as closed frontier models pulled ahead on capability.

   Source: Menlo Ventures, 2025

Training Costs, Compute, and Energy

The cost of building frontier models keeps climbing even as the cost of using them collapses. Energy has become the binding constraint on the industry's growth.

1. 12

   2.4x per year is how fast frontier AI training costs are growing, with the largest models projected to cost over $1 billion to train by 2027, per Epoch AI.

   Source: Epoch AI, 2024

2. 13

   3.3x per year is the growth rate of global AI compute capacity since 2022, with Nvidia hardware providing over 60 percent of it.

   Source: IEEE Spectrum (Stanford AI Index), 2026

3. 14

   72,816 tons of CO2 equivalent were generated training xAI's Grok 4, versus an estimated 5,184 tons for GPT-4 and 8,930 tons for Llama 3.1 405B.

   Source: Stanford AI Index, 2025

4. 15

   29.6 GW of AI data center power capacity was online as of the 2026 AI Index, about what it takes to power New York State at peak demand.

   Source: Stanford AI Index, 2026

5. 16

   945 TWh is the projected global data center electricity consumption by 2030, roughly double 2024 levels and just under 3 percent of world electricity, with AI the biggest driver, per the IEA.

   Source: IEA, 2025

Token Prices and Benchmark Progress

Using an LLM of a given capability gets about ten times cheaper every year, while the frontier keeps advancing on benchmarks that were designed to be hard.

1. 17

   280-fold is how much the inference cost of GPT-3.5-level performance fell between late 2022 and late 2024, per the Stanford AI Index.

   Source: Stanford AI Index, 2024

2. 18

   $60 per million tokens was the cost of GPT-3-quality output in November 2021, falling to $0.06 per million tokens three years later, a 1,000x drop that works out to roughly 10x per year, per a16z.

   Source: a16z, 2024

3. 19

   62-fold is how much prices fell for GPT-4-level performance between March 2023 and late 2024.

   Source: a16z, 2024

4. 20

   77.3 percent was the AI success rate on real-world tasks measured in the 2026 AI Index, up from 20 percent a year earlier.

   Source: Stanford AI Index, 2026

5. 21

   50 percent accuracy on Humanity's Last Exam was surpassed by frontier models in April 2026, a benchmark built specifically to resist saturation.

   Source: IEEE Spectrum (Stanford AI Index), 2026

Enterprise Adoption and API Usage

Nearly every large organization now uses AI somewhere, and the enterprise LLM market has a clear leader: Anthropic, powered by demand for coding models.

1. 22

   88 percent of organizations use AI in at least one business function, up from 78 percent a year earlier, per McKinsey's late-2025 global survey of nearly 2,000 respondents.

   Source: McKinsey, 2025

2. 23

   40 percent of enterprise LLM API spending went to Anthropic in 2025, ahead of OpenAI at 27 percent and Google at 21 percent, with the top three controlling 88 percent of usage, per Menlo Ventures.

   Source: Menlo Ventures, 2025

3. 24

   54 percent of enterprise LLM spending on coding goes to Anthropic models, versus 21 percent for OpenAI.

   Source: Menlo Ventures, 2025

4. 25

   $47 billion was Anthropic's annualized revenue run rate in May 2026, up from $9 billion at the end of 2025, per company announcements.

   Source: Simon Willison (Anthropic announcements), 2026

5. 26

   3.2 quadrillion tokens per month were processed across Google's AI systems as of May 2026, roughly 7 times the volume of a year earlier.

   Source: Google, 2026

Sources

Every statistic on this page links to the publication where the number appears. Data comes from 14 sources.

• Precedence Research: https://www.precedenceresearch.com/large-language-model-market
• Precedence Research (GlobeNewswire): https://www.globenewswire.com/news-release/2026/02/26/3245397/0/en/large-language-model-market-forecasted-to-reach-usd-149-89-billion-by-2035-driven-by-ai-automation-and-open-source-adoption.html
• Stanford AI Index: https://hai.stanford.edu/news/inside-the-ai-index-12-takeaways-from-the-2026-report
• Menlo Ventures: https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/
• IEEE Spectrum (Stanford AI Index): https://spectrum.ieee.org/state-of-ai-index-2026
• Hugging Face: https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026
• Meta: https://about.fb.com/news/2025/03/celebrating-1-billion-downloads-llama/
• Epoch AI: https://epoch.ai/blog/how-much-does-it-cost-to-train-frontier-ai-models
• IEA: https://www.iea.org/reports/energy-and-ai/executive-summary
• Stanford AI Index: https://hai.stanford.edu/ai-index/2026-ai-index-report
• a16z: https://a16z.com/llmflation-llm-inference-cost/
• McKinsey: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
• Simon Willison (Anthropic announcements): https://simonwillison.net/2026/May/29/anthropic/
• Google: https://blog.google/innovation-and-ai/sundar-pichai-io-2026/

Related Statistics