Google CEO Sundar Pichai says APIs now process nearly 32 billion tokens per minute
The figure roughly doubles an earlier Google Cloud benchmark and points to fast-rising demand for Gemini models
Google’s first-party model APIs now handle nearly 32 billion tokens every minute, according to CEO Sundar Pichai.
To decode the jargon: a token is a small chunk of text, often a word or a piece of one, that an AI model reads or writes. Every chatbot reply, code suggestion and document summary gets chopped into tokens before a model can do anything with it.
So token throughput works a bit like electricity usage for AI. The more tokens flowing through the meter, the more work customers are actually asking the models to do.
From 16 billion to nearly 32 billion
The new figure lands well above earlier benchmarks. Google Cloud had previously reported direct API usage of more than 16 billion tokens per minute as of April 2026.
That April number was itself a step up. Earlier reports put the figure at 10 billion tokens per minute in the prior quarter.
Put those three data points in a row and the curve is steep. Throughput went from 10 billion, to over 16 billion, to nearly 32 billion tokens per minute, roughly doubling from the April benchmark.
Pichai’s latest metric covers first-party model APIs specifically. Those are the interfaces developers and businesses use to plug Google’s own models straight into their apps and services.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The broader ecosystem is bigger still. Total token processing across Google’s AI footprint has reportedly reached multi-quadrillion volumes on a monthly basis.
Gemini is doing the heavy lifting
Much of the growth traces back to Google’s Gemini family of models. That includes the newer releases known as Gemini 4 Argon and Gemini 4 Carbon.
Pichai said the Argon model processed 25 trillion tokens within its first 72 hours of operation.
More than 8.5 million monthly active developers are building with Gemini models, according to the research findings.
Supporting all of this takes a lot of hardware. Google is investing up to $185 billion in capital expenditures for 2026 to keep pace with AI demand, according to estimates cited in the research.
Why big tech keeps publishing token counts
Token metrics have become a favorite stat for AI executives. They tend to show up during earnings calls and major conferences, where companies want to prove that demand for AI infrastructure is real and growing.
That said, token counts are not revenue. Pricing per token varies by model and customer, so a doubling in throughput does not automatically mean a doubling in sales.
Google has been using these figures to position Google Cloud as a serious contender in AI services. Its main rivals in that arena include OpenAI and Microsoft Azure, both of which are also racing to sign up developers and enterprise customers.
What this means for Google and the AI market
For investors, the headline number helps answer the question hanging over Google’s spending plans. Capital expenditure on the scale of up to $185 billion needs visible demand to justify it, and a near doubling in API throughput is the sort of evidence Google wants on the table.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

