OpenAI Reportedly Moves From Renting Google TPUs to Buying Them

OpenAI plans to buy Google’s AI chips, The Information reported on Tuesday, citing a person familiar with the matter. 

The verb is the detail worth noticing. In June 2025, Reuters reported that OpenAI had begun renting Google’s chips, and a purchase is a bigger commitment, made with a company that competes with OpenAI in AI models. 

Nothing changes today for the ChatGPT users or developers. But the report shows where AI computing supply is heading, and Alphabet reports results later today.

What was reported, and what wasn’t

The report says the move would add Google’s tensor processing units (TPUs) to OpenAI’s computing resources as demand for chips to train and run models surges. 

TPUs have emerged as an alternative to Nvidia’s GPUs, which dominate the market. OpenAI is still among the largest buyers of those GPUs as flagships like GPT-6 Astra run on them.

Much is missing. Coverage of the report gives no price, chip count, TPU generation, delivery schedule, or workload. 

It is also unclear whether OpenAI would put the chips in its own data centres or reach them through partners such as Google Cloud or Oracle. Neither company had confirmed the plan, though.

How a rental became a purchase plan

The public record reveals a clear progression over time. 

In June 2025, as Reuters notes, OpenAI began renting Google chips, though OpenAI framed the effort as early testing rather than a large-scale rollout. 

By early 2026, Google started selling TPUs directly to select buyers beyond its standard cloud services. OpenAI then unveiled its own custom silicon developed with Broadcom, targeting a deployment late in the year. 

Now, OpenAI is moving to buy Google hardware outright while actively hiring specialized engineers to optimize its flagship models for TPU architecture. 

Competitor Anthropic has similarly scaled up TPU buys while hiring former Google TPU lead Amir Salek to drive its custom silicon efforts. 

What to watch, and what it means if you build on AI

With Alphabet reporting earnings today, Wall Street is watching closely to see if heavy AI investments are paying off. 

Major tech firms are pouring hundreds of billions into infrastructure, so any updates on TPU supply will carry weight. 

Key details still need clearing up; namely, whether OpenAI and Google publicly confirm the deal, how many chips are involved, and where they will run.   

For developers and engineering teams, this hardware shift changes OpenAI’s backend setup, not user pricing.

Still, it highlights a smart development practice: keep model calls hidden behind a clean internal adapter.

When teams avoid tying their code to a single vendor, switching models or spreading workloads across providers becomes a simple settings tweak instead of a massive rewrite.

Source: OpenAI, a Longtime Nvidia Customer, Plans to Buy Google’s AI Chips