OpenAI is turning to Google's AI chips in a move to challenge Nvidia.
OpenAI is turning to a rival to get ahead in the AI race. The ChatGPT maker is reportedly planning to buy AI chips from Google, a move that could add an unexpected twist to the race for computing power.
Google owns one of the few serious alternatives to Nvidia's dominant AI processors, and OpenAI's reported interest suggests that even the biggest AI players are looking beyond a single supplier.
According to a Bloomberg report, OpenAI intends to purchase Google's chips, although the final terms remain unclear. The deal could have significant implications for Nvidia's dominance and the broader race for computing power.
**Why OpenAI wants Google's AI chips**
Building powerful AI models takes enormous computing resources. Running them for millions of users adds another layer of demand, making access to advanced chips a critical part of the business.
OpenAI is therefore looking to diversify its AI infrastructure, which includes the chips, data centres, networks and software needed to train and run its models. Relying on multiple suppliers could give the company more flexibility as demand grows and competition for computing capacity intensifies.
The move also puts Google in an interesting position. The two companies compete in AI products, but Google could also become a supplier to one of its rivals.
**Google's TPUs offer an alternative to Nvidia**
Google's AI processors, known as tensor processing units (TPUs), were developed more than a decade ago to accelerate the company's own computing workloads. They have since become an important part of its AI infrastructure and an alternative to Nvidia's graphics processing units (GPUs).
Nvidia remains a dominant force in AI computing, supported by widespread use of its GPUs across cloud providers, startups and large technology companies. However, the cost and availability of advanced chips have pushed businesses to explore alternatives.
Google is also expanding access to its processors beyond its own infrastructure. Oracle has announced plans to offer Google TPUs through its cloud platform, giving customers another route to access the hardware.
**OpenAI is building chips of its own**
Buying from Google is not OpenAI's only hardware strategy. The company has also been working on its own AI chips, developed with semiconductor company Broadcom.
According to the report, these in-house chips, called Jalapeno, are expected to support OpenAI's AI models later this year. Richard Ho, who leads OpenAI's internal chip team, previously worked on Google's semiconductor efforts, including TPUs.
That connection highlights how competition for specialised chip talent is becoming part of the wider AI race. Companies are not only competing to build better models. They are also investing in the hardware needed to train and operate them.
**Nvidia may face more competition, not an overnight replacement**
OpenAI's reported interest in Google's chips does not mean it is abandoning Nvidia. Instead, it points to a more diversified approach to securing computing capacity, managing costs and supporting growing AI workloads.
OpenAI could unlock another source of computing power while its in-house chip programme develops. Google, meanwhile, could turn its TPUs into a bigger business by supplying a company it also competes with in the AI market.
The AI race is increasingly being fought on two fronts: building smarter models and securing the hardware to run them. Sometimes, that means buying chips from a company you are also trying to beat.