OpenAI is quietly shifting parts of its workload onto Google’s custom TPU chips in what appears to be a calculated move to cut costs.

Nvidia shares dropped 0.6% on Monday.

According to a report from The Information, which cites a single anonymous source, the change is recent and motivated by the growing demands on OpenAI’s infrastructure, particularly following the launch of its new image generation platform earlier this year.

Until now, OpenAI has leaned heavily on Nvidia’s GPUs, courtesy of its deep partnerships with Microsoft and Oracle. Oracle has been especially keen to play hardware middleman, stockpiling one of the industry’s largest inventories of Nvidia GPUs. That stockpile has turned Oracle into a vital cog in OpenAI’s machine-learning pipeline.

But with Nvidia’s hardware prices climbing and supply staying stubbornly tight, even the most GPU-hungry companies are being forced to explore alternatives. Enter Google, which is  pitching its TPU chips to third-party cloud infrastructure providers. The aim is to create a parallel market that challenges Nvidia’s current dominance, especially in high-performance AI training and inference.

If Google can tempt more providers into its TPU ecosystem, it could slowly siphon off chunks of Nvidia’s market share. That’s no small feat, given Nvidia’s current status as the de facto standard for training large language models and other generative AI systems.

Google is in a rare position. It not only builds AI models but designs the hardware to run them. The firm uses TPUs to train its own Gemini AI system, which is now being slathered across everything from Gmail to Search. That dual capability makes Google one of the few tech giants that can offer an end-to-end AI stack.

Should OpenAI increasingly depend on Google’s chips, it would mark the first meaningful disruption of Nvidia’s tight hold on the market. The shift also signals a more fragmented hardware landscape for AI, where no single chipmaker can rest easy.

OpenAI’s apparent change in direction comes just months after it partnered with Oracle for the Stargate programme, a high-profile alliance that looked like a strategic snub to Microsoft, its long-time ally. Now, by entertaining Google’s silicon, OpenAI appears willing to look beyond historical allegiances in order to stay competitive and manage its bottom line.

The move could set the tone for other AI companies eyeing alternatives to Nvidia. If Google's TPUs prove viable at scale, the long-standing dominance of Nvidia in the AI sector may finally be facing a serious contender, driven not by preference but by financial pressure and supply chain reality.