Billion-dollar AI chip deal between Google and Meta could be on the cards — would involve renting Google Cloud TPUs next year, outright purchases in 2027

Meta may be on the cusp of spending billions on Google AI chips to power its future developments, as the social-media giant is reportedly in talks to both buy and rent Google compute power for its future AI endeavours, as reported by The Information, via Reuters. The ongoing negotiations reportedly involve Meta renting Google Cloud Tensor Processing Units (TPU) in 2026, before purchasing them outright in 2027.

This news shows continuing collaboration between the companies, despite a recent pause on their undersea cable projects.

To date, Google has mostly leveraged its TPUs for its internal efforts, so this move, if it comes to fruition, would be a change of tactic that could help it capture a sizeable portion of the AI chip business. Considering that few, if any, companies have figured out how to turn a profit from developing AI just yet, Google may be looking to get in on Nvidia's act. The long-time GPU maker has made untold billions since the start of the AI craze, propelling it to become the world's most valuable company within a short timeframe.

Indeed, Reuters reports some Google Cloud executives believe that the shifting strategy would give it the chance to capture as much as a 10% slice of Nvidia's data center revenue. Considering Nvidia made over $51 billion from data centers in Q2 2025 alone, Google cornering that much of Nvidia's revenue would be worth 10s of billions of dollars.

Markets reacted to the rumors of this deal, sending Meta and Google stock upwards. Alphabet rose several percent in pre-market trading, and Reuters has it on track to become the next $4 trillion company potentially as soon as later today. Meta stock prices are up, too, but Nvidia took a 3% hit.

Even if Google does clinch this deal and secures a huge order and long-term revenue stream for its TPUs outside of internal use, it's still going to be swallowed up by the AI industry as a whole. There isn't enough compute power, fabrication capacity, or supply-chain logistics to provide the enormous uptick in demand for AI data center buildouts that have been ongoing this year.

Memory prices are skyrocketing, GPU prices are expected to jump up next year, and just about everything electronic could be more expensive this time next year.

That's if the bubble doesn't burst, of course. Even 2026 feels a long way off when it comes to this ever-changing industry, but 2027 is a lifetime away. Who knows what the state of AI hardware will be like then, and there's no telling whether Google's TPUs will have any longer shelf life than Nvidia's top GPUs. Especially with an aggressive annual release schedule.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.