Google spins up two new TPU variants in challenge to Nvidia

Google is rolling out two fresh spins on its custom tensor processing units, according to Bloomberg News, which reported the news Monday. The variants are dubbed TPU 8tx and 8i Max. Alphabet shares climbed more than 2 per cent in premarket trading on the report, though the stock is still down 9 per cent so far this year.

These aren't brand-new silicon. Google's chief technologist for AI infrastructure, Amin Vahdat, told Bloomberg the two chips are reconfigured versions built around workloads customers keep asking for. The 8tx targets secure, confidential computing for government agencies and companies in regulated industries, while the 8i Max is aimed at AI applications that need long context windows.

Google's homegrown chips were once reserved for internal use. They have since turned into a genuine challenger to Nvidia's pricey graphics processors, which dominate AI workloads worldwide. Alphabet did not immediately respond to a Reuters request for comment, and Reuters said it could not independently verify the Bloomberg report.

The report lands as Google leans harder into its own AI hardware stack, putting its in-house design team in direct conversation with the same customers weighing Nvidia's GPUs.