Tech & AI
Google Adds Two Specialised TPU Chips as It Widens Challenge to Nvidia, Bloomberg Reports
Alphabet shares rose about 2.3% in premarket trading after the report of secure and long-context variants of its in-house AI chips.
HedgeWeekly Tech & AI Desk ·
Google is introducing two new versions of its tensor processing units (TPUs, the in-house chips it designs to train and run AI software), Bloomberg reported on Monday. The TPU 8tx is aimed at confidential work for government agencies and regulated industries, and pairs with an AMD processor to encrypt data at the chip level. The TPU 8i Max puts two TPU chips in one package to handle long-context tasks, where a model must process very large inputs in a single request, such as coding or drug discovery.
Google's chip chief, Amin Vahdat, told Bloomberg the designs are reworked versions of this year's chips, and that the variants should be available next year. That takes the eighth-generation family to four chips, built to different jobs. Google rents TPUs through its cloud business rather than selling them outright; Apple, Meta Platforms and Oracle are among named customers, and Oracle has said it plans to offer them to its own cloud clients from next year. Alphabet rose about 2.3% in premarket trading, GuruFocus reported.
Demand already exceeds supply: Google said in its latest earnings report that every TPU it has available is contracted. Bloomberg Intelligence estimates TPU sales, including related hardware and software, will pass $135 billion by 2028, from about $12 billion this year, and says the chips cost roughly 30% less than Nvidia's GPUs while competing on throughput. It also notes TPU margins are thinner than Nvidia's. Broadcom helps design and coordinate production of the chips.
Why it matters: each specialised chip gives large buyers another reason to split spending between Nvidia and Google, which bears on Nvidia's pricing power and on the order books of Broadcom and other design partners. The limits are real: memory and manufacturing shortages cap how quickly Google can ship, and the new parts are not due until next year. Investors should watch Alphabet's capital spending guidance, any new TPU customer announcements, and whether Nvidia answers with its own specialised parts.