Google rolls out specialized AI chips to take on NVIDIA
What's the story
Google is releasing two new versions of its tensor processing units (TPUs), custom-made semiconductors designed to develop and run artificial intelligence (AI) software. The move is part of a broader strategy by Alphabet, Google's parent company, to take on NVIDIA in the multi-billion dollar AI chip market. The new TPUs will enable confidential computing and allow models to access more information while answering queries.
Product evolution
TPU family expansion
Google's chip unit has sped up its product development cycle, launching two new AI processors this year instead of the usual one every two years.
The latest additions to the TPU family are part of this accelerated pace.
Google chip chief Amin Vahdat said in an interview that "the future is going to be about specialization, and the upsides for specialization continue to be significant."
Market strategy
TPU models reconfigured to address specific customer requests
The new TPU 8tx and 8i Max models are not entirely new silicon designs, but rather reconfigured versions to tackle tasks that customers have been asking for.
The 8tx is focused on secure, confidential work for government agencies and companies in regulated industries, while the 8i Max is designed to run AI applications with long-context windows.
This capability is essential for drug discovery, math calculations, and coding applications.
Competitive edge
TPU sales projections
TPUs are competing in a market dominated by NVIDIA's graphics processing units (GPUs).
Google started developing this technology over a decade ago with Broadcom, and has now begun selling them outside its cloud-computing operation.
According to Bloomberg Intelligence, by 2028, TPU sales will exceed $135 billion.
This is up from an estimated $12 billion this year and doesn't include sales from clients using the chips through Google Cloud.
Clientele expansion