Google Splits Its TPU Line for Sensitive Data and Big Contexts
Google has introduced two versions of the tensor processing unit it designs itself, one built for customers who cannot let sensitive data sit exposed while it is processed and one built to hold very large amounts of information at once.
The chips are the 8tx and the 8i Max, both redesigned versions of existing processors and both tailored to specific customer needs rather than to a single general workload. Bloomberg reported the plan earlier in the day, and Alphabet shares were up more than 3 per cent in early trading.
And the reason the line exists at all is that Google now sells these chips rather than only using them. Its tensor processing units ran internal products such as Search, YouTube and Gemini for years, then became a cloud offering to win enterprise AI customers, and Alphabet said TPU sales made a meaningful contribution to results in its latest quarter as Google Cloud revenue more than doubled to USD 30.71 billion.
The 8tx is the part aimed at regulated industries. It targets government agencies and companies that must keep data protected while it is being processed, pairs Google's chip technology with AMD processors, encrypts data inside the hardware and delivers up to twice the speed of the previous version with virtually no loss of performance.
The 8i Max goes the other way, aimed at AI applications that need to work through very large amounts of information at once. It combines two chips into a single package for additional memory and computing power, which the company says suits coding assistants, mathematical reasoning and drug discovery.
But the competitive frame is Nvidia, and it is worth being precise about the size of the challenge. Google is selling a chip it designed for its own workloads into a market where Nvidia sets the price and the software standard, and a custom part that wins on encryption or memory capacity wins a segment rather than the market.
The encryption claim is the one with the clearest customer. Regulated industries have spent years being told that AI means moving data to someone else's servers, and encryption inside the hardware during processing answers a compliance question that a faster chip does not.
So the two variants read as one strategy. Google is chasing the workloads where a general purpose accelerator is the wrong shape instead of raw throughput, and the 8tx gives regulated buyers a reason to stay inside Google Cloud.
Google introduced two tensor processing unit variants, the 8tx and 8i Max
The 8tx targets government and regulated industries that need protected processing
The 8i Max packs two chips in one package for memory and compute
Alphabet said TPU sales contributed meaningfully to quarterly results
Google Cloud revenue more than doubled to USD 30.71 billion
Source: Reuters
