Google has introduced two new AI chips, the TPU 8tx and TPU 8i Max, due to ship next year. The notable shift is the breadth of the line-up: Google is pitching one chip at regulated workloads and another at AI tasks that need long context windows.
Watch Desk analysis
What happened
The 8tx is designed to keep data encrypted for government, banking and other regulated customers, and is paired with processors from AMD. The 8i Max puts two chips in one package for requests involving long context windows, where a model must work with more information at once. The two additions join the 8t and 8i announced in April, giving Google four TPU systems introduced in a single year.
Google’s chip push comes with ambitious forecasts. Bloomberg Intelligence, as cited by Yahoo Finance, projects TPU sales rising from about $12 billion this year to more than $135 billion by 2028, excluding chips rented through Google Cloud. That is a forecast, not booked sales. Yahoo Finance also cites Epoch AI estimates that TPU shipments doubled in computing power in one quarter, reaching roughly 800,000 chips.
Yahoo Finance’s report says Anthropic is a major customer, with Meta, OpenAI and Oracle also buying Google chips. It cites a SemiAnalysis finding that Google’s current Ironwood TPU can deliver nearly three times as much AI output per dollar as Nvidia’s Blackwell chips on agent workloads, while Nvidia’s top system remains faster when raw speed is the priority.
Why it matters
More specialised chips give cloud customers another option beyond buying or renting Nvidia hardware. Google is making a case for different jobs and different trade-offs: encryption for sensitive workloads, long-context capacity for demanding requests, and price efficiency for AI agents. That could matter to customers weighing cost and performance, not just to chipmakers watching market share.
Our read
This is a meaningful expansion of Google’s AI hardware menu, rather than a claim that one chip beats Nvidia at everything. The reported price-performance advantage is workload-specific, and the sales projections are forecasts. Still, four systems in a year and named customers suggest the TPU business is becoming a serious alternative worth comparing, rather than a footnote to Google Cloud.
What to watch
- Whether the 8tx and 8i Max ship on schedule next year.
- Whether Google or customers publish comparable performance and cost results for specific workloads.
- Whether the sales forecasts translate into sustained demand beyond Google Cloud.
Discussion spark: For AI workloads, would you prioritise a cheaper chip tailored to the task, or the fastest hardware available even at a higher cost?
Sources and evidence
- Google’s chip business is on a faster track than Nvidia’s own AI boom – Yahoo Finance (9 November 2026, 13:54 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.