Google is preparing two specialised AI chips for launch next year: one designed to encrypt data at the hardware level, the other to give AI models more memory and computing power for longer context windows. The designs point to a more specialised contest for AI infrastructure, not simply another round in the GPU arms race.
Watch Desk analysis
What happened
Google chip division chief Amin Vahdat said the TPU 8tx and TPU 8i Max are being announced, according to TechCentral’s account of Bloomberg reporting. The 8tx is aimed at businesses in heavily regulated sectors and government agencies. Its integrated AMD processor enables hardware-level data encryption, avoiding the need to decrypt data in system memory during processing, the report says.
The TPU 8i Max targets workloads that need expanded context windows, the amount of information a model can process in one query. It combines two TPU chips to increase memory and computing power, with software engineering, mathematics and pharmaceutical research named as potential uses. Both products are expected to launch next year.
Why it matters
These are different answers to different constraints: keeping sensitive data encrypted during processing, and giving models room to handle larger inputs. That could make specialised accelerators more appealing to organisations whose workloads or rules do not fit a one-size-fits-all chip. The practical test, of course, is what customers can run on them and at what cost.
TechCentral also reports that Oracle, Meta and Apple are among Google’s TPU customers, and that regulatory filings show Anthropic has agreed to purchase up to $150 billion of TPUs between 2028 and 2031. Those are substantial reported commitments, but they do not establish how many chips will be delivered or what share of demand these new designs will serve.
Our read
The interesting shift is towards tailoring AI hardware to specific needs, rather than treating every workload as a reason to build a bigger general-purpose accelerator. Hardware encryption and larger context are concrete pitches, not yet proof of performance or adoption. For operators, the useful question is whether these chips offer enough security or capacity to justify adding another platform to an already complicated data centre.
What to watch
- Whether Google publishes detailed specifications, pricing and performance comparisons.
- When the chips become available and which cloud customers offer them.
- Whether the 8tx’s hardware encryption meets the requirements of regulated customers.
- How much of the reported TPU demand translates into delivered capacity.
Discussion spark: Would hardware-level encryption or the ability to process longer AI contexts be the more compelling reason for your organisation to choose a specialised accelerator?
Sources and evidence
- Source update (10 November 2026, 11:47 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.