Google just showed two new AI chips. Headlines say it’s a big swing at Nvidia.
It’s a real move. But the details are more specific, and smaller, than most summaries make it sound. Let’s go through what’s actually known.
Quick answer
What did Google unveil? Two specialized versions of its eighth-generation TPUs: the TPU 8tx and the TPU 8i Max. Bloomberg reported it first, on November 9, 2026.
What are they for? The 8tx is for confidential computing. The 8i Max is for long-context AI work.
Are they brand-new chips? No. They’re new configurations built on existing eighth-generation designs.
Is Nvidia’s dominance over? No. Nvidia is still the leader. Google is widening its challenge.
What is the TPU 8tx?
The TPU 8tx is built for people who can’t risk their data leaking, like government agencies and heavily regulated industries.
It uses the same silicon as Google’s TPU 8t, paired with custom AMD processors. The idea is hardware-level protection that isolates data and AI model parameters while they’re being used. Normally, data has to be decrypted in memory to be processed.
Google claims the 8tx delivers up to twice the speed of the 8t on some operations, with almost no performance penalty from the security. That’s Google’s claim, not independently tested. Some commenters have also questioned what customers can verify about the software and hardware inside a protected environment.
What is the TPU 8i Max?
The TPU 8i Max puts two TPU 8i processors in one package. That doubles compute, on-chip memory and high-bandwidth memory.
It’s meant for AI that handles huge context. Think entire codebases or hours of video. Google lists coding assistants, mathematical reasoning, scientific research and drug discovery as uses.
No performance figures were published for it.
Where do these fit in Google’s lineup?
The two new systems bring the eighth-generation family to four products:
- TPU 8t: large-scale training
- TPU 8i: real-time reasoning and inference
- TPU 8tx: confidential computing
- TPU 8i Max: long-context workloads
Google’s chip chief, Amin Vahdat, has framed the strategy as specialization. In his words, specialisation will play a crucial role in shaping the future of AI computing.
Does this match what you’ve read online?
Not fully. Some summaries describe “edge deployment, multi-modal training” and chips tuned for transformers and real-time agents. I found none of that in the reporting based on Bloomberg.
One article describes generic “inference” and “training” variants with big throughput and cost claims. That description doesn’t match the reporting on the actual chip names and purposes above. I’d treat those numbers with caution until Google publishes specs.
Is it cheaper than Nvidia?
Possibly. Market analysts estimate TPUs cost about 30% less than Nvidia’s popular GPUs, with competitive speed and lower power use, according to The Hans India. That’s an analyst estimate, not a Google price list.
Nobody has published pricing for the 8tx or 8i Max. And there’s no head-to-head test against Nvidia chips. The only speed claim is against Google’s own 8t.
Is it an escape from GPU shortages?
Here’s the twist. Google says its current TPUs are already committed to customers, and demand exceeds supply. Reports also say Google faces memory-component and manufacturing shortages. It’s reportedly exploring partnerships with Marvell and MediaTek to diversify supply.
So TPUs aren’t a magic way around hardware shortages. They’re another scarce chip, just from a different supplier.
When can you get them?
Unclear. One summary says Google plans to release the two chips next year, in 2027. Another report on the launch gives no availability date. None of the sources I read give pricing or a firm shipping date.
If you were planning to rent them next week, don’t. Check Google Cloud’s announcements.
Why does this matter?
Google Cloud revenue rose 103% year over year to $30.7 billion in Q3, per Anadolu’s report. Alphabet describes TPU system sales as a meaningful contributor. Alphabet shares rose more than 3% in early trading after the news.
Bloomberg Intelligence estimates TPU sales could climb from about $12 billion this year to more than $135 billion by 2028. That’s a projection, not a result.
The bigger picture is simple. Big buyers want more than one chip supplier. Google, AMD and others are giving them options. That helps price and supply over time.
What about Nvidia’s market share?
You’ll see “90%+” in some summaries. The sources I read don’t back an exact number. One says Nvidia holds over 80% of the AI accelerator market, another gives no figure. Estimates vary. Nvidia clearly still leads.
What we still don’t know
- Exact release and availability dates
- Pricing for both chips
- Performance numbers for the 8i Max
- How the 8tx security holds up under independent scrutiny
- Real-world comparisons with Nvidia hardware
FAQ
What are Google’s new AI chips?
The TPU 8tx, built for confidential computing, and the TPU 8i Max, built for long-context AI. Both are based on Google’s eighth-generation TPUs.
Are they completely new designs?
No. They are reconfigured versions of existing eighth-generation chips.
What does the TPU 8tx do?
It protects data and AI model parameters during processing, for government and regulated industries. It’s paired with custom AMD processors.
What is the TPU 8i Max?
It combines two TPU 8i processors in one package, doubling compute and memory for long-context workloads like big codebases.
Are Google’s TPUs cheaper than Nvidia GPUs?
Analysts estimate about 30% cheaper, but no pricing has been published for these chips.
Can I rent the new TPUs now?
Availability isn’t clear. One summary says 2027. No firm date was given in the sources I read.
Does this end Nvidia’s dominance?
No. Nvidia still leads AI accelerators. Google is expanding a real alternative.
Bottom line
Google isn’t dropping a “Nvidia killer.” It’s adding two focused chips: one for secure computing, one for huge context. That’s smart, and it widens the choices for big buyers.
But there are no prices, no independent tests and no clear ship date. Watch for those before you call this a shift.