Google expands its TPU chip family: 8tx for sensitive data and 8i Max for giant context
Direct answer
On November 9, 2026, Google introduced two new systems in its eighth-generation TPU chip family: the TPU 8tx, built for confidential and secure inference and small-scale training, and the TPU 8i Max, for massive context windows and memory-intensive agent workloads. The announcement came from Amin Vahdat, Google's senior vice president and chief technologist for AI and infrastructure. With these two, the eighth-generation family now totals four systems in a single year: 8t for large-scale training, 8i for real-time reasoning, 8tx and 8i Max. Google published no prices or availability dates.
What each new system does
- TPU 8tx, the confidential one: it protects sensitive information and model weights while the AI processes them. It is the system for companies that need to run AI on regulated or confidential data without exposing it.
- TPU 8i Max, the memory one: it combines two chips for memory-intensive work and offers, according to Google, the highest compute density per rack. It is the system for agents that work with very long contexts and heavy documentation.
- The full family in one year: 8t to train large models, 8i for real-time responses, 8tx for secure inference and 8i Max for massive context. Four specialized systems in twelve months.
- Competition is tightening: Bloomberg reported that the expansion intensifies Google's competition with NVIDIA in AI infrastructure. Google builds its own chips and rents them on its cloud; NVIDIA sells its chips to everyone.
What it means for your business
- AI is specializing per task, just like the models: just as Anthropic launched one model for technical work and another for general use, Google launches different chips for confidential inference and giant context. Buying AI is picking the right piece for each job.
- Sensitive data stops being a reason not to use AI: confidential inference exists as a product. If your company held back on AI because of regulated data, the question is now which provider offers that safeguard and at what price.
- Giant context makes memory-hungry agents cheaper: agents that read full files, contracts or histories need huge context windows. The infrastructure for that is already sold as a product, and its price will fall as competition grows.
- Rent capacity in the cloud: these systems arrive through Google's cloud. For a small business, the decision is which cloud and which model to use.
Note: this article summarizes Google's November 9, 2026 announcement, published by Amin Vahdat, with coverage from Bloomberg and the November 10 AI news digests. Google published no prices or availability dates for the two new systems.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- Business AI spending cooled: what Ramp's numbers say
- Google launched the universal Gemini agent: the digital coworker with its own email
- Google is building its first Wyoming data center: $17 billion and 702 MW
Frequently asked questions
What are the TPU 8tx and TPU 8i Max?
Two new systems in Google's eighth-generation TPU chip family, announced November 9, 2026. The 8tx is for confidential and secure inference; the 8i Max is for huge context windows and memory-intensive agent workloads.
How many eighth-generation TPU systems are there now?
Four in a single year: 8t for large-scale training, 8i for real-time reasoning, 8tx for confidential inference and 8i Max for massive context.
Can I buy a TPU 8tx for my company?
Google published no prices or availability dates. TPUs are used through Google's cloud, so the practical route is renting capacity in the cloud.
Why does confidential inference matter?
Because it lets you run AI on sensitive or regulated data while protecting the information as it is processed. It is the missing piece for companies with delicate data to use AI without exposing it.
