Google has added two systems to its eighth-generation Tensor Processing Unit (TPU) portfolio, positioning them for what the company calls the "agentic era" of AI: the TPU 8i Max, designed for very large context windows and the highest compute density per rack, and the TPU 8tx, designed for secure and confidential inference.
Eighth-generation TPU · Google
Google splits its new TPU line into two systems for the "agentic era"
The TPU 8i Max targets very large context windows and the highest compute density per rack. The TPU 8tx targets secure and confidential inference. Google treats these as distinct hardware needs rather than one general-purpose chip.
TPU 8i Max
Long context, high density
Processes very large amounts of context at once
Built for AI agents reasoning over long documents, extended conversations, and multistep task histories
Company claims the highest compute density per rack in the lineup
TPU 8tx
Secure, confidential inference
Runs AI models on sensitive data
Designed to protect that data while it is being processed
Security claims are company statements until independently tested
Why the hardware is splitting: the shift to agentic AI
STEP 1
Single prompts
One question, one answer
→
STEP 2
Sequences of actions
Agents carry out multistep tasks
→
STEP 3
More context, more compute
Sustained demand on hardware
→
STEP 4
Two distinct needs
Density and long context vs. security
Positive reading
Matches hardware to distinct agent workloads
Targets sensitive and regulated data with confidential inference
Continues investment in custom silicon beyond general-purpose GPUs
Open questions
No detailed technical specifications or benchmarks yet
No pricing or availability timelines
No third-party evaluations of performance or efficiency
Bottom line
Adoption will depend on cost, performance per watt, ease of deployment, and how customers use confidential inference.
Compute density and security claims remain unverified
Further technical documentation or independent testing is needed
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...