k
keynews.ai
keynews.ai
ChipsThe story, in brief

Our eighth generation TPUs: two chips for the agentic era

Google just shipped two new TPUs built for agents. Here's what that means for inference costs.

Reporting by Hacker News 1 min brief
Paper-cut illustration of an amber microchip with circuit paths extending into a row of data-center cabinets.
The infrastructure powering AI.AI illustration by KeyNews
The KeyNews take

Why it matters

Google's eighth-generation TPUs represent a major infrastructure play designed explicitly for agentic workloads—signaling that the compute hierarchy is shifting from training-centric to agent-execution-centric silicon.

The key facts

5 to know

  1. Eighth-generation TPUs announced

  2. Two-chip design for agentic era

  3. Published April 22, 2026

  4. Google Cloud infrastructure announcement

  5. Positioning for agent inference workloads

Go to the source

Hacker Newsblog.google

Publisher excerpt: Article URL: Comments URL: Points: 13 # Comments: 5
Read original report
Back to today's editionMore chips news

Keep reading

Related stories

More from Chips
Chips

Nvidia and CoreWeave tackle the CPU bottleneck in agentic AI infrastructure

Nvidia and CoreWeave are solving agentic AI's CPU bottleneck—the execution layer between model decisions that GPU-only stacks can't handle.

SiliconAngleChips

Sleepy European telcos turn to data centre craze

European telcos are finally pivoting from stagnant connectivity to data-centre capex — but the real question is whether they can compete with hyperscalers in the AI buildout.

Financial Times TechnologyChips

Transforming the data path: Dell’s new releases move enterprises closer to the agentic data center

Dell extends its AI Data Platform with three new orchestration capabilities to manage agentic workloads — but the post provides no pricing, availability date, or measured performance impact.

SiliconAngle