
Share
Two new eighth-gen TPUs push Google's custom silicon into fresh territory: hardware-level privacy for training sensitive models and doubled memory for data-hungry inference, as the race to dethrone Nvidia heats up.
Google just gave its chip lineup two new specialists. On Monday, the company unveiled TPU 8tx and TPU 8i Max, a pair of tensor processing units built on top of its eighth-generation chip family that debuted back in April. These aren't incremental speed bumps. They're targeted answers to two problems that keep showing up in enterprise AI deployments: how do you train on sensitive data without exposing it, and how do you handle workloads that need more memory than a single chip can offer.
Tensor processing units are Google's homegrown silicon, custom-built for AI workloads instead of general computing. The company has leaned on them for years inside its own data centers, and increasingly sells access to them through Google Cloud as an alternative to renting Nvidia GPUs. That alternative pitch is central to why these two chips exist. Google, like every other hyperscaler, is trying to cut how much it depends on outside chipmakers for AI infrastructure.
TPU 8tx is the more conceptually interesting of the two. Built in partnership with AMD and based on the earlier TPU 8t chip, it's designed specifically for confidential training and inference. Google says the chip "builds security directly into the hardware layer, keeping sensitive data and model weights protected and isolated even while in active use."
That's a meaningful distinction. Most data protection schemes cover data at rest or in transit, encrypted on disk or encrypted over the wire. Protecting data while it's actively being computed on is a harder problem, and it's the kind of guarantee enterprises in regulated industries, think healthcare, finance, or defense, tend to ask for before they'll put proprietary data anywhere near a third-party cloud.
TPU 8tx also ships with a custom inferencing network that lets it run reasoning alongside training. In practice, that means:
That last point is the real sell. A lot of enterprise AI projects stall out not because the modeling is hard but because legal and compliance teams won't sign off on moving sensitive data across systems. Hardware-level isolation is Google's attempt to remove that friction.

TPU 8i Max takes a simpler approach to a different problem. It's essentially two of the inference-focused TPU 8i chips from April, fused into a single package. That packaging doubles memory capacity, processing power, and bandwidth, letting models handle a wider range of context during a single query. If you've hit memory ceilings running long-context inference or serving models that need to hold a lot of state at once, this is Google's answer: don't redesign the chip, just give it more room to work.
The timing here isn't incidental. Microsoft rolled out a new AI inference chip back in January. Amazon has been selling its own AI chips more aggressively since June, explicitly positioning them as a challenge to Nvidia. Even AI labs are getting into silicon: OpenAI unveiled its first custom chip, built by Broadcom, in June, and Anthropic has been building out a chip design team since August. Everyone with enough capital and enough workload volume is trying to own more of their hardware stack.
Nvidia, for its part, is dealing with a distraction that has nothing to do with silicon roadmaps. CEO Jensen Huang admitted over the weekend to having an affair with someone U.S. officials believe may be a Chinese intelligence asset tied to Megaspeed, a Singapore-based AI infrastructure firm that has reportedly purchased $4.8 billion in Nvidia chips over the past two years and faced prior scrutiny over whether it illegally routed chips to China. It's a scandal with real national security implications given how central Nvidia chips are to the broader U.S.-China AI competition, and it's landed at a moment when Nvidia's hardware rivals are shipping new products on a near-monthly cadence.
None of that changes the fundamentals of what Google shipped, though. The practical takeaway for teams evaluating cloud AI infrastructure:
Whether confidential computing on TPUs becomes a real differentiator or stays a niche feature depends on how many enterprises actually have data sensitive enough to need it. But the fact that Google built it at the hardware level, with a named partner in AMD, suggests the company sees real demand coming. Worth watching whether Microsoft or Amazon answer with confidential-computing chips of their own.
Tags
Original Sources
Google broadens its AI hardware portfolio with two new specialized chips | TechCrunch
↗ https://techcrunch.com/2026/11/10/google-broadens-its-ai-hardware-portfolio-with-two-new-specialized-chips
Ring, known for its doorbell cameras, launches a home ...
↗ https://techcrunch.com/2026/11/10/ring-known-for-its-doorbell-cameras-launches-a-home-security-drone-and-an-enterprise-security-robot
About the author
Kai built ML infrastructure at a Bay Area startup before developing an obsession with transformer architectures and inference optimisation that eventually pulled him out of product work entirely. A stint at a compute research lab sharpened his instinct for what actually matters in a model release versus what is marketing. He writes from the inside — from the perspective of someone who has debugged the systems he is describing at three in the morning. He is allergic to hype and instinctively drawn to the unglamorous plumbing questions that everyone else skips over.
More from The Engineer →This Week's Edition
11 November 2026
14 articles
Related Articles

Tether Evo's New Research Tackles BCI's Calibration Problem with Cross-Subject Models
Models & Research · 5 min

Why Some Engineers Think the Ocean Floor Is the Future of Data Centers
Environment & Climate · 5 min

Europe's Digital ID Push Tests a New Privacy Tool: Zero-Knowledge Proofs
Policy & Regulation · 5 min
Related Articles

Tether Evo's New Research Tackles BCI's Calibration Problem with Cross-Subject Models
Models & Research · 5 min

Why Some Engineers Think the Ocean Floor Is the Future of Data Centers
Environment & Climate · 5 min

Europe's Digital ID Push Tests a New Privacy Tool: Zero-Knowledge Proofs
Policy & Regulation · 5 min
More Stories
© 2026 Cedar & Bloom. All rights reserved.