Google broadens its AI hardware portfolio with two new specialized chips

Google on Monday unveiled two new variants of its tensor processing units (TPUs), extending its in-house chip lineup into confidential inference and larger-memory workloads as the company works to cut its reliance on externally made chips for its cloud computing business.

The two units, TPU 8tx and TPU 8i Max, belong to Google's eighth generation of TPUs, which was introduced in April and originally emphasized inference capacity and support for large training requirements. The new parts build on that generation rather than starting a new one, broadening the range of use cases the portfolio covers.

TPU 8i Max takes a different approach: it packages two of the inference-focused TPU 8i chips from April into a single unit, doubling memory capacity, processing power and bandwidth. The result is a chip that lets models consider a wider range of information during queries, addressing memory pressure for large workloads rather than adding new security features.

The launch lands amid a broader scramble for AI-specific semiconductors. Microsoft, Amazon and AI labs including OpenAI and Anthropic are all developing their own chips and supporting software to reduce dependence on Nvidia's dominance of the market. TechCrunch's report also flagged a personal controversy involving Nvidia CEO Jensen Huang disclosed over the weekend, arriving at what it characterized as a vulnerable moment for the company.

What remains open is how and when these chips reach customers: Google did not outline availability, pricing, or which cloud regions will carry them, nor how the confidential-training features will be billed relative to existing TPU tiers. Also unaddressed is whether TPU 8tx's on-chip reasoning loop will become a distinct product surface for distilled-model workflows or stay a feature inside the standard training stack, and how quickly rivals respond with comparable memory-doubling inference parts.