AMD lands a key role in Google’s new TPU 8tx AI system

Semiconductor stocks have had a rough stretch. The PHLX Semiconductor Sector index fell 12% over the past three months, and AMD was not spared from the turbulence.

Then Google Cloud handed AMD a meaningful vote of confidence. On November 9, 2026, AMD announced that its 5th-generation EPYC processors will power Google Cloud’s new TPU 8tx system, and the stock surged approximately 13% in response.

What Google actually picked

The TPU 8tx is Google’s newest AI accelerator system, built for sensitive work like inference and reinforcement learning. Inference is the stage where a trained AI model answers questions. Reinforcement learning is how models improve through trial and feedback.

Google’s TPU 8tx uses hardware-enforced confidential computing. The data is protected not only at rest or in transit, but also while it is being processed.

AMD’s contribution is its Infinity Guard security feature set, built into the EPYC chips. Those protections are meant to keep data intact and shielded during computation.

The 5th-generation EPYC line is based on AMD’s Turin design, using the Zen 5 architecture. These are server CPUs, the general-purpose chips that run alongside Google’s specialized TPU accelerators and keep the whole system coordinated.

The Axion detail matters

The most interesting part of the deal is what Google did not choose. The TPU 8tx relies on AMD CPUs rather than Google’s own Axion CPUs.

For this particular system, Google reached past its own silicon and went with AMD. The stated reasoning centers on AMD’s security capabilities for confidential AI deployments, which marks a shift from Google’s prior reliance on Axion for this kind of setup.

The relationship is not new, either. AMD’s EPYC processors already run several Google Cloud virtual machine families, so the TPU 8tx deal extends a partnership that has been building for years rather than starting one from scratch.

The numbers behind the reaction

Investors responded quickly. AMD shares climbed approximately 13% after the announcement.

AMD’s own results tell a steadier story. The company reported $8.5 billion in data-center revenue for the quarter ending September 26, 2026, a sign that its core server business kept growing even as sentiment around the sector soured.

AMD’s EPYC chips are also deployed across more than 10 hyperscalers and AI labs.

What this means

For AMD, the clearest takeaway is positioning. Confidential computing is becoming a selling point for cloud customers handling sensitive AI workloads, and Google’s choice suggests AMD’s security features can win business even against in-house alternatives.

Investors watching AMD will want to track whether the data-center segment keeps building on that $8.5 billion quarterly figure, and whether other hyperscalers follow Google’s lead on confidential computing with EPYC.

The third is how Google balances Axion and AMD going forward. If the TPU 8tx turns out to be a one-off, the strategic signal weakens. If it becomes a pattern for secure AI deployments, AMD’s role inside Google Cloud could grow well beyond the virtual machine families it already powers.