AMD and Google Cloud have expanded their collaboration, with 5th Gen EPYC processors providing the host infrastructure and security foundation for Google Cloud’s new TPU 8tx system.
There’s a quiet but meaningful shift happening in how cloud companies handle sensitive AI workloads — and AMD and Google Cloud are at the centre of it. On 9 November 2026, AMD announced that its 5th Gen EPYC processors power the host infrastructure for Google Cloud TPU 8tx, a new system Google describes as the first cloud Tensor Processing Unit platform to deliver hardware-enforced confidential computing.
That’s a fairly technical claim, but the practical point is straightforward: organisations running AI on this system get protection for their data and AI model weights not just when that information is stored or sent across a network, but while it’s actually being processed. That’s the bit that’s historically been harder to lock down.
What TPU 8tx Actually Does
Google Cloud’s TPU 8tx is built on the same silicon as the TPU 8t, which is designed for large-scale AI training. The 8tx variant pairs that silicon with 5th Generation AMD EPYC CPU hosts and adds a confidential-computing layer directly into the hardware. As Google Cloud put it, the system is “built on the same silicon as TPU 8t and paired with 5th Generation AMD EPYC CPU hosts, TPU 8tx builds security directly into the hardware layer, keeping sensitive data and model weights protected and isolated even while in active use.”
So this isn’t a software patch bolted on afterwards. The protection is baked in at the chip level.
AMD says its EPYC processors handle workload orchestration, data preparation, and high-throughput communication between workloads and the TPU accelerators themselves. Think of EPYC as the engine room — managing the flow of work — while the TPUs handle the heavy computational lifting for AI tasks.
Where AMD’s Security Technology Fits In
AMD brings its Infinity Guard suite to the arrangement, and specifically its Secure Encrypted Virtualisation capabilities. Combined with Google Cloud’s TPU confidential-computing architecture, the aim is to keep sensitive data and AI model weights protected even from the cloud infrastructure itself during processing.
AMD’s own announcement puts it plainly: “5th Gen AMD EPYC processors power the host infrastructure for Google Cloud TPU 8tx, helping enterprises protect sensitive AI models.”
And AMD also describes TPU 8tx as “the first cloud Tensor Processing Unit (TPU) system to deliver hardware-enforced confidential computing.” That’s a notable claim in a market where cloud providers are competing hard on AI infrastructure credentials.
The Workloads This Is Built For
The collaboration is aimed at three specific categories of AI work: inference, fine-tuning, and reinforcement learning. These aren’t abstract use cases. Inference is what happens when an AI model actually answers a question or makes a prediction. Fine-tuning is the process of adapting a general model to a specific task or dataset. And reinforcement learning involves training AI systems through feedback and reward signals — a method behind some of the most capable AI systems currently in use.
For industries handling sensitive data — healthcare, finance, legal services, defence — the ability to run these workloads with hardware-enforced confidentiality could change the calculus around what’s safe to put in the cloud.
But it’s worth being clear about what hasn’t been disclosed. AMD and Google Cloud haven’t published detailed performance benchmarks for TPU 8tx, nor have they announced pricing, customer numbers, or deployment volumes. The announcements also don’t explain why Google selected AMD EPYC hosts specifically for this system over other options. And as with any security claim, the real-world effectiveness of confidential computing depends on implementation, access controls, governance, and independent assurance — not just the hardware specification.
Google Cloud’s Broader TPU Family
TPU 8tx sits alongside two other systems in Google Cloud’s current TPU 8 lineup. The TPU 8t is intended for large-scale training workloads, while the TPU 8i is aimed at real-time reasoning. The 8tx variant carves out a specific niche: sensitive AI workloads where data protection during processing is a hard requirement rather than a nice-to-have.
The AMD and Google Cloud partnership on EPYC processors isn’t new — this announcement expands an existing collaboration. But adding hardware-enforced confidential computing to a cloud TPU system represents a step forward in how the two companies are positioning their joint infrastructure for regulated industries and security-conscious enterprise customers.
What happens next will likely depend on how quickly regulated sectors adopt the system and whether independent security researchers validate the confidential-computing claims in practice.
What This Means for Kent Residents
There’s no direct Kent footprint here — no local data centre, no regional jobs announcement, and no named partnership with Kent public bodies. But organisations across the county, from NHS Kent and Medway ICB to Kent County Council and local businesses handling sensitive information, could eventually access Google Cloud AI services built on this infrastructure remotely, subject to pricing, data-governance requirements, and procurement rules. For any Kent organisation already weighing up cloud AI services, the arrival of hardware-enforced confidential computing is the kind of feature worth asking a supplier about — chiefly where patient records, financial data, or other sensitive information is involved.
Source: @AMD
AMD EPYC Processors Power Google Cloud's First Confidential AI Computing System, TPU 8tx Quiz
5 questions