Business•

AMD Shares Climb About 13% After Google Picks EPYC Over Its Own Axion Chips for the Confidential TPU 8tx

Google Cloud will run its new TPU 8tx confidential AI system on AMD's fifth-generation EPYC processors rather than its in-house Axion CPUs, and AMD stock rose roughly 13%.

Written By :MINRK
AMD Shares Climb About 13% After Google Picks EPYC Over Its Own Axion Chips for the Confidential TPU 8tx

Photo by Luis Gonzalez on Unsplash

1. AMD Google TPU deal at a glance

Advanced Micro Devices (AMD) said on November 9, 2026, that its fifth-generation EPYC server processors will act as the host CPUs inside Google Cloud's new TPU 8tx system, an AI accelerator platform built for confidential computing. AMD shares rose roughly 13% that day, as investors focused on Google choosing AMD silicon instead of its own Axion processors.

The AMD Google TPU arrangement is narrow in one sense and significant in another. AMD is not designing or manufacturing Google's Tensor Processing Units, which remain Google's own custom accelerators. What AMD supplies is the general-purpose processor that sits beside those accelerators, directs the work they do and, crucially for this system, helps keep the data they handle sealed off from prying eyes.

That distinction matters for understanding why the market reacted so strongly. Google has spent years building its own chips, including the Arm-based Axion CPU line, and a decision to reach outside that internal portfolio for a new flagship system sends a message about where AMD's technology currently stands in the security-sensitive corner of the AI market.

2. What is the Google TPU 8tx?

The TPU 8tx is a member of Google's newest family of AI accelerator systems, aimed primarily at inference and reinforcement learning. Inference is the stage at which a model that has already been trained is put to work answering questions, generating text or making predictions for users. Reinforcement learning is a training approach in which a model refines its behavior by trying actions, receiving feedback and adjusting accordingly.

Both workloads have become central to how large AI systems are deployed and improved. Inference in particular now accounts for a growing portion of the computing that cloud providers sell, because every query sent to a chatbot or AI agent has to be processed somewhere. Reinforcement learning, meanwhile, has become a key technique for sharpening models after their initial training run.

What sets the TPU 8tx apart is its emphasis on hardware-enforced confidential computing. Google has positioned it as the first of its cloud TPUs to offer that kind of protection, which is designed to shield data not only while it is stored or moving across a network but also while it is actively being processed by the chips.

3. How confidential computing works in the system

Traditional data security focuses on two states: data at rest, such as files sitting on a disk, and data in transit, such as information crossing a network. Encryption has long covered both. The weak point has historically been the moment of processing, when data must be decrypted so that a processor can work on it. During that window, a compromised host system, a malicious insider or a flaw in the software stack could in principle expose sensitive information.

Confidential computing aims to close that gap by using the processor's own hardware to create protected environments where data stays encrypted in memory and is isolated from the rest of the system, including the cloud operator's own management software. For companies handling medical records, financial data, proprietary models or other sensitive material, that third layer of protection can determine whether a workload moves to the public cloud at all.

In the TPU 8tx, AMD's contribution comes through its Infinity Guard security feature set, which is built directly into EPYC processors. A core component of that suite is Secure Encrypted Virtualization, or SEV, which encrypts the memory of virtual machines with keys managed by the processor itself. AMD says its EPYC CPU works together with the TPU so that this protection extends to the attached accelerators, helping keep sensitive data protected as it moves between the host and the AI chips.

It is worth being precise about the limits. Confidential computing reduces the attack surface and limits who can see data during processing, but it is not a guarantee against every conceivable security threat. Its value lies in making certain categories of exposure far harder, which is often what regulated customers need before they will commit a workload to shared infrastructure.

4. The role of the fifth-generation EPYC host CPU

AMD's fifth-generation EPYC chips are built on the design code-named Turin, which uses the company's Zen 5 processor architecture. These are general-purpose server CPUs, the same broad class of chip that runs databases, web services and enterprise software in data centers around the world.

In an accelerator system like the TPU 8tx, the host CPU acts as a coordinator. It handles workload orchestration, deciding which tasks go where and in what order. It prepares data so the accelerators receive it in a usable form. And it manages the high-speed communication that keeps the TPUs supplied with work. Accelerators are extremely fast at the specialized mathematics behind AI, but they depend on the host to keep them busy and to stitch their output back into a coherent result.

Because the host also anchors the security model, the choice of CPU in a confidential system is not a minor component decision. The processor's built-in protections set the baseline for how well data is guarded across the whole platform, which helps explain why Google's selection drew so much attention.

5. Why passing over Axion stands out

The most striking aspect of the announcement is what Google did not use. Axion is Google's in-house CPU, part of a broader effort by large cloud providers to design their own chips rather than rely entirely on outside suppliers. Building custom silicon lets those companies tailor performance to their own software, control costs and reduce dependence on vendors.

Against that backdrop, choosing AMD's EPYC for a new system that Google is promoting as a step forward in confidential AI suggests that AMD's security capabilities carried decisive weight. The reasoning, as described, centers on what AMD's processors can offer for confidential deployments, an area where its memory-encryption technology has been developed and refined across several product generations.

The decision does not necessarily mean Google is retreating from Axion. In-house chips and third-party processors often coexist in large cloud fleets, each deployed where it fits best. But a flagship confidential system running on AMD hardware is a visible endorsement, and it shows that even companies with deep internal chip programs will still look outside when a specific capability matters enough.

6. An expansion of an existing partnership

The TPU 8tx is not the first meeting point between the two companies. AMD's EPYC processors already power several families of Google Cloud virtual machines, the rentable computing instances that customers use to run their applications. AMD has also said that its SEV technology protects a range of Google Cloud confidential services, including Confidential VMs and Confidential GKE, Google's managed Kubernetes offering.

Seen in that light, the new system extends a relationship that was already built around AMD's security features. What is different is the setting. Moving from general-purpose virtual machines into the host role of an AI accelerator platform brings AMD closer to the center of the most closely watched spending category in the technology industry.

7. How the market reacted

Investors responded quickly to the November 9 announcement, sending AMD shares up roughly 13% in a single session. A move of that size is unusual for a company of AMD's scale and reflects how much weight the market places on signals of demand from the largest cloud providers.

The rally also came during a difficult stretch for semiconductor stocks more broadly, with sentiment toward the sector having cooled in recent months. That made the jump stand out further, since it suggested investors were willing to reward company-specific evidence of momentum even as enthusiasm for the group as a whole had faded.

Neither company has disclosed pricing, unit volumes or a general availability date for the TPU 8tx, so the financial contribution of the arrangement remains unclear. The reaction instead appears to reflect what the deal implies about AMD's competitive position rather than any specific revenue figure attached to it.

8. What the AMD Google TPU deal means for the industry

The broader lesson is that confidential computing is becoming a real selling point for cloud customers running sensitive AI workloads. As companies move from experimenting with AI to deploying it on proprietary and regulated data, questions about who can see that data during processing become harder to set aside. Hardware that can answer those questions convincingly has a commercial edge.

For AMD, Google's choice indicates that its security features can win business even when a customer has a capable in-house alternative available. That is a meaningful argument in a market where the largest buyers increasingly design their own chips and where third-party suppliers must show clear advantages to keep their place.

For the cloud providers themselves, the arrangement illustrates the practical balance between building and buying. Custom silicon offers control and efficiency, but specialized capabilities developed by outside vendors over many years can still be the deciding factor for particular systems.

9. What to watch next

Three questions will shape how significant the announcement proves to be. The first is whether AMD's data center business continues to build on its recent momentum in the quarters ahead, giving investors concrete evidence that wins like this one translate into sales.

The second is whether other large cloud operators follow Google's lead and turn to EPYC processors for their own confidential AI offerings. A pattern of similar decisions would strengthen the case that AMD holds a durable advantage in this niche.

The third concerns Google's own balancing act between Axion and AMD. If the TPU 8tx turns out to be a one-off, the strategic signal will fade. If it marks the start of a broader pattern, AMD's role inside Google Cloud could grow well beyond the virtual machine families it already supports.

Related Articles

NEWSLETTERS

Don't miss another story.

Subscribe to the MINRK Newsletter today.

Tick the newsletters you want. No selection? You'll get MINRK Headlines.

By signing up, you will receive emails about MINRK products and you agree to our terms of use and privacy policy.

MINRK Daybook

Market analysis for crypto traders and investors.

EVERY WEEKDAY

MINRK for Advisors

Defining crypto, digital assets and the future of finance for financial advisors.

EVERY THURSDAY

MINRK Protocol

Exploring the tech behind crypto one block at a time.

WEEKLY

MINRK Long & Short

A must read for institutions. Insights, news and analysis delivered weekly.

EVERY WEDNESDAY

MINRK Headlines

The biggest crypto news and ideas of the day.

EVERY WEEKDAY

MINRK Policy

Examining the intersection of cryptocurrency and government.

WEEKLY

MINRK Research

Join thousands of readers who rely on MINRK for data-driven insights on the latest digital asset trends.

MONTHLY