On Monday, November 9, 2026, Alphabet (GOOGL) unit Google introduced TPU 8tx, a TPU system built for confidential inference, according to Google. The host processors running alongside it come from Advanced Micro Devices (AMD). That choice stands out because Google already designs and uses its own host chip.

When it unveiled TPU 8t and TPU 8i on Wednesday, April 22, 2026, Google said both run on its own Axion processor. TPU 8tx is built on the same silicon as TPU 8t, yet Google paired it with AMD’s fifth-generation EPYC processors.

Google said it built the system for regulated enterprises and governments running frontier or proprietary models. The company noted that “some highly regulated customers are asking for specialized encryption capabilities for unique workloads.”

Also read: AMD, Nvidia and Intel diverge as Morningstar resets the scorecard

That is the fear the hardware targets. Sensitive data and model weights can be exposed while a chip is processing them.

AMD’s edge here is a security track record

Confidential computing keeps data encrypted even while a processor works on it. Traditional encryption covers stored and moving data. Data being processed has been harder to protect.

AMD says its system extends that protection beyond the CPU to the connected accelerators. Prompts, model weights, and intermediate data stay encrypted in CPU memory and in transit, AMD said in a press release.

An accelerator’s identity and security configuration are verified before it can reach protected memory.

Neither announcement explains why TPU 8tx uses AMD rather than Axion, but AMD’s release points to its record inside Google. Its SEV technologies already help protect workloads across Google Cloud’s Confidential VMs, Confidential GKE and Confidential Space, among other services.

Google security experts have also conducted deep reviews of AMD’s security processor firmware, AMD said. That history means Google is not starting from zero on trust. Mark Lohmeyer, a Google Cloud vice president, said customers get “silicon-level data protection.” He said pairing EPYC with Google’s TPU design marks the first time cloud TPUs get hardware-enforced confidential computing.

Google paired its confidential TPU 8tx system with AMD’s fifth-generation EPYC processors, even though its TPU 8t runs on Google’s own Axion chip.

jetcityimage / Getty Images

Custom chips still need a trusted host

Google’s TPUs compete with AMD’s Instinct GPUs for AI workloads. TPU 8tx shows there is a second layer to that rivalry.

Accelerators do not work alone. AMD executive Dan McNamara said the strongest AI systems need a CPU that can keep pace with the accelerators it hosts.

In TPU 8tx, EPYC processors manage workload orchestration, data preparation, and high-throughput communication with the TPUs, according to AMD. AMD says EPYC is the host CPU of choice for more than 10 leading hyperscalers and frontier AI labs.

That gives AMD a way to earn from growth in custom chips even when its GPUs are not chosen. The catch is that the biggest buyers can design hosts too, as Axion shows. TPU 8tx suggests that, at least for confidential workloads, Google still sees value in an AMD host.

Monday’s AMD rally had more than one driver

AMD stock reached this news after a violent week. Shares fell 17.26% on Thursday, November 5, 2026, then rebounded 11.04% on Friday, November 6, 2026, according to StockAnalysis.

AMD closed at $508.58 on Monday, November 9, 2026. That is a 13.26% gain from Friday’s close of $449.04, based on StockAnalysis data. It also puts shares back above their $488.78 close on Wednesday, November 4, 2026, the day before the drop.

More Artificial Intelligence:

  • AI investments fail when companies ignore the hard part of transformation
  • Morgan Stanley issues a fresh warning about AI stocks
  • CrowdStrike gives companies a ‘kill switch’ as bots go off script

The Google news did not move the stock alone. Oppenheimer analyst Rick Schafer upgraded AMD to Outperform from Perform with a $600 price target, 24/7 Wall St said. Neither announcement included pricing, volumes, or revenue figures for TPU 8tx, so analysts have no new figures to model yet.

As of Friday, November 6, 2026, 53 analysts tracked by StockAnalysis rated AMD a consensus Strong Buy. Their average price target was $658.53. That average sits $2.28 below the stock’s 52-week high of $658.52.

Oppenheimer’s $600 target is below that average, making the upgrade less aggressive than it sounds. Eight analysts rated the stock a Hold, and none rated it a Sell, according to StockAnalysis. Recent calls on StockAnalysis include:

  • An $800 target came with Citi analyst Atif Malik’s Buy rating on Tuesday, October 6, 2026.
  • A $755 target came with Phillip Securities analyst Yik Ban Chong’s Buy rating on Friday, November 6, 2026.

Privacy is becoming a reason to pick a chip

AI chip contests are usually framed around speed and cost. TPU 8tx adds another test: whether a customer’s compliance team will sign off. That test favors suppliers with long security records, and those records take years to build.

The next signal is whether other cloud providers pair their own accelerators with AMD hosts for sensitive work. If Google later moves this system onto Axion, AMD’s window narrows. Until then, the most valuable chip in an AI system may not be the fastest one. It may be the one compliance teams already trust.

More Stocks News

  • Stock Market Today (Nov. 10, 2026): Nasdaq falls after key inflation report
  • Costco members are paying more for gas and filling up anyway
  • Disney’s next earnings come with an unusual $600 million boost
  • Elon Musk sends strong signal to Tesla stock investors