AI Infrastructure Shift: Meta Activates Google Chips in Production

For businesses and developers relying on artificial intelligence tools, a major hardware option just became available. Meta, the company behind the popular PyTorch framework, has begun running AI workloads in production using Google’s custom chips. This move signals a practical change for users of these platforms, as the underlying infrastructure now supports a broader range of hardware without requiring extensive code rewrites.

The update centers on the release of new versions for the SGLang and vLLM inference engines. These tools are essential for running large language models efficiently. Previously, using Google’s tensor processing units, or TPUs, required developers to use Google’s JAX framework. The new versions, announced on Google’s developer blog, allow these engines to work natively with PyTorch. This development removes a significant technical barrier for the many engineers who prefer or depend on the PyTorch ecosystem.

Zongwei Zhou, Meta’s lead for PyTorch on TPUs, confirmed that the SGLang-TorchTPU version is already active in Meta’s production environment. He described the result as a remarkably mature serving stack that works directly within PyTorch. This integration was the product of a close three-way collaboration between Meta, Google, and the SGLang team. Zhou’s team operates at RadixArk, a startup founded by the creators of SGLang, which helped build the engine alongside the tech giants.

Jana van Greunen, director of PyTorch engineering at Meta, emphasized the strategic importance of this flexibility. She stated that the company cannot be constrained by a single hardware architecture. By enabling the deployment of PyTorch workloads on Google TPUs, Meta gains a critical alternative path. Google confirmed that it co-developed this TorchTPU technology with Meta’s PyTorch team to ensure smooth integration.

This is not the first time Meta has experimented with Google hardware. Google identifies PyTorch/XLA, an older connection method, as the predecessor to the new system. Reports from SemiAnalysis indicated that Meta research teams used TPUs through this older bridge between 2020 and 2023 before those contracts were canceled. The current push appears specifically tailored to accommodate Meta’s refusal to fully migrate to the JAX framework in the past.

While the exact location of these production chips is not public, industry talks last year suggested a timeline involving Google Cloud rentals in 2026 and potential purchases in 2027. Meta is also reported to have signed a multi-year rental agreement. This timing aligns with the current announcement, suggesting a long-term commitment to diversifying its hardware stack.

The news arrived just one day after Amazon open-sourced TorchNeuron, a similar backend for its Trainium chips. Both moves appear designed to loosen the software lock-in that has long benefited Nvidia. For the AI industry, this trend suggests a future where hardware choice is less about proprietary software ecosystems and more about performance and availability.

Performance metrics are already showing competitive results. A tuned vLLM setup on the seventh-generation Ironwood TPU reportedly exceeds 700 tokens per second for a single user. This test, conducted by Inferact, used 16 Ironwood chips with speculative decoding. It outperformed a similar setup on 16 Nvidia GB200s, which reached 452 tokens per second. The new kernels also deliver significantly higher throughput at small batch sizes compared to the Nvidia baseline.

The ecosystem is set to expand quickly. SGLang-TorchTPU plans to add support for models like DeepSeek V4.1 Flash and Reflection Beam soon. RadixArk aims to provide Day 0 support for new open models by the first quarter of 2027. Google, Meta, and RadixArk will share further details at the SGLang Summit in San Francisco on November 12 and 13.