Meta's AI Runs on Google's Silicon

Google's custom AI accelerators are now powering production workloads at Meta, marking a significant shift in the hardware landscape for large-scale artificial intelligence. The integration was made possible through a new version of the SGLang inference engine, which is now compatible with PyTorch and has been announced on Google’s developer blog. This development enables Meta to utilize Google's tensor processing units (TPUs) for its AI operations, a move that has been in development for some time.

The core of this update is the capability of the hardware to function natively within the PyTorch framework. Previously, running TPUs required the use of Google's JAX AI framework. The new software versions, including a matching open-source version of vLLM, are designed to allow such operations largely without requiring extensive code rewriting. This technical shift is intended to simplify the process for developers who are already working within the PyTorch ecosystem, which is widely used across the industry.

“SGLang-TorchTPU is running in production at Meta today,” stated Zongwei Zhou, Meta’s PyTorch-TPU lead. He described the development as the outcome of “a close three-way collaboration between Meta, Google, and the SGLang team.” The engine, he said, provides Meta with “a remarkably mature, PyTorch-native serving stack.” Jana van Greunen, director of PyTorch engineering at Meta, also commented on the new capability, stating that Meta “cannot be constrained by a single hardware architecture” and that TorchTPU “enables us to deploy our PyTorch workloads to Google TPUs.” Google has confirmed that it co-developed TorchTPU with Meta’s PyTorch team.

This is not the first time Meta has attempted to utilize Google's TPUs. Google has described the older PyTorch/XLA bridge to TPUs as a “predecessor” to the new TorchTPU. According to a report from SemiAnalysis, Meta research teams used TPUs through this bridge from 2020 to 2023 before Meta canceled those contracts. Google's PyTorch-on-TPU push was aimed mainly at Meta, which did not want to migrate to JAX. It remains unclear where Meta’s production TPUs are physically located. Reports from last year indicated that discussions included Google Cloud TPU rentals in 2026 with purchases in 2027, and Meta later reportedly signed a multi-year rental agreement for such services.

The announcement came one day after Amazon said it was open-sourcing TorchNeuron, its own native PyTorch backend for its Trainium2 and Trainium3 chips, at the same PyTorch Conference. Both moves appear to be aimed at loosening Nvidia’s software lock-in. The timing of these announcements highlights a growing effort among major cloud providers to offer alternatives to Nvidia's hardware and software stack for AI workloads.

Performance claims regarding the new integration are based on specific testing scenarios. A tuned vLLM setup serving Moonshot AI’s open-weight, 2.8-trillion-parameter Kimi K3 model reportedly tops 700 tokens per second for a single user on the seventh-generation Ironwood TPU. That speed figure comes from Inferact, the startup founded by vLLM’s creators that helped Google build vLLM-TorchTPU. Its test ran 16 Ironwood chips with speculative decoding, a method of speeding up large language model inference with a small draft model, against vLLM’s published setup on 16 Nvidia GB200s, which reached 452 tokens per second. Its kernels for Kimi K3 and Qwen 3.8 27B deliver roughly 1.4 to 2 times the GB200 baseline’s throughput at small batch sizes without speculative decoding.

The SGLang-TorchTPU project will continue to expand its model support, beginning with DeepSeek V4.1 Flash and Reflection Beam. RadixArk has stated that it is working toward providing Day 0 TPU support for new open models, meaning support on release day, by the first quarter of 2027. Google, Meta, and RadixArk plan to provide further details at the SGLang Summit 2026, scheduled for Nov. 12-13 in San Francisco.