GoKawiilMeta runs Google TPU inference in production via new PyTorch-native SGLang engine
Google and Meta jointly released a PyTorch-native version of the SGLang inference engine, called TorchTPU, letting Google's tensor processing units serve AI workloads without requiring a switch to Google's JAX framework. Meta's PyTorch-TPU lead Zongwei Zhou said the engine is already running in production at Meta, built with help from RadixArk, the startup founded by SGLang's creators. A matching PyTorch-native version of the vLLM inference engine was also released as open source.
GoKawiil's interpretation of the reporting above, not reported fact.
The move gives Meta a path to use Google TPUs for inference while staying within its own PyTorch ecosystem, reducing its dependency on Nvidia's CUDA-based hardware stack for at least some workloads. Meta previously used TPUs via an older bridge from 2020 to 2023 before canceling those contracts, suggesting this new arrangement reflects renewed, deeper cooperation between the two companies. Reported talks of TPU rentals in 2026 and purchases in 2027 indicate the production use could expand into a larger commercial hardware relationship.
- Google and Meta co-developed TorchTPU, a PyTorch-native version of the SGLang inference engine for TPUs.
- Meta's PyTorch-TPU lead says SGLang-TorchTPU is already running in production at Meta.
- Meta previously tried and abandoned TPU use between 2020-2023, and reports point to expanded TPU rental and purchase deals in 2026-2027.
Source: tomshardware.com — Shane Downing, 2026-10-22
Published there as: “Google’s TPUs for AI workloads now run in Meta's data centers”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.