Meta Puts Google Chips to Work, Testing Nvidia's Dominance

Google has officially expanded its reach in the artificial intelligence hardware market, confirming that its Tensor Processing Units (TPUs) are now running production AI workloads for Meta. This development marks a significant milestone, as it represents the first time Google has sold its custom chips to an external customer at this scale. The move positions Google directly against Nvidia, the long-standing leader in GPU sales for AI training and inference, by offering a competitive alternative that integrates seamlessly with major software frameworks.

The core of this announcement is a new PyTorch-native version of the SGLang inference engine, detailed on Google’s developer blog. Previously, running AI models on TPUs required using Google’s own JAX framework. However, the new infrastructure allows the hardware to function as a native option for PyTorch, the dominant AI framework developed by Meta, without requiring extensive code rewrites. A matching open-source version of vLLM, another popular inference engine, was also released to support this transition.

Zongwei Zhou, the lead for PyTorch-TPU at Meta, stated that the new SGLang-TorchTPU engine is "running in production at Meta today." He credited the progress to a "close three-way collaboration" between Meta, Google, and the SGLang team. This team operates out of RadixArk, a startup founded by the creators of SGLang, which built the engine in partnership with both tech giants. Zhou described the new stack as "a remarkably mature, PyTorch-native serving stack," highlighting its readiness for enterprise use.

Jana van Greunen, director of PyTorch engineering at Meta, emphasized the strategic importance of hardware diversity. She noted that Meta "cannot be constrained by a single hardware architecture" and that the new TorchTPU integration "enables us to deploy our PyTorch workloads to Google TPUs." Google confirmed it co-developed this solution directly with Meta’s PyTorch team, underscoring the collaborative nature of the project.

This is not Meta’s first encounter with Google’s hardware, although it is a new chapter in their relationship. Google identified its older PyTorch/XLA bridge as a "predecessor" to the current TorchTPU. Reports from SemiAnalysis indicated that Meta research teams utilized TPUs through this older bridge between 2020 and 2023 before those contracts were canceled. The recent push appears specifically designed to accommodate Meta, which had previously resisted migrating its entire stack to Google’s JAX framework.

While the physical location of Meta’s production TPUs remains unclear, industry talks last year suggested a phased approach involving Google Cloud TPU rentals in 2026, with direct purchases potentially occurring in 2027. Meta was later reported to have signed a multi-year rental agreement, providing a financial foundation for this expanded partnership.

The timing of this announcement coincides with a broader industry shift away from Nvidia’s software ecosystem. Just one day prior, Amazon announced it was open-sourcing TorchNeuron, a native PyTorch backend for its own Trainium2 and Trainium3 chips, at the PyTorch Conference. Both Google and Amazon appear to be strategically loosening Nvidia’s software lock-in to capture a larger share of the AI workload market.

Performance metrics suggest Google’s hardware is competitive with Nvidia’s latest offerings. A tuned vLLM setup on the seventh-generation Ironwood TPU reportedly exceeded 700 tokens per second for a single user when serving Moonshot AI’s Kimi K3 model. This test, conducted by Inferact, utilized 16 Ironwood chips with speculative decoding. In comparison, a similar setup on 16 Nvidia GB200s reached 452 tokens per second. For smaller batch sizes without speculative decoding, the new kernels deliver roughly 1.4 to 2 times the throughput of the GB200 baseline.

Looking ahead, SGLang-TorchTPU will continue to expand its model support, starting with DeepSeek V4.1 Flash and Reflection Beam. RadixArk aims to achieve "Day 0" TPU support for new open models by the first quarter of 2027, ensuring new releases are supported on the day they launch. Google, Meta, and RadixArk plan to share further details at the SGLang Summit 2026, scheduled for November 12-13 in San Francisco.