# OpenAI Reportedly Plans to Buy Google TPUs to Ease Nvidia Reliance > OpenAI plans to purchase Google's custom TPU accelerators to diversify its compute infrastructure away from Nvidia, backed by compiler engineering hirings. - URL: https://roobrief.com/en/posts/openai-google-tpu - Language: en - Published: 2026-11-02T01:00:00.000Z - Tags: ai, hardware - Source: https://www.theinformation.com **OpenAI reportedly plans to purchase Google's custom TPU hardware directly to diversify its AI accelerator infrastructure away from exclusive reliance on Nvidia.** According to a report from The Information citing people familiar with the matter, OpenAI is planning to buy custom Tensor Processing Units (TPUs) from Google. Neither company has officially announced the deal, and spokespeople for both sides did not immediately respond to requests for comment. OpenAI has already been renting Google Cloud TPU capacity since mid-2025. Moving toward a direct silicon purchase, however, represents a shift from temporary cloud provisioning to deploying Google's specialized silicon within its own infrastructure footprint. ## What is confirmed and what remains unverified The broader commercial cooperation between OpenAI and Google on infrastructure is well established. OpenAI has leased compute resources on Google Cloud Platform (GCP) and explored making its frontier models available through the Google Cloud marketplace. Crucial commercial details of this direct purchase arrangement remain undisclosed. The hardware generation (whether TPU v5p or the newer Trillium architecture), delivery timeline, shipment volume, and total contract value have not been made public. It is also unconfirmed how OpenAI plans to divide this hardware between pre-training clusters and high-volume inference(using a finished model to produce answers) fleets serving ChatGPT and developer APIs. ## From an Nvidia monoculture to multi-vendor silicon OpenAI is one of the world's largest buyers of Nvidia GPUs. However, as compute expenditure for frontier training runs and real-time inference scales exponentially, relying on a single silicon vendor introduces substantial financial and supply-chain risk. Nvidia's flagship accelerators carry steep price tags and frequent supply constraints. To mitigate this dependency, OpenAI has been co-developing custom AI silicon with Broadcom and TSMC, targeting production deployments around 2026. Tape-outs and enterprise data center deployments take years to mature. Because Google's TPU line is already battle-tested on large language models and global user traffic, acquiring mature TPU hardware provides OpenAI with immediate cost leverage while its proprietary silicon pipeline develops. ## Internal engineering signals at OpenAI Internal organizational movements point to a concerted technical effort to support Google's architecture. OpenAI recently listed job openings for TPU compiler engineers tasked with porting and optimizing frontier models for Google's hardware. The goal is to adapt model architectures and execution graphs, historically built around Nvidia's CUDA(NVIDIA's GPU computing toolkit) ecosystem, toward Google's XLA (Accelerated Linear Algebra) and related open compiler toolchains. Leadership connections also tie the engineering teams together. Richard Ho, who heads OpenAI's hardware systems division, previously led TPU system engineering at Google for years. Having a hardware executive with intimate knowledge of Google's interconnects and distributed training topologies signals a deliberate, long-term technical commitment. ## What to watch next Purchasing Google hardware does not imply an abrupt exit from Nvidia. Nvidia GPUs remain the benchmark for advanced AI model research, and OpenAI's massive clusters on Microsoft Azure continue normal operations. Two operational questions will be critical going forward: whether adopting TPU clusters noticeably drives down inference unit economics for GPT-tier workloads, and how deeply two direct competitors in generative AI and search will integrate their silicon supply chains. ## Sources - [OpenAI Plans to Buy Google AI Chips](https://www.theinformation.com) (The Information) - [OpenAI diversifying away from Nvidia with Google TPU and in-house chips](https://www.reuters.com) (Reuters) - [OpenAI Careers — Systems and Hardware Optimization](https://openai.com/careers) (OpenAI)