OpenAI plans to buy Google’s Tensor Processing Units, The Information reported on October 27, citing a person familiar with the discussions. The potential purchase would give the ChatGPT developer another source of AI computing hardware alongside Nvidia GPUs and its own custom chips. Neither company has confirmed a purchase, and the reported deal’s size, price and delivery schedule remain unknown.
Bloomberg subsequently reported the plan, while Reuters noted that OpenAI had previously explored renting Google’s processors. Buying TPUs outright would be a different arrangement, potentially giving OpenAI more direct control of its computing infrastructure.
The Information’s October 28 technology program discusses the reported OpenAI purchase, Oracle’s TPU arrangement and competition in AI chips.
Google’s AI chips are moving beyond its own cloud
Google originally developed Tensor Processing Units for its own services, but is increasingly selling complete processor systems to outside customers. Its eighth-generation hardware separates high-throughput model training on TPU 8t from low-latency inference on TPU 8i. Google says that specialization improves cost and energy efficiency; results will depend on the workload and system configuration.
Google Cloud explains the different roles of its eighth-generation TPU 8t and TPU 8i processors.
The company’s broader push is already visible in its financial results. Google Cloud reported $30.7 billion in third-quarter revenue, with TPU systems contributing to growth. Selling the hardware requires major upfront spending on production, memory, networking and inventory, not just chip design.
On October 27, Oracle and Google announced a separate, confirmed agreement to make TPUs available through Oracle Cloud Infrastructure beginning in 2027. Oracle says it will supply the data-center power, cooling and networking required to operate the systems. That rollout is planned, not yet a generally available service.
Google Cloud’s hardware overview introduces the eighth-generation TPU system for training and inference.
More competition, but no simple replacement for Nvidia
AI companies increasingly want choices beyond a single accelerator supplier. OpenAI also developed its Jalapeño inference chip with Broadcom, while Google is developing software support for popular machine-learning frameworks. Each option still requires compatibility work, appropriate network architecture and sufficient data-center capacity.
Google Cloud’s technical conversation with Amin Vahdat and Jeff Dean examines the architecture and memory demands of eighth-generation TPUs.
The real comparison extends beyond headline computing speed. Training efficiency depends on productive accelerator utilization, while inference costs depend on latency, throughput, memory and electrical consumption. Those trade-offs also shape demand for optical networking components and the switching technology connecting AI systems.
For now, Oracle’s 2027 rollout is a public commitment; the OpenAI purchase remains reported but unconfirmed. The next milestones are a formal agreement, disclosed order details and evidence that TPU installations can deliver competitive performance at scale.
BitcoinVersus.Tech — independent reporting on computing, infrastructure and technology.
Editor’s Note: We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support our work and help further secure the integrity of our research initiatives, please scan the QR code below to donate 5,000 sats.
Scan this QR code with a compatible Bitcoin wallet while viewing the article on a laptop or larger screen. Donation address: 3C9o19EH5HSiwEPyCTmEKzxhNCbo2X6TTb.
BitcoinVersus.tech is not a financial advisor. This media platform reports on financial subjects purely for informational purposes.
Leave a Reply