Nvidia doubles down on partnerships

Nvidia is tying its hottest chips more tightly to a handful of giant cloud partners instead of treating them like auction inventory. (finance.yahoo.com) Chief executive Jensen Huang said on the Dwarkesh Podcast that Nvidia “never” sells chips to the highest bidder and instead tries to keep supply predictable for major data-center customers that plan years ahead. Yahoo Finance reported the remarks on April 23, 2026. (finance.yahoo.com) A day earlier, Nvidia said it was expanding its Google Cloud alliance with support for Nvidia Blackwell graphics processing units, confidential computing on Blackwell, and Google’s Gemini models on Google Distributed Cloud. Nvidia also said Google Cloud’s new A5X bare-metal instances will use Nvidia Vera Rubin chips and scale to nearly 1 million GPUs. (blogs.nvidia.com) The basic market shift is straightforward: the scarce product is no longer just a chip, but a full stack of hardware, software, networking and cloud capacity bundled together. Nvidia’s announcement described that stack as spanning optimized libraries, silicon, systems, and orchestration with Google Cloud. (blogs.nvidia.com) That is a change from the shortage phase of the artificial-intelligence boom, when investors and customers focused on how many graphics processing units Nvidia could ship each quarter. Huang’s allocation comments point to a model where long-term customers get steadier access than whoever offers the highest spot price. (finance.yahoo.com) Google Cloud framed the A5X system as part of its AI Hypercomputer platform, with single-site clusters of up to 80,000 Rubin GPUs and federated deployments reaching 960,000 GPUs across sites. That scale targets customers building very large training and inference systems rather than renting a few servers at a time. (blogs.nvidia.com) The Google tie-up also reaches customers that cannot run public-cloud workloads in the usual way. Nvidia said Gemini on Google Distributed Cloud will run on Blackwell systems in on-premises, air-gapped and edge environments, with confidential computing designed to keep prompts and model data encrypted in use. (blogs.nvidia.com) Nvidia and Google are not starting from scratch. Nvidia said the companies have worked together for more than a decade, and Nvidia had already previewed Gemini support on Google Distributed Cloud with Blackwell infrastructure in 2025. (blogs.nvidia.com (blogs.nvidia.com) The immediate effect is that Nvidia is selling customers a queue position inside a cloud ecosystem as much as a processor. In this market, access, placement and integration are becoming part of the product. (finance.yahoo.com)