Alphabet confirmed on April 24, 2026, a major expansion of its strategic partnerships with Nvidia and Broadcom, aimed at scaling its artificial intelligence infrastructure. The announcements, finalized during the Google Cloud Next 2026 conference in Las Vegas, introduce next-generation hardware and software solutions designed to support agentic and physical AI workloads.
The collaboration with Nvidia centers on the debut of the A5X bare-metal instances, which utilize Nvidia’s new Vera Rubin architecture, the successor to the Blackwell platform. According to Google Cloud, these systems are engineered to deliver a tenfold reduction in inference costs per token and a tenfold increase in token throughput per megawatt compared to previous generations. The A5X instances integrate Nvidia ConnectX-9 SuperNICs with Google’s proprietary Virgo networking fabric, allowing clusters to scale to 80,000 GPUs at a single site or up to 960,000 GPUs across multisite configurations.
Mark Lohmeyer, vice president and general manager of AI and computing infrastructure at Google Cloud, stated that the next decade of AI will be defined by the ability to run demanding workloads on an integrated, AI-optimized stack. As part of this co-engineered AI Hypercomputer architecture, Google Cloud will also offer confidential computing for Nvidia Blackwell GPUs, enabling encrypted processing for sensitive models.
Broadcom, which expects AI-related revenues of 10.7 billion dollars in the second quarter of fiscal 2026, is also deepening its technical integration with Google Cloud. The companies launched Cloud Network Insights, a native Google Cloud observability service powered by Broadcom’s AppNeta technology. This tool provides end-to-end visibility across hybrid and software-as-a-service environments, helping engineers isolate performance bottlenecks in distributed AI applications.
The partnership with Broadcom also extends to custom silicon development. Alphabet confirmed it has secured an agreement with Broadcom to continue the supply and co-development of Tensor Processing Units through 2031. This includes the deployment of the eighth-generation TPU line, which has been split into two specialized chips: the TPU 8t for large-scale training and the TPU 8i for inference and reinforcement learning. Amin Vahdat, Google’s chief technologist for AI and infrastructure, described the split as a fundamental shift to meet the differing requirements of training and inference in the agentic era.
Furthermore, a three-way agreement between Alphabet, Broadcom, and AI startup Anthropic was detailed. Under this arrangement, Google Cloud and Broadcom will provide 3.5 gigawatts of TPU-based computing capacity to Anthropic beginning in 2027. This infrastructure deal coincides with Alphabet’s confirmation on April 24 of a 40 billion dollar investment in Anthropic, consisting of 10 billion dollars in immediate cash and 30 billion dollars in milestone-contingent funding.