AI infrastructure scales up

Artificial intelligence is getting a bigger industrial backbone: Nvidia backed Vast Data at a reported $30 billion valuation, and Nvidia and Google Cloud widened their infrastructure partnership on April 22. (cnbc.com) (blogs.nvidia.com) Vast said April 22 that it closed a $1 billion Series F round at a $30 billion valuation, up from a $9.1 billion Series E valuation in late 2023. The company said Drive Capital led the round, Access Industries co-led, and Nvidia joined the financing. (vastdata.com) (cnbc.com) Vast sells software that manages and serves large pools of data for artificial intelligence systems, and CNBC reported the company says its platform supports projects running on millions of graphics processing units, or GPUs. Vast was founded in 2016 and has pitched its platform as a way to keep expensive AI chips fed with data instead of sitting idle. (cnbc.com) (vastdata.com) A data system here works like a warehouse and conveyor belt for AI: it stores files, pulls the right pieces quickly, and delivers them to chips that do the math. Google Cloud and Nvidia said their latest expansion is aimed at “agentic” AI systems and simulations that need more coordinated compute, software, and cloud services than a single model demo. (cloud.google.com) (blogs.nvidia.com) At Nvidia’s GTC 2026 conference, Google Cloud said it would add support for Nvidia Vera Rubin NVL72 systems, expand Nvidia software across Vertex AI Training and Model Garden, and integrate Nvidia Dynamo with Google Kubernetes Engine Inference Gateway. Google also highlighted G4 virtual machines powered by Nvidia RTX PRO 6000 Blackwell Server Edition chips. (cloud.google.com) (nvidia.com) Those announcements land as large companies are trying to move AI from pilots into production systems that run continuously, not just in test environments. Nvidia said the joint Google Cloud platform is meant to help customers turn experimental agents and simulations into deployed systems for coding, security, fleet management, and factory operations. (blogs.nvidia.com) (cloud.google.com) Hospitals and health systems sit downstream from this buildout because many medical AI tools run on the same cloud, chip, and data layers sold to other enterprises. When a hospital buys an imaging model, a documentation assistant, or a retrieval system for clinical records, it is often also inheriting the vendor’s cloud region, storage design, and model-serving dependencies. (cloud.google.com) (vastdata.com) That pushes procurement questions beyond model accuracy. Health systems weighing new AI contracts may need to ask where patient data is stored, which subcontractors handle inference, whether workloads can move between clouds, and how outages or pricing changes at an upstream provider would affect clinical operations; those are reasonable inferences from the cloud and infrastructure arrangements the companies described. (cloud.google.com) (blogs.nvidia.com) (vastdata.com) Nvidia’s central role in the stack has grown fast. CNBC reported last week that Nvidia’s data center segment now accounts for 91.5% of its revenue, a sign that the company’s business is increasingly tied to AI infrastructure rather than its older gaming base. (cnbc.com) The near-term story is not a new chatbot or a new hospital app. It is a race to control the storage, compute, and cloud plumbing underneath them — the layer that decides how fast enterprise AI can move from demo to daily use. (cnbc.com) (cloud.google.com)