NVIDIA revenue widens beyond hyperscalers

NVIDIA is trying to change the way investors think about its business. The old story was simple — a few giant cloud companies bought almost everything, and NVIDIA rode that wave. The new story is broader. At GTC 2026, Jensen Huang said 60% of NVIDIA’s business now comes from the top five hyperscalers, while the other 40% comes from a much wider mix of customers, including sovereign clouds, enterprises, industrial users, robotics, and edge computing. (investing.com) ### Why does that 40% matter? Because it changes the risk profile. If nearly all growth comes from a handful of buyers, one spending pause can wreck the whole narrative. A 40% share outside the biggest cloud players suggests NVIDIA is becoming less of a pure hyperscaler trade and more of a full-stack infrastructure company with demand (investing.com)s.” (nvidia.com) ### Who is in that “other 40%” bucket? Basically, everyone trying to build AI without being Amazon, Microsoft, or Google. That includes regional cloud providers, national “sovereign AI” projects, enterprise IT shops building internal inference systems, and newer physical-AI markets like robotics and autonomous machines. GTC 2026 was packed with that message. NVIDIA was talking as much about (nvidia.com)aining. (hashrateindex.com) ### Is this just a talking point? Probably not. The broader demand story lines up with the rest of NVIDIA’s product roadmap. Rubin is aimed at agentic AI and reasoning workloads, and NVIDIA’s own fiscal 2026 annual report says production shipments are expected in the second half of fiscal 2027. The company is clearly planning for a market where inference spreads beyond a few giant training clusters and into many more real-world deployments. (stocklight.com) ### So where’s the catch? The catch is that revenue diversification is not the same thing as supply-chain diversification. Bloomberg data, echoed in coverage picked up by Tom’s Hardware and Yahoo Finance on May 4 and May 5, says Asian suppliers now account for roughly 90% of NVIDIA’s production costs, up from about 65% a year earlier. (tomshardware.com) ### Why is that a real operational risk? Because the same parts keep showing up in more products. NVIDIA still depends on a concentrated set of suppliers for advanced packaging, high-bandwidth memory, wafers, and server assembly — names like TSMC, SK hynix, Samsung, Foxconn, and Quanta keep coming up in the reporting. If robotics, automotive, edge servers, and data-center GPUs all pull on the same constrai(tomshardware.com)ply for another. (msn.com) ### Why does memory matter so much here? Because AI systems are not just compute-bound anymore — they are memory-bound too. Blackwell-class systems need enormous amounts of HBM, and newer embedded and robotics platforms are also competing for advanced memory and leading-edge manufacturing capacity. Even Tom’s Hardware’s recent reporting on Jetson product changes points to memory shortages as a live issue in NVIDIA’s edge lineup. (tomshardware.com) ### What’s really changed in the NVIDIA story? NVIDIA is widening from a hyperscaler supplier into the operating system for AI infrastructure. That is the upside. But the industrial base under that expansion is still narrow, regional, and bottleneck-prone. So the company’s demand story is getting healthier at the exact moment its manufacturing dependencies are getting more concentrated. (investing.com)i-expansion-and-strategic-partnerships-93CH-4564073)) ### Bottom line The important shift is not that hyperscalers stopped mattering — they still make up most of NVIDIA’s business. It’s that the next leg of growth looks more distributed across enterprise, sovereign, and physical-AI customers. But turns out the supply chain underneath all of it is still heavily concentrated in Asia, which means the ceiling on growth may depend as much on memory and packaging as on demand. (investing.com)