Wells Fargo flags Blackwell scarcity

Nvidia’s Blackwell story is turning into a capacity story. Not a “is demand real?” story. Not even really a “how big is AI?” story. The live question now is how fast Nvidia and its supply chain can physically ship enough systems, power gear, cooling, and networking to satisfy everyone trying to build around Blackwell at once. Wells Fargo made that point explicit on May 12 when it raised its Nvidia price target and said compute demand still looks greater than supply. ### What changed on May 12? Wells Fargo analyst Aaron Rakers raised his Nvidia price target to $315 from $265 and kept an overweight rating. The interesting part wasn’t just the higher target. He rebuilt the valuation around a gigawatt-capacity model for AI infrastructure, basically treating Nvidia’s data-center revenue as a function of how much AI power gets deployed into the field. (thestreet.com) ### Why use gigawatts instead of just GPU counts? Because the bottleneck has moved up a level. A modern AI cluster is not just chips. It is racks, power conversion, liquid cooling, networking, batteries, and buildings that can actually feed the load. Wells Fargo’s model scales from 9.2 gigawatts of AI infrastructure in fiscal 2026 to 15.7 gigawatts in fiscal 2027, then 20.8 and 25.2 in the following two years. That framing tells you the constraint is now whole-system deployment, not a simple unit tally. (thestreet.com) ### So where does Blackwell scarcity show up? In the gap between demand and shippable capacity. Rakers said the backdrop still points to compute demand greater than supply, and that Nvidia’s data-center revenue depends on its ability to scale deployed gigawatts. Wells Fargo’s revenue numbers are huge — $354.5 billion for fiscal 2027 data center, then $504.5 billion and $628 billion in the next two years — and they sit above consensus. (thestreet.com) That only works if Blackwell systems keep ramping despite constrained availability. ### Is this still mostly a hyperscaler story? Not anymore. Fortinet said on May 12 that it is accelerating its FortiAIGate offering with Nvidia AI platforms and software, aiming to secure AI workloads, data, and autonomous agents in data centers and the cloud with GPU-accelerated security. That is a different kind of buyer pressure. It means Blackwell-class infrastructure is spreading into enterprise security stacks, not just giant model-training clusters. (thestreet.com) ### What about the edge piece? That is the other expansion path. Zero Latency launched a closed beta for Zerogrid last week, describing it as a distributed inference grid that routes workloads to edge capacity based on latency, data locality, and burst constraints. In plain English — more AI demand is shifting toward low-latency inference near users and data, which favors smaller but still scarce accelerated deployments rather than one giant centralized cluster. (fortinet.com) ### Why are suppliers suddenly part of the story? Because if Nvidia wins by selling modular AI factories, the supporting gear becomes strategic. DigiTimes highlighted Delta Electronics as a beneficiary of this shift, and Delta itself has been showcasing 800 VDC power racks, battery backup, liquid cooling, and microgrid gear built for next-generation AI factories at Nvidia GTC 2026. That is the infrastructure layer Blackwell needs to turn demand into revenue. (markets.businessinsider.com) ### What does this mean for buyers? The catch is that scarcity is no longer just about getting GPUs on a PO. Buyers need the full stack lined up — power, cooling, networking, security, and edge placement — before the procurement window closes. If Blackwell demand keeps broadening from hyperscalers into enterprise AI security and edge inference, lead times can tighten even if Nvidia keeps improving chip output. That last part is an inference, but it follows directly from the shift toward gigawatt-scale deployment and system-level dependencies. (digitimes.com) ### Bottom line? Wells Fargo’s call matters because it says the Nvidia debate has moved from valuation theater to industrial throughput. Blackwell looks scarce not because demand is fragile, but because too many parts of the market now want in at the same time. (thestreet.com)