#NVDA : AI DATA CENTER POWER SHORTAGE BECOMES A BIGGER STORY NVIDIA’s Rubin Ultra Could Push AI Rack Power Toward 600 kW NVIDIA’s next-generation Rubin Ultra systems are expected to push data-center power requirements to another level. The Rubin Ultra NVL576 “Kyber” rack is designed around roughly 600 kW of power per rack, with deployments expected from 2027. That creates a major challenge for data-center developers: getting the GPUs is one thing, getting enough electricity to run them is another. Morgan Stanley has been warning that power availability could become one of the biggest bottlenecks for U.S. data-center growth through 2028. Its research has estimated a large gap between projected data-center demand and available power access. ⚡ Why Existing Power Is Becoming More Valuable A data-center site with land is no longer enough. Developers increasingly need: • Available grid power • Secured power capacity • Substation and transmission access • High-density cooling infrastructure • Fast construction timelines This makes already-powered or near-powered data-center sites particularly valuable because securing a new grid connection can take years. NVIDIA itself is preparing for this shift. Its 800VDC architecture is designed to support racks from around 100 kW to more than 1 MW as AI infrastructure scales. 🏢 Companies Getting Attention This power constraint is putting several AI data-center operators and infrastructure developers on investors’ radar, including: #NBIS – AI infrastructure and data-center expansion #CRWV – CoreWeave, focused heavily on AI compute infrastructure #IREN – Large-scale data-center and power infrastructure development #WULF – Expanding toward AI/HPC data-center capacity #CIFR – Developing data-center infrastructure around its power assets #APLD – Building high-density AI/HPC data-center capacity #CORZ – Converting existing power infrastructure toward high-performance computing The common theme is simple: secured power can become as important as the physical data-center building itself. 🔥 The Bigger AI Infrastructure Story The AI race is moving beyond GPUs. Investors are increasingly looking at the entire chain: GPU → Data Center → Power → Cooling → Grid → Land NVIDIA can keep making faster and more powerful AI systems, but those systems ultimately need electricity and infrastructure capable of supporting them. That is why the next phase of the AI buildout could create opportunities for companies that already control power, land and data-center capacity.