Reflection AI secures $1B compute deal with Nebius

NVIDIA argues that performance per watt is the critical metric for AI infrastructure efficiency, as power constraints directly determine token generation capacity and profitability in AI factories. The company claims its Blackwell NVL72 platform delivers up to 25x performance per watt over Hopper on frontier models like DeepSeek V4 Pro, achieved through system-wide codesign spanning silicon, software, and networking. As agentic AI increases token demand, infrastructure choices made today will determine which organizations can scale in a power-constrained environment.

New York Governor Kathy Hochul has signed the nation's first statewide data center moratorium, blocking new environmental permits for hyperscale facilities over 50 megawatts for up to one year. The move aims to give the state time to develop regulations addressing energy costs and environmental concerns. A separate bill passed by state lawmakers with a lower 20 megawatt threshold awaits the governor's signature.

Meta announced Monday it will invest an additional $40 billion to more than double the planned computing capacity of its Louisiana data center to five gigawatts from two gigawatts originally announced in 2024. The expansion makes the facility Meta's largest computing infrastructure to date. The move reflects Meta's escalating capital commitments to AI infrastructure as competition intensifies in the sector.

Google is expanding its tensor processing unit business beyond internal use and Google Cloud, now actively marketing TPUs to emerging cloud providers that have previously focused exclusively on renting Nvidia GPUs. The company has approached neocloud operators like Nscale as part of a competitive push against Nvidia's dominant market position in AI chips. This marks a strategic shift from Google's historical practice of keeping TPUs confined to its own infrastructure.