Andree Jacobson

Microsoft's Fairwater Wisconsin went live today, ahead of schedule. It's the world's most powerful AI data center. Hundreds of thousands of NVIDIA GB200 GPUs, wired to behave as a single computer. The physical build is biblical. 315 acres. 1.2 million square feet across three buildings. 26.5 million pounds of steel. 46.6 miles of foundation piles. 120 miles of underground cable. The storage wing alone runs the length of five football fields. Each rack holds 72 GPUs sharing 14 terabytes of pooled memory, pushing 865,000 tokens per second. One rack generates more text per second than you've written in your entire life. Fairwater holds thousands of racks. The speed of light became a real bottleneck. Cable lengths between racks were introducing latency that slowed training. Microsoft's fix was a two-story layout so racks sit directly above and below each other. We are now architecting around special relativity to train AI. The power numbers are where it gets surreal. Phase 1 pulls 400 megawatts. Full build-out approaches 900 megawatts. Roughly a nuclear reactor, feeding one computer. The grid draw equals 300,000 homes. Fairwater Wisconsin is only the first. Atlanta is already online. Norway and the UK are next. Microsoft is stringing them together over 120,000 miles of dedicated fiber into one AI superfactory with 2+ gigawatts of total capacity. The buildings outlast the chips. GB200 gets swapped for GB300, then Rubin. The cells are modular. Microsoft's real bet is on the real estate, the power hookups, and the fiber. AI progress is now gated by concrete and megawatts.

We are looking for excellent people to help build our vertically integrated AI stack. Numerics, quantization, HW simulators, compiler, runtime, kernel performance, RTL, verification, emulation, DFT, physical design, post Si bringup. Join us at Tesla!