The AI Power Crisis — Part 1 - NuttyCLD
nuttycld.substack.com · 2,771 words · saved by 1 readers
Two Languages of Electricity
We are building the most power-hungry machines in human history. The frontline of the GPU war has shifted. Compute speed and memory capacity alone no longer decide the winner. The variable that will define the next chapter is power itself. The NVIDIA H100 SXM’s thermal design power (TDP) of up to 700W was already substantial. The Blackwell B200 crossed 1,000W. The upcoming Vera Rubin platform is reported at 2,300W per GPU (Max-P) based on industry sources. A single chip consuming as much power as a small air conditioner. At the rack level, the scale shifts again. Current production…
saved by
related reading
- The AI Power Crisis — Part 2 - Nuttynuttycld.substack.com
- Siemens Energy · Technical document · Letter format portrait – Templatep3.aprimocdn.net
- Power 2026 - Electricity Pricing in the Age of AIpower2026.ai
- Roadmap: The AI data center stack - Bessemer Venture Partnersbvp.com
- Why American data centers can't plug inworksinprogress.co
- Can AI scaling continue through 2030? | Epoch AIepoch.ai
- IIIa. Racing to the Trillion-Dollar Cluster - SITUATIONAL AWARENESSsituational-awareness.ai
- Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI computedwarkesh.com
- How to Build the Future of AI in the United States | IFPifp.org
- The Datacenter Bible - From Layman to Expert in 100 Pagestscsw.substack.com
- The Moon Should Be a Computerpalladiummag.com
- Vik's Newsletter | Vikram Sekar | Substackviksnewsletter.com