Sovereign AI Capacity Costs $60 Million a Megawatt, in a Bite

AM Intelligence placed two orders on 5 October for 20,000 Nvidia Rubin GPUs, and the numbers behind them are the ones the other sovereign AI cluster announcements left out.

• Order book now runs to 29,000 GPUs and about 100 megawatts of contracted capacity, deployed across India and Malaysia as Vera Rubin NVL72 rack-scale systems.
• The new capacity is about 70 megawatts, delivered in the second quarter of 2027. A first order of 9,000 GPUs went in during August for the Hyderabad AI factory.
• The first 100 megawatts were funded with $6 billion committed, which is $60 million per megawatt. The next 300 megawatts carry an estimated investment of more than $20 billion, or over $66 million per megawatt.
• Rack density lands near 250 kilowatts and is liquid cooled throughout, so the same physical footprint can take successive generations of silicon.
• The three orders are the first committed tranche of a pipeline of about 400 megawatts, and the company plans to sell 1 gigawatt as compute-as-a-service inside a 5 gigawatt program.

Read the full analysis.