Nvidia already sells about twelve gigawatts of AI a year. Consensus needs half as much again next year; the price assumes the power stops growing after that.
The callLONG
Price at publication
US$204.64close, 30 October 2026
Target
US$245.00+19.7% · 12 months, to October 2027
Conviction
Low
What proves it wrongData Center revenue falls from one quarter to the next in any quarter Nvidia reports before October 2027, which would show chips running ahead of the power to use them; or its non-GAAP gross margin falls below 70 per cent in any of those quarters; or Nvidia makes a payment under any of its land, power and shell guarantees.
Nvidia designs the chips, racks and networks inside AI data centres. Data Center revenue was US$89.0 billion in the quarter to 26 July 2026, up 117 per cent, and 92.5 per cent of the total.
Its own filings now treat power as the constraint. Nvidia has guaranteed up to US$108.5 billion of land, power and building costs for customers, mostly for a 4.25 gigawatt campus in Ohio that OpenAI will lease, ready from 2028.
Nvidia's own figures for that campus put about US$41 billion of its revenue on each gigawatt of computers. On that yardstick this year's revenue equips about 12 gigawatts, and the fiscal 2028 consensus needs about 51 per cent more, a pace the queues on this site make hard.
I assume the gigawatts grow 20 per cent and then 15 per cent, so my fiscal 2028 earnings are 23 per cent below consensus. Yet at 16 times, the price needs fiscal 2029 earnings only 4 per cent above my fiscal 2028.
My call is long, with low conviction and a target of US$245. My base value is US$244, 19 per cent above the price, within a range of US$114 to US$394. It is wrong if Data Center revenue falls on the quarter before, if gross margin drops below 70 per cent, or if Nvidia pays out on a guarantee.