Nebius' Earnings Call: "We are typically seeing 4 or more customers competing for every GPU we bring online..."

OK, let's break it down:

1. Nebius notified investors in today's call that all their newly deployed capacity is sold out again.

2. "We are typically seeing 4 or more customers competing for every GPU we bring online."

3. That makes sense, we are short on compute that is deployed and functional.

4. As of December 31, 2025, just one quarter ago, Nebius had only 170MW of active power consisting of H100, H200, GB300, GB200, B200, B300 and some others. We are talking about dozen of thousands GPUs only.

5. According to the last unaudited ER from Nebius this morning, they added around $1.6B to their PP&E, slightly less than $1 billion worth of GPUs. Hopefully they are all operating. That means around 17,500 to 22,000 more GPUs, and this is if we are generous with Nebius, and with their ability to bring such an amount of GPUs online.

6. Which means around 40MW deployed in one quarter.

7. Just as a reminder, Anthropic rented a 300MW datacenter from xAI for the demand. 40MW is tiny and not enough.

8. Now these numbers include the GPUs reserved as part of the second tranche for Meta and Microsoft.

9. Which means Nebius deployed merely a low few thousands of new GPUs to their short-term customers, and we are generous here.

10. According to NVIDIA's Q1 estimated datacenter revenue of approximately $70B, NVIDIA is expected to ship/sell/etc. around 1.2M to 1.5M GPUs.

11. Therefore, we are short on deployed compute, but there are lots of GPUs. Lots.

Q.E.D