Twitxy Saved

James Rothmann

Verified account This account is verified. Learn more

@JamesRothmann

Cape Town, South Africa Joined July 2009
1.1K Following 1.1K Followers 1.4K Posts

Latest posts 5

James Rothmann Verified account @JamesRothmann
Though Experiment:

If all inference tokens in the world today ran on GB300 NVL72 Racks and DeepSeek v4.1 Flash, we would only need 200 Racks.

Let's say, Opus 5.5 needs 15 times this, then we need 3000 racks.

Nvidia is cranking out like 60,000 (equivalent) racks this year.
Chon Tang Verified account @chontang Oct 27
This is really interesting.

All of OpenRouter's Deepseek demand (roughly 30% of total OR token traffic) could be satisfied by 5-10 GB300 NVL72 racks. Even if that 10x over the next year, it's 50-100 racks.

NVIDIA is shipping 90k racks next year? What am I missing?
View quoted post on X
2 replies 2 reposts 9 likes 1.9K views View on X