> Collectively there are less than <50 teams worldwide who would be in the market for 16 nodes of H100s (or much more), at any point in time, to do foundation m

> Collectively there are less than <50 teams worldwide who would be in the market for 16 nodes of H100s (or much more), at any point in time, to do foundation model training At best 100 and this number will go down as many would fail to make money. Even traditional 100 software development companies would have a very low success rate and here we're talking about products that themselves work probabilistically all the way down.