Razor's Edge: Jensen Compute Box Avanlanche

Interesting WSJ article.

"Then he went on to describe the situation the company was in. Major cloud providers were running away from Nvidia, making plans to use alternative chips and lessen their reliance on the firm. Even worse, Huang said, the vast supply of chip wafers and memory Nvidia had committed to buying were coming at him like an “avalanche.”

Lot of time spent last six months talking about Nvidia's moat is that Jensen's locked down the supply chain for AI compute. What this article tells you is he kind of has operated for the last few years with a a strategy that his scale and effective 80% mkt share of compute mkt must be mantained as the mkt exponentially scaled or effectively as he scaled it. And generally this has meant securing ever increasing supply for longer and longer lead times as he viewed that as his most critical asset. But this also tells u that as the market has boomed he has had to get in front of the demand and to an extent help create it himself with allocations/financing/investment.

So, the timing, nature of this meeting and the use of the word 'avalanche' tell you that as recently as August he was concerned that he has overshot near-term real demand.

That also explains this Berlin conference fable..

Berlin during a conference, Huang, accompanied by his top sales executives, broadcast a new message to a half-dozen CEOs of companies that are major buyers of Nvidia chips who were seated around a circular table. He opened with a story that sounded like a fable, according to people who attended.

Huang had once found a soup vendor in Japan, he said, with a long line of customers. He joined the line, and after a long wait, enjoyed one of the best bowls of soup he’d ever had. When he asked for another serving, the soup-seller told him no—he had too many customers waiting and not enough soup for all of them.

“He’s a better businessperson than all of you,” Huang told the cloud computing CEOs. The soup-man doesn’t sell out all his inventory to one or two customers, but instead keeps prices high and his customer base diverse, the CEO said.

For me this is Jensen basically expressing frustration that his customers can't drive healthy demand elasticity despite all that he's doing for them.

Anyway, after reading this article I concluded one thing for certain, Jensen's competitive strategy has contributed notably in squeezing the component market and created an element of the compute shortage all in a very reflexive manner. I also am convinced the SpaceX talk about 7-20gw for 2027 that are all Nvidia now makes a lot more sense. Post August Jensen went looking for a swing buyer with a balance sheet capable of absorbing whatever margin capacity he has committed to.

At that scale, the sequence seems pretty obvious..

1. Robust demand and increasing competitors cause Nvidia to reserve more supply. Manufacturing and memory availability tightens, reinforcing the perception of overwhelming compute demand.
2. Suppliers expand capacity against Nvidia’s commitments. Consequently, its forecast becomes a major driver of future industry capacity.
3. Nvidia helps finance additional demand with its balance sheet. That supports orders for the production it has already arranged.

The upstream shortage and the eventual downstream surplus can therefore originate from the same strategic decision made by one fabless company trying to mantain 80% share. You do not need two unrelated events or an industry wide collapse in AI demand to get from one to the other. My Dark Matter analogy from couple weeks ago now appears even more compelling.

Anyway, with Jensen out this gets more interesting. Nvidia is gonna make all these chips because strategically their view is by the end of 2028 they have like 2.5trl in compute base across 4 generations which effectively locks everyone else out of a shot at any meaningful share for years to come. On the flip side, if the demand is not there Nvidia likely tightens the screws itself and an infra bust happens which crushes everyone else far worse then them. Also a reminder LTA's kind of worthless if the glut is potentially as big as it might look to Nvidia and there share is as big as it is.