Nvidia just paid $6B to license Poolside's tech + hire 100+ engineers, plus $1B invested at $12B pre-money. Target? China.
Open-weight models are free to download and modify — which is why Chinese models (DeepSeek, Kimi, GLM) are getting adopted globally. They now trail US frontier capability by only a few months. US labs built closed models instead. Nvidia is buying into that fight, which also means competing with OpenAI and Anthropic — two of its largest customers.
But here's the number that matters: Poolside had six weeks to raise $2B for a 40,000 GB300 cluster. They missed it. They lost the cluster. A customer with a signed allocation couldn't finance it.
Nvidia reports Wednesday. Two weeks ago it cut its OpenAI backstop from $250B to under $120B.
Demand exists. Financing is the constraint.
Watch backlog conversion, not backlog size.
$NVDA $SPY
Open-weight models are free to download and modify — which is why Chinese models (DeepSeek, Kimi, GLM) are getting adopted globally. They now trail US frontier capability by only a few months. US labs built closed models instead. Nvidia is buying into that fight, which also means competing with OpenAI and Anthropic — two of its largest customers.
But here's the number that matters: Poolside had six weeks to raise $2B for a 40,000 GB300 cluster. They missed it. They lost the cluster. A customer with a signed allocation couldn't finance it.
Nvidia reports Wednesday. Two weeks ago it cut its OpenAI backstop from $250B to under $120B.
Demand exists. Financing is the constraint.
Watch backlog conversion, not backlog size.
$NVDA $SPY