Every AI lab wants more compute, and PaleBlueDot AI thinks it can sell them the middle layer. The Palo Alto company said October 1 it closed a Series C of $200M at a $3.2B valuation, led by ComputeCore with participation from existing investor B Capital.
PaleBlueDot’s model stacks three products. The first owns and operates its own GPU clusters. The second runs a marketplace, reselling capacity that partner suppliers own. Serverless inference is the third. Founded in 2024, the company markets the bundle as a Super Intelligence Infrastructure Platform. The fresh capital goes toward more clusters, widening the hardware and locations customers can choose from.
Validation from Nvidia has become part of the sales pitch. In Japan, PaleBlueDot’s HGX B300 cluster was granted Exemplar Cloud status for large-model training. To earn it, the cluster cleared a 95 percent performance bar on every recipe in the program. Each node carries eight Blackwell Ultra GPUs. NVLink ties them together, and an 800Gb/s Quantum-X800 InfiniBand fabric feeds the node.
The order book has grown. By the end of September the company said it had signed more than $5B in customer contracts. US and Japanese customers together generate over half of monthly revenue. In July it closed a $255M credit refinancing.
The plan, according to CEO Stephen Watts, is a wider customer base. He named frontier labs, Neolabs and US enterprises as targets. Behind the strategy sits a simple bet: that demand for training and inference keeps outrunning supply.
