Upper90 lends General Compute $400M against SambaNova SN50s

A $400 million loan changed hands this week that says more about where AI infrastructure is heading than most of the flashier funding rounds. Inference cloud startup General Compute, according to TechCrunch, took the money from Upper90 and reportedly put up SambaNova's SN50 chips as collateral, which if it holds up to scrutiny is the first debt facility backed by inference-specific silicon rather than Nvidia GPUs.

The lender is not a random face. Upper90 co-founder and CEO Billy Libby, a former Goldman Sachs quantitative trader, financed Crusoe's GPU buy in 2021, which he believes was the first loan against the value of advanced chips. His pitch for doing something similar now is candid: "When we financed Nvidia GPUs as the first group to do that, the market was inefficient. We could really put together something as an early participant." The through-line is that lenders make outsized fees on hardware nobody else knows how to price yet, and inference silicon looks like the current gap.

Why it matters beyond one deal: for two years the working assumption behind AI capex has been that Nvidia GPUs are the only assets a bank will lend against, which is a big part of why the debt market has been so concentrated. If Upper90 can underwrite SambaNova SN50s, Intel-backed silicon that General Compute claims runs inference 16 times faster than GPU-based clouds, then other inference chipmakers like Groq, Cerebras and TensorWave have a template for how their buyers can raise cheap money without diluting equity. General Compute CEO Finn Puklowski frames it exactly that way, calling the deal "the first signal of capital organizing itself and the fragmenting of Nvidia's monopolistic dominance."

The honest caveat is that the 16x speed figure comes from General Compute itself, and the reporting does not disclose the loan's tenor, interest rate, advance rate, or which customers are on the hook for filling the SN50 capacity. Residual value on a chip family with no real secondary market is a genuinely hard thing to underwrite, and Libby's 2021 GPU thesis is only vindicated because a very specific compute cycle went his way.

What is worth watching is whether the second and third of these deals appear in the next few quarters. If they do, the moat around Nvidia is not really the chips anymore, it is how quickly the debt markets learn to price the alternatives.