General Compute, a Boston-based AI inference cloud startup, has secured a debt facility of up to $400 million from investment firm Upper90 — a deal backers describe as the first major AI infrastructure financing collateralized by inference-specific chips rather than the Nvidia GPUs that have dominated AI lending to date. The facility starts at $100 million and scales with customer demand.
Chips as collateral, not GPUs
Founded by CEO Finn Puklowski and CTO Jason Goodison, General Compute raised a $15 million seed round in May to build an inference “neocloud” around silicon from SambaNova, the Intel-backed chipmaker. The new facility is secured by SambaNova SN50 chips, which are purpose-built for running already-trained models rather than training them, and which the company says deliver several times the throughput per dollar of general-purpose GPUs on inference workloads while skipping the water-cooling infrastructure GPU clusters typically require.
That distinction is the deal’s real novelty: prior GPU-backed financings have relied on Nvidia hardware’s resale value and broad software compatibility as collateral. Lenders underwriting a loan against newer, more specialized inference silicon are betting that dedicated inference chips hold comparable resale and utilization value — a bet that reflects how quickly the inference market has grown relative to training.
Why it matters
The deal is another data point in a broader shift in AI infrastructure spending: as more frontier and open-weight models move from training to production, the bottleneck is increasingly serving millions of concurrent queries cheaply and quickly rather than building bigger models. Financing structured around inference-optimized silicon, rather than the GPUs used to train frontier models, signals that lenders now see the inference market as large and durable enough to underwrite on its own terms — a shift that could open a new financing channel for the wave of neocloud startups built around alternative AI chips rather than Nvidia’s.