General Compute's $400M Loan Signals Serious Capital Moving Into Non-Nvidia Inference
General Compute, an AI inference startup that raised a $15 million seed round in May 2026, has secured a $400 million loan from Upper90. The collateral is the chips themselves: SambaNova SN50 inference silicon, backed by Intel.
Upper90 Has Done This Before
In 2021, Upper90 financed GPU purchases by Crusoe in what is believed to be the first loan structured against the value of advanced chips. General Compute's deal follows the same structure, roughly 27 times larger.
The logic is the same: chips hold value, they can be collateralized, and a lender willing to underwrite that value can fund infrastructure that operators couldn't otherwise afford upfront.
The SambaNova SN50
The SN50 is purpose-built for inference, not training. It doesn't require water cooling. General Compute claims SN50-based inference runs 16 times faster than GPU-based clouds.
That 16x figure is General Compute's claim. No independent benchmark was cited.
The business case rests on running open source models at lower cost than frontier LLM providers. That market is growing. Kimi's K3 is reportedly competitive with Anthropic and OpenAI models on coding benchmarks, and OpenRouter and Fireworks both raised new funding rounds at large valuations. Demand for cheap inference against open weights is real, even if 16x needs verification.
TensorWave Is Making a Similar Bet
General Compute is not alone. TensorWave is building AI infrastructure around AMD through a direct partnership, another non-Nvidia path to inference at scale.
The pattern: capital is moving toward open source inference infrastructure, and multiple teams are betting that infrastructure doesn't have to be Nvidia.
What to Watch
Whether SambaNova SN50 delivers on efficiency promises at production scale is still an open question. The chips are inference-specific, which limits flexibility if workload patterns shift.
Upper90's willingness to put $400 million behind non-Nvidia silicon is a meaningful data point. It's not a verdict on the technology, but it suggests at least one firm that has been in chip financing since 2021 thinks the hardware holds durable value.
Source: Techcrunch