AI infrastructure startup General Compute has secured a $400 million loan from Upper90, marking a notable shift in how the market values specialized computing hardware. The financing is believed to be among the first to use inference-focused chips as collateral.
Unlike training chips built for developing large AI models, inference chips are designed to run already-trained systems with greater speed and efficiency. General Compute says this approach can deliver faster performance while reducing the energy and cooling demands often associated with large-scale AI operations.
Founded by CEO Finn Puklowski, the company raised a $15 million seed round in May to build an AI neocloud around SambaNova's SN50 chips. These chips are engineered for inference and can be deployed across more data centers without the need for expensive water-cooling infrastructure.
Upper90's backing reflects a broader trend in AI finance: investors are increasingly looking beyond the most expensive frontier-model systems and toward infrastructure that supports open-source models at lower cost. As more alternatives to Nvidia-based setups emerge, chip diversity is becoming a strategic advantage for cloud providers.
The deal also highlights how capital is adapting to a new phase of AI growth, where efficiency, specialization, and flexible hardware choices matter as much as raw computing power. The next wave of AI infrastructure may be shaped by faster inference, broader chip ecosystems, and more accessible model deployment.
