Inference Chips Emerge as New Collateral in $400 Million Deal
AI-generated image · US National Wire
A shift toward cost-efficient inference infrastructure signals a market pivot away from the expensive GPU-heavy training phase.
The financial architecture supporting AI is shifting from model training to operational efficiency. TechCrunch reports that General Compute, an AI inference cloud startup, has secured a $400 million loan from tech investment firm Upper90. Notably, this deal may be the first to use inference-specific chips as collateral—hardware designed to run existing models rather than the costlier chips required to build them.
Founded by CEO Finn Puklowski and CTO Jason Goodison, General Compute is building an inference "neocloud" utilizing SN50 chips from Intel-backed chipmaker SambaNova. According to General Compute, these chips provide 16 times faster inference than GPU-based clouds and are more power-efficient, removing the need for expensive water-cooling systems. This allows for faster deployment across a wider range of data centers.
Upper90 CEO and co-founder Billy Libby, a former Goldman Sachs quantitative trader, noted that while his firm previously financed GPU purchases for Crusoe in 2021, the market for GPUs may now be over-bought. Libby told TechCrunch that the firm is now targeting the inference wave, citing the importance of open-source models.
This move reflects a broader industry trend toward reducing the cost of AI tokens and tools. TechCrunch notes that other infrastructure providers, such as TensorWave, are making similar bets via partnerships with AMD. Puklowski told TechCrunch that the deal with Upper90 represents a signal of capital organizing to challenge Nvidia's market dominance by scaling chips with superior total cost of ownership.