General Compute, an AI inference-focused neocloud startup, secured a $400 million debt facility from Upper90 Capital Management on July 17, marking what appears to be the first time inference-specific chips (SambaNova SN50 ASICs) were used as collateral rather than traditional Nvidia GPUs. The credit line begins at $100 million and scales with customer demand. General Compute has already placed $300 million in orders for SambaNova chips and targets 16x faster inference than GPU-based clouds.
Upper90 co-founder Billy Libby previously financed Nvidia GPUs for Crusoe Energy in 2021 as a first mover in chip-backed lending. Since then, CoreWeave scaled GPU financing to $8.5 billion. Now, as GPUs are 'comparatively well understood and perhaps overbought,' Upper90 is pivoting to inference ASICs and non-Nvidia infrastructure. General Compute built on SambaNova's transformer-optimized chips; SambaNova itself just closed its Series F at an $11 billion valuation (July 8), a fivefold jump from $2.2B in February.
For architects: this is a watershed moment. Chip-backed financing is now flowing to alternatives outside Nvidia's ecosystem. For inference clouds and open-model deployments, SambaNova's power-efficient ASICs offer a real collateral story for lenders. The downstream risk: unlike Nvidia GPUs with an established secondary market, SambaNova ASICs have no resale history and carry architecture-dependent obsolescence risk if model formats shift. But Upper90's confidence signals that inference silicon is becoming a bankable asset class.