A new financing model is emerging in AI infrastructure: lenders are now willing to back loans using specialized inference chips as collateral, a shift that could fragment Nvidia's dominance in the compute market. General Compute, an AI inference cloud startup, just secured a $400 million loan from Upper90, a tech investment firm, using SambaNova inference chips as collateral. This may be the first major deal to put inference-specific chips at the center of a large-scale financing arrangement, signaling that markets are ready to bet on alternatives to the graphics processing units (GPUs) that have dominated AI infrastructure spending.

Why Are Lenders Suddenly Interested in Inference Chips?

For years, GPU financing was considered risky. When Upper90 first financed GPU purchases for Crusoe Energy in 2021, the market viewed it as a bold, uncertain bet. But as CoreWeave proved that chip-backed lending could work, and eventually went public on the strength of that model, traditional lenders began to understand the value proposition. Now that GPU markets are mature and potentially oversupplied, Upper90 is looking to the next wave of AI infrastructure.

Inference chips are different from the GPUs used to train large language models (LLMs), which are AI systems trained on vast amounts of text data. Inference chips are optimized to run already-trained models quickly and efficiently, which is what most businesses actually need. General Compute's SambaNova SN50 chips, built by an Intel-backed chipmaker, are designed specifically for this task. They consume less power, don't require expensive water-cooling systems, and can be deployed across a wider variety of data centers than traditional GPUs.

"When we financed Nvidia GPUs as the first group to do that, the market was inefficient. We could really put together something as an early participant, and kind of get compensated for the risk," said Billy Libby, co-founder and CEO of Upper90.

Billy Libby, Co-founder and CEO, Upper90

General Compute claims its SambaNova-based infrastructure will deliver 16 times faster inference than GPU-based clouds, a significant performance advantage for companies running AI models in production.

What Does This Mean for the Broader AI Market?

The financing deal reflects a growing market thesis: open-source AI models are becoming enterprise infrastructure, and the companies that can serve them cheaply will win. Companies like OpenRouter and Fireworks, which provide access to open models, have raised new funding rounds at substantial valuations. Meanwhile, new open models like Kimi's K3 have proven competitive with the latest releases from Anthropic and OpenAI on coding benchmarks, suggesting that enterprises don't always need the most expensive frontier models.

The shift also highlights a fundamental challenge for any startup trying to scale AI infrastructure: getting access to the chips you need. Nvidia's dominance means that alternative chip suppliers often struggle to find buyers. By partnering with Upper90 to finance chip purchases, General Compute is solving a chicken-and-egg problem: the startup gets the hardware it needs to scale, and Upper90 gets exposure to the next generation of AI infrastructure.

How Are Other Companies Positioning Themselves in This Shift?

General Compute is not alone in betting on non-Nvidia alternatives. TensorWave, another AI infrastructure company, is making a similar bet by partnering with AMD. As more alternative chip suppliers emerge and prove their value, compute providers that aren't locked into exclusive Nvidia deals may gain a competitive advantage in offering cost-efficient inference services.

• Cost Efficiency: Inference-specific chips like SambaNova's SN50 offer lower total cost of ownership than GPUs, making them attractive for companies running production AI workloads at scale.
• Deployment Flexibility: These chips don't require expensive water-cooling infrastructure, allowing them to be deployed across a broader range of data centers and edge locations.
• Market Fragmentation: As alternative chip suppliers gain traction and attract capital, Nvidia's historical monopoly on AI compute is beginning to fragment, creating opportunities for specialized infrastructure providers.

"There are a bunch of chips that are starting to scale that have amazing total cost of ownership, or that can operate much faster than Nvidia, but there's not too many buyers for them. By getting together with Upper90, this is not just a cool startup getting some money to buy some compute. Like, this is the first signal of capital organizing itself and the fragmenting of Nvidia's monopolistic dominance," said Finn Puklowski, CEO of General Compute.

Finn Puklowski, CEO, General Compute

General Compute itself is a young company. Founded by CEO Finn Puklowski and CTO Jason Goodison, it raised a $15 million seed round in May 2026 to build what's known as a neocloud, a computing infrastructure purpose-built for AI workloads rather than the general-purpose infrastructure offered by traditional cloud providers like Amazon Web Services (AWS) or Microsoft Azure.

The $400 million loan from Upper90 represents a significant validation of this model. It's not just capital flowing to a startup; it's a signal that sophisticated investors believe inference infrastructure, powered by specialized chips outside Nvidia's ecosystem, is the next frontier of AI infrastructure investment. As enterprises continue to seek cost-effective ways to run AI models in production, this financing trend is likely to accelerate.