A New Era in the AI Market: General Compute Secures $400 Million Investment

With the development of AI technologies, market participants are now focusing not only on training models but also on the efficient inference stage. The startup General Compute has secured $400 million in credit from the investment firm Upper90. A unique aspect of this deal is that it may be the first major financial transaction in history to use specialized inference chips as collateral. This was reported by Techcrunch.com reports.

Until now, the AI market has primarily focused on expensive GPUs from companies like NVIDIA. However, the SambaNova chips used by General Compute are designed to run pre-trained models faster and more cheaply. According to TechCrunch, the market is now striving to run open-source models on cost-effective infrastructure, providing an alternative to the expensive services of giants like OpenAI or Anthropic.

The advantage of inference chips and the new strategy

General Compute was founded by Finn Puklowski and aims to create a "neocloud" designed exclusively for AI workloads, unlike traditional general-purpose cloud services like AWS or Azure. The company's SN50 chips are energy-efficient and do not require expensive water-cooling systems, allowing for faster deployment in various data centers.

Company representatives state that the new chips can perform inference 16 times faster than GPU-based cloud systems. This metric is crucial for enterprises running AI applications in real-time, as the cost and speed of each token directly impact business efficiency.

Changes in the investment market

Upper90 co-founder Billy Libby, formerly of Goldman Sachs, pioneered the practice of lending against GPU chips as collateral in 2021. At the time, traditional banks avoided such deals due to the rapid depreciation and high risk of the hardware. While NVIDIA GPUs dominate the market today, Upper90 predicts that the next wave will be driven by specialized inference chips.

Interest in new chip manufacturers like Groq and Cerebras is currently growing. Furthermore, new models like Kimi’s K3 have proven they can compete with products from leading laboratories in coding and complex tasks. This is further increasing the demand for infrastructure that supports open-source models.

In conclusion, this major deal between General Compute and Upper90 signals a diversification of AI infrastructure. Specialized systems that perform everyday tasks quickly and cheaply are becoming a key focus of the global technological race, alongside supercomputers.