General Compute invests up to $400 million for AI infrastructure

General Compute invests up to US$400 million for AI infrastructure Credit: Disclosure General Compute announced the raising of a committed credit line of up to US$400 million to accelerate the expansion of its artificial intelligence infrastructure. The financing, provided by Upper90 Capital Management, will allow the company to expand its inference cloud, a platform designed to run large language models (LLMs) with higher performance than traditional GPU-based infrastructures. "Upper90 was early to the AI ??infrastructure opportunity, remaining selective about where we allocate capital. We focus on differentiated approaches that make computing cheaper, faster and more accessible for customers. General Compute's partnership with SambaNova gives you access to inference silicon that is up to six times more power efficient than traditional GPUs. Our role is to ensure the company has the capital it needs to scale with rapidly growing demand." —Billy Libby, co-founder and CEO of Upper90. Initially, the company will have access to US$100 million, with the possibility of increasing this amount as customer demand grows. According to the company, the objective is to quickly increase its operational capacity to serve organizations that use advanced AI models in production. AI market requires new infrastructure The accelerated growth of generative artificial intelligence has caused important changes in the way computing infrastructure is built. If in the early years of generative AI the biggest challenge was training increasingly larger models, currently the focus is on inference, the stage in which these models respond to user requests in real time. This change occurs because millions of queries are carried out daily by AI tools, requiring platforms capable of delivering quick responses, with low operational costs and high availability. According to projections cited by General Compute, global token consumption is expected to grow approximately 24 times over the next three and a half years, significantly increasing the need for specialized infrastructure for inference. Bet on specialized chips instead of GPUs Instead of following the model adopted by most large computing clouds, General Compute chose to develop its platform using specialized chips from SambaNova. These processors, known as ASICs (Application-Specific Integrated Circuits), were designed specifically for inference loads in artificial intelligence, unlike GPUs, which have a more generic architecture. According to the company, this choice eliminates part of the limitations faced by conventional infrastructures, such as high energy demand, dependence on few suppliers and the need for complex liquid cooling systems. In addition, the company claims to have more than US$300 million in previously secured hardware supplies, allowing it to accelerate the expansion of its capacity without depending on the availability of the GPU market. Up to 16 times faster inference General Compute claims that its infrastructure delivers significant performance gains when compared to traditional clouds. Among the main indicators released by the company are: Inference up to 16 times faster; Generation of the first token up to seven times faster; Throughput up to 8.5 times higher; Processing of up to 1,000 tokens per second; Application migration in less than 30 seconds. According to the company, these results allow for a much more fluid experience for applications based on artificial intelligence, reducing response time in chatbots, virtual assistants, intelligent agents and other corporate solutions. Lower energy consumption and accelerated implementation Another difference presented by the company is related to energy efficiency. While racks equipped with GPUs can consume more than 120 kilowatts and require liquid cooling systems, the chips used by General Compute operate at approximately 20 kilowatts per rack, using only air cooling. In practice, this reduces operational costs and simplifies infrastructure deployment. As there is no need for major adaptations to data centers, new equipment can be installed in a few weeks, while traditional projects often require years to come into operation. According to the company, this feature also makes its platform around six times more energy efficient when compared to conventional GPU-based solutions. Compatibility with different AI models General Compute's infrastructure was developed to operate with different language models available on the market. According to the company, its platform supports models from organizations such as OpenAI, DeepSeek, MiniMax and other AI labs, allowing customers to migrate their workloads quickly. The solution is also compatible with traditional APIs and the Model Context Protocol (MCP), a standard used in the integration between intelligent agents and different corporate applications. Upper90 reinforces investment in infrastructure for AI Upper90, responsible for the credit line, stated that the decision to invest in General Compute is related to the growth potential of infrastructure focused on artificial intelligence. According to Billy Libby, co-founder and CEO of the manager, the objective is to support companies that develop technologies capable of making computing faster, more efficient and accessible. In addition to financing, Upper90 also has an equity stake in General Compute, aligning its credit investment with the company's growth. Expansion accompanies growth in generative AI General Compute believes that the demand for specialized inference infrastructure will continue to grow as intelligent agents, corporate copilots and applications based on large language models become more present in companies. “We are the only neocloud capable of serving premium tokens — frontier-level intelligence and the largest models, served quickly,” said Finn Puklowski, co-founder and CEO of General Compute. "That's where workloads are migrating. Because we're new, we're not handcuffed to a single chip vendor. This line of credit allows us to deploy the fastest inference silicon on the planet in existing data centers and meet that demand now." With the new financing, the company intends to quickly expand its capacity using specialized SambaNova chips, seeking to meet the growing need for artificial intelligence processing at scale. The strategy reinforces a trend observed in the market: more than just investing in training increasingly larger models, companies are now directing efforts to build an inference cloud capable of offering speed, energy efficiency and scalability to support the next generation of AI applications.