Bessent, an AI inference startup, has raised $1.5 billion across two funding rounds, reaching valuations of $11 billion and $13 billion by providing scalable compute resources and software to help enterprises efficiently deploy and customize open source AI models. The company addresses supply constraints through a multi-cloud strategy and enables businesses to leverage fine-tuned models for specialized workflows, positioning itself as a key player in the growing AI inference market that balances cost, control, and capability against closed frontier models.
Bessent, an AI inference startup specializing in delivering software and computing capacity for companies utilizing open source AI models, has raised $1.5 billion across two funding rounds, achieving valuations of $11 billion and $13 billion respectively. The company’s CEO, Tuhin Srivastava, highlighted the growing demand for AI inference as open source models improve and post-training techniques enhance their specialization for specific tasks. To meet this surge in demand, Bessent is focused on acquiring substantial compute resources and hiring top infrastructure and research engineers to build robust software layers.
A key challenge Bessent faces is securing sufficient compute capacity in a supply-constrained market. Srivastava explained that the company maintains strong relationships with hardware providers like NVIDIA but also diversifies its compute sources across 18 different cloud providers and 19 clusters. This multi-cloud strategy provides the flexibility needed to fulfill customer demand efficiently. The company’s approach reflects the broader industry trend of balancing cost, capability, and control when deploying AI models at scale.
Investor Apoorv Agrawal from Ultimoto Capital emphasized the enormous market potential of AI inference, describing it as possibly the largest market in the world. He noted that AI applications have evolved beyond simple Q&A to complex multi-agent systems requiring thousands of inference requests per user interaction. Agrawal also stressed the importance of control and capability alongside cost, highlighting how enterprises seek to leverage their unique data and workflows without relinquishing control to external providers. Bessent’s platform enables customers to harness a portfolio of fine-tuned models tailored to specific workflows, compounding value over time.
The discussion also touched on the evolving landscape of open source AI models. Srivastava traced the journey from early models like LLaMA 3 through periods of stagnation to recent breakthroughs such as GLM 5.2, which approaches frontier-level performance. Bessent’s value proposition lies in making these models accessible and easy to use for enterprises that lack the infrastructure teams of major AI labs. The company supports a range of model sizes and types, focusing on delivering practical utility rather than just raw scale.
Finally, the conversation addressed the ongoing debate between closed frontier models and open source alternatives. While acknowledging the superior intelligence and reasoning capabilities of closed models from leading labs, Agrawal pointed out that only a handful of companies dominate that space. For many enterprises, especially those outside the top tier, post-trained open source models offer better cost efficiency, control, and the ability to build proprietary advantages. Bessent positions itself as a critical enabler for these companies, helping them achieve frontier-level capabilities without the high costs and limitations of closed models.
