A source familiar with the fundraising details has revealed that US-based startup Applied Compute, founded just a year ago, is in discussions for a new funding round that could value it at approximately $3 billion on a post-money basis. This would represent a more than doubling of its valuation compared to the funding round disclosed just four months ago.
The company specializes in helping businesses deploy and customize open-source large language models using their own proprietary data. The new fundraising initiative comes on the back of a significant surge in the company’s revenue over recent months. According to the source, Applied Compute’s current annualized recurring revenue stands at roughly $50 million, a nearly threefold increase from the figure CEO Yash Patil disclosed in November of last year.
The source indicated that investor Elad Gil is leading the negotiations for this funding round, which is expected to be worth several hundred million dollars. It is currently unclear whether the $3 billion valuation includes the new capital injection. The funding round has not yet closed, and the terms are still subject to change.
Many companies are now seeking to reduce the costs of using AI by moving away from paid, closed-source models from providers like Anthropic and OpenAI. Instead, they are opting for and customizing open-source large language models, a trend that is driving up market demand. This shift is also benefiting other similar startups, including OpenRouter, which aggregates hundreds of models for developers, and inference service providers like Baseten and Fireworks, which offer training, fine-tuning, and deployment services for open-source models.
Founded in 2025 by former OpenAI researchers Yash Patil, Risam Gal, and Linden Li, Applied Compute is headquartered in San Francisco. The company builds and trains industry-specific, customized large models for sectors like finance and law. Its core technology relies on reinforcement learning to optimize models: it provides positive rewards for behaviors that achieve specific goals and imposes penalties for deviations. Its clients include food delivery platform DoorDash, code intelligence startup Cognition, and data labeling service provider Mercor.
By using these customized models, enterprises can build their own AI agents to perform various business operations on behalf of employees. Applied Compute is also developing continuous learning technology that allows these agents to accumulate experience and improve iteratively from real-world scenarios. The company generates revenue from two main sources: consulting fees for model fine-tuning services and usage fees for the computing power clients consume when running their custom models on Applied Compute’s infrastructure. The company has previously raised $160 million in funding from investors including Kleiner Perkins, Benchmark, Sequoia Capital, and Lux Capital.