Applied Compute has raised $350 million at a $3.25 billion valuation, adding fresh capital as the AI cloud startup reports annualized recurring revenue above $100 million. The company announced the round on Monday, with prominent angel investor Elad Gil leading it and Nvidia, Shopify Ventures, HubSpot Ventures and more than a dozen others participating.
The company said it will use the money to grow its market presence and buy more infrastructure. Its funding announcement also named Microsoft and DoorDash among its customers, while describing a business built around cloud computing for artificial intelligence development and use.
Applied Compute Raises $350M as Revenue Tops $100 Million
Applied Compute provides a cloud platform developers use to train large language models and run inference, the process of using a model to respond to queries. It also has machine learning researchers who assist customers, including with quality evaluations for newly developed models.
The company’s platform targets the work involved in training language models. A typical training stack has one component that creates model copies and assigns tasks, turning their results into training data. Another component uses that data to refine the model’s weights.
Those components usually operate synchronously, so each waits for the other to finish. Applied Compute said its asynchronous reinforcement learning approach allows data generation and weight adjustment to run at the same time, speeding up model development.
Training and Inference Tools Share One Platform
The platform also supports self-distillation, an optimization method that removes the need for teacher models. Researchers use teacher models, which are specialized neural networks, to supervise training runs.
A built-in assistant can summarize skills a language model learned during a training run. The company says this can replace researchers’ review of lengthy reasoning logs for that information.
For inference workloads, an autoscaling engine adjusts the hardware assigned to a model based on user traffic. An online reinforcement learning feature periodically uses prompts received by a model to fine-tune its weights and keep it aligned with customer requirements, according to the company.
Applied Compute offers those training and inference features alongside Shift-1, its model router. The tool directs coding agents’ requests to a language model that can fulfill each one in the most cost-efficient manner, and it can divide complex tasks among multiple models.
New Funding Follows a Revenue Milestone
Applied Compute disclosed that annualized recurring revenue has topped $100 million. The company also said it launched less than 18 months ago, giving the funding announcement a revenue milestone alongside its financing news.
The round adds to other startup financing reported by ValiantCEO, including an Active Label funding round tied to production growth. Other coverage has described capital going to AI businesses, such as MOSI’s Pre-A funding and Manus’s new funding.
Applied Compute’s founders, Rhythm Garg, Yash Patil and Linden Li, described training, inference and routing as the first parts of a larger stack in a blog post. They wrote that the company is building infrastructure for companies to own their intelligence, from models to systems that run, monitor and secure them.
The company’s stated next steps are to expand its market presence and purchase more infrastructure with the capital. Its announcement places that spending plan alongside a platform spanning model training, inference and request routing, as it reports annualized recurring revenue above $100 million.
Read SiliconANGLE’s funding report for the company’s announcement and product details.
