By Hamza Faisal
Applied Compute has raised $350 million at a $3.25 billion valuation, less than 18 months after the AI infrastructure startup was founded. The round, announced on Monday, underlines how quickly investors are willing to pay up for companies sitting in the picks-and-shovels layer of the AI boom.
The funding was led by prominent angel investor Elad Gil, with participation from Nvidia, Kleiner Perkins, Lux Capital, Hanabi Capital, Blackstone, Shopify Ventures and HubSpot Ventures, among more than a dozen others. Alongside the raise, the San Francisco company disclosed that its annualized revenue run rate has passed $100 million, up from zero at founding.
A valuation ladder built in 17 months
Applied Compute's valuation has climbed steeply at each step. Founded in 2025 by former OpenAI researchers Yash Patil, Rhythm Garg and Linden Li, the company first raised $20 million from Benchmark's Victor Lazarte at a $100 million valuation in mid-2025, then an $80 million round led by Kleiner Perkins at $1.3 billion. The latest round lifts that figure to $3.25 billion, roughly 32 times its annualized revenue run rate.
The company builds and runs custom AI models for enterprises on top of open-weight foundations. Its customers include Microsoft, DoorDash, Shopify, Harvey, Cognition and Decagon, and its inference service now handles around one trillion tokens a day, just 12 weeks after launch. About 40 percent of its customers use it for both custom training and inference, according to industry coverage of the announcement.
What the platform actually does
Applied Compute operates a cloud platform that developers use to train large language models and run inference, the process of getting a trained model to answer queries. Its training approach, called asynchronous reinforcement learning, lets a model's data-generation and weight-adjustment steps run at the same time rather than waiting on each other, speeding up development. A second technique, self-distillation, removes the need for separate teacher models that supervise training runs.
On the inference side, an autoscaling engine adjusts hardware allocation to match user traffic, and an online reinforcement-learning feature periodically fine-tunes a model's weights using the prompts it receives. The company also offers Shift-1, a model router that sends each request to the language model that can answer it most cost-efficiently, or splits complex tasks across several models.
Why Nvidia's cheque matters
Nvidia's participation is the detail rivals will watch closest. Every training run and inference workload on Applied Compute's platform creates demand for accelerators, and a stake in the layer that decides which chips get bought is strategically valuable for the chipmaker. The company said the new capital will expand its market presence and buy more infrastructure, a familiar refrain in a sector where compute capacity is the binding constraint.
The round also arrives as the inference market heats up across the board. Fal, another AI infrastructure startup, doubled its valuation to $16 billion in a $1.3 billion round announced the same week, while rivals such as Baseten, Fireworks AI and Modal are all raising at higher valuations. The question for Applied Compute is whether its revenue curve keeps bending fast enough to justify a multiple that assumes it does.
