Applied Compute said it has raised $350 million in new funding at a $3.25 billion valuation, less than 18 months after its founding, after growing from zero to a $100 million annualized revenue run rate. The round was led by investor Elad Gil, with participation from Kleiner Perkins, Lux Capital, Hanabi Capital, NVIDIA, Blackstone, Shopify Ventures and HubSpot Ventures, among others. The San Francisco startup, founded in 2025 by three former OpenAI researchers, builds infrastructure that lets enterprises train, deploy and continually improve AI models tailored to their own work.
From zero to $100 million
The speed of the climb is the story. Applied Compute’s founders, Yash Patil, Rhythm Garg and Linden Li, started the company on the bet that businesses would want to own and improve the AI powering their products, rather than rent a generic model. Open-weight models gave them the foundation, and customer data and expertise supplied the rest.
Since its last fundraise, the company said it has more than doubled the number of enterprises it serves. Customers include Microsoft, Shopify, Mercor, Harvey, DoorDash, Cognition, Clay and Decagon, across life sciences, cybersecurity, hardware design, legal, financial services and logistics. That breadth is itself a signal: customization demand is not confined to one sector.
The customer examples show what that customization is worth in practice. For Harvey, Applied Compute trained a model for its Review Table product that improved answer and citation quality at less than half the inference cost of a frontier alternative. For Clay, it trained and deployed a router that cut inference costs by up to 47% without sacrificing quality.
Training and inference, together
Three months ago the company introduced AC2, its Agent Cloud, bringing training, inference and improvement into one platform. Since July, customers have run 1,700 training runs on AC2, and its inference service has grown to serve more than 1 trillion tokens a day. Forty percent of customers use Applied Compute for training and inference.
The team has tripled since January, and 85% of its staff are researchers and engineers. That ratio matters because the company’s pitch depends on research talent: customizing a model for a law firm or a chip designer is work. The moat is not the model weights, which anyone can download, but the systems that train, route, monitor and secure them.
Our read
This round belongs to a clear pattern in AI infrastructure funding. Money is moving from the companies that train frontier models to the companies that make those models useful inside a specific business, and investors are now following that shift.
Trajectory, another post-training startup, raised $45 million days earlier for the same thesis. The two rounds together suggest investors now see post-training as a market in its own right, rather than a feature of the big labs, and Applied Compute’s $100 million run rate puts it well ahead of that pack.
The valuation still deserves a note of caution. At $3.25 billion against a $100 million revenue run rate, the company is priced at more than 30 times annualized revenue, a multiple that assumes the run rate keeps compounding and today’s customers expand rather than churn.
The figures are the company’s own, not audited results, and deal databases record the $350 million round against a September date even though the announcement wave arrived in early November. The valuation is therefore best read as a bet on the next 18 months rather than a verdict on the last ones.
Why It Matters
For enterprises, the significance is practical. If a custom model can beat a frontier model on a task at half the inference cost, the default choice for production workloads starts to shift. That is the same bottleneck behind the year’s largest infrastructure rounds, including Fal’s rise to a $16 billion valuation on inference demand.
For the open-weight ecosystem, the round is a vote of confidence in a direction. Rather than consolidating around a handful of closed labs, enterprise AI spending is spreading to specialists that tune open models for regulated, data-sensitive work. If that spread continues, the winners will be the companies that combine research depth with reliability: uptime, cost control and quality gains.
Background
Applied Compute was founded in 2025 by Patil, Garg and Li, former OpenAI researchers, and its rounds included a Series B led by Kleiner Perkins that valued the company at $1.3 billion. The round more than doubles that valuation, and the company said it plans to use the funding to expand compute capacity, grow its team and accelerate deployment.
Source: company announcement.
0 Comments