Applied Compute

Less than 18 months after it was founded, Applied Compute says it reached 100 million dollars in annualized revenue run rate. Its own post of 2 November 2026 says the company raised 350 million dollars at a 3.25 billion dollar valuation, with the round led by Elad Gil. The business sells the idea that a company should own the model behind its products rather than rent one, and it builds that model with the customer.

The AI Supercloud for open models

The home page describes the offer as an AI supercloud for open models: train, run inference on, and keep improving any model. The platform is called AC2, short for Applied Compute Agent Cloud, introduced about three months before the funding post. It brings training, serving and continual improvement into one product. A router feature lets a customer train and serve a single endpoint that picks the best model for each request.

Train it. Serve it. Improve it.

Training covers tool-using, long-horizon agents across text, images, code and structured data, using the customer's own harness and graders, with rewards calibrated to product measures. Serving is described as production inference where the deployment can be tuned for throughput, latency and concurrency. Improvement means mining production traces for training data and updating live deployments from user feedback, with A/B tests for release candidates. The post gives examples: a model for a legal-review product that matched answer quality at under half the inference cost of a frontier alternative, and a custom router that cut one customer's inference costs by up to 47 percent.

Deployed with you from day one

Two working modes are listed. In the embedded mode the company's staff sit with a customer's engineers, from the first evaluation to deployment. In the managed mode the customer runs experiments while the company handles training, serving and infrastructure. The site states that it holds SOC 2 certification and ISO 42001 certification, offers serverless or VPC deployment, and says the customer owns the resulting model. The post names Microsoft, Shopify, Mercor, Harvey, DoorDash, Cognition, Clay and Decagon among its customers, and says its team has tripled since January, 85 percent of them researchers and engineers.

Key facts

Founded: Less than 18 months before November 2026, per its own post People: Team tripled since January, 85% researchers and engineers (company statement) Customers: Microsoft, Shopify, Mercor, Harvey, DoorDash, Cognition, Clay, Decagon (named in its post) Products or Services: AC2 platform, AC Inference, custom model training, routers, embedded and managed partnership Funding: 350 million dollars at a 3.25 billion dollar valuation, led by Elad Gil (company statement, 2 November 2026) Certifications: SOC 2, ISO 42001 (company statement) Website: appliedcompute.com