Applied Compute Raises $350 Million at $3.25 Billion Valuation

Applied Compute has raised $350 million at a $3.25 billion valuation, less than 18 months after its founding, on the back of a $100 million annualized revenue run rate.

Elad Gil led the round. The list of participants runs long: Kleiner Perkins, Lux Capital, Hanabi Capital, NVIDIA, Blackstone, Shopify Ventures, HubSpot Ventures, Diffusion, Definition, Timeless, Layer Global, Neo, Abstract, Nova, Quantum Light, First Harmonic, and Swish.

The speed of the climb is the story. A $3.25 billion valuation on $100 million of annualized revenue works out to roughly 32 times run-rate sales — a multiple that investors are clearly willing to pay for infrastructure that sits underneath enterprise AI, not on top of it.

What does Applied Compute actually build?

The company provides infrastructure for training, deploying, evaluating, and improving AI models tailored to specific enterprise requirements. Its core premise: businesses increasingly want greater control over the AI systems supporting their products and operations.

Advances in open-weight models have opened the door for companies to customize AI models using proprietary data and specialized expertise. Applied Compute supports that work through an integrated platform designed to simplify the technical demands of enterprise AI development.

Since its previous financing round, the company has more than doubled the number of enterprises it serves.

Customers include:

• Microsoft
• Shopify
• Mercor
• Harvey
• DoorDash
• Cognition
• Clay
• Decagon

That base spans life sciences, cybersecurity, hardware design, legal services, financial services, and logistics.

In short: it sells the plumbing for companies that want to own their AI models rather than rent them.

Where are the cost numbers?

Applied Compute reports that its custom model development has helped customers improve AI performance while cutting inference costs. Two customer examples anchor the claim.

For legal technology company Harvey, Applied Compute trained a model for its Review Table product that improved answer and citation quality while operating at less than half the inference cost of a frontier model alternative.

For Clay, the company trained and deployed a customized routing system that reduced production inference costs by up to 47% without reducing quality.

Those are the kinds of figures that matter to enterprise buyers watching their AI budgets. Roughly halving inference cost, at scale, changes the math on what workloads make economic sense to run.

How fast is platform adoption moving?

The company recently introduced AC2, its Agent Cloud platform, which combines model training, inference, and continuous improvement in one system.

The early usage numbers:

• More than 1,700 training runs completed on AC2 since July 2026
• More than one trillion tokens processed daily on its inference infrastructure
• About 40% of customers now use the platform for both training and inference

That last figure is the strategic one. It reflects growing adoption of Applied Compute's integrated services — customers who train and infer on the same platform are harder to displace.

The company says it intends to expand beyond these capabilities into a broader technology stack for developing, operating, monitoring, and securing enterprise AI systems.

Adoption is compounding, and the training-run count is the clearest proof.

What changes with $350 million in the bank?

Applied Compute's workforce has tripled since January 2026, and researchers and engineers make up approximately 85% of employees — a ratio that signals where the money goes.

The new capital will fund three things: expansion of computing capacity, recruitment, and accelerated deployment of its technology across enterprise markets.

The bet underneath all of it is straightforward. With a $3.25 billion valuation and $100 million annualized revenue run rate, Applied Compute is positioning its platform around growing demand for customized AI infrastructure and model ownership — and its investors are betting that enterprises will keep paying for control over the models that run their businesses.