The archive · AI & Models · Strategic decision · 2025–2026
Applied Compute bets open models beat frontier APIs inside companies
Ex-OpenAI trio sells enterprise "Specific Intelligence"; $50M run rate and $3B valuation talks within a year.
Applied Compute
What the business is
Applied Compute provides software infrastructure and embedded engineers so enterprises can train and deploy custom AI agents in-house on open-weight models.
Starting capital：Launched October 2025 with $80M from Benchmark, Sequoia, Lux, and others; April 2026 added $80M led by Kleiner Perkins at $1.3B, total $160M
How it started
Yash Patil, Rhythm Garg, and Linden Li left OpenAI and founded Applied Compute in 2025, arguing that general frontier models were "brilliant strangers" useless for specific business workflows without a company's data, standards, and judgment.
What happened
It emerged from stealth in October 2025 with $80M and early adopters including Cognition, DoorDash, and Mercor. In April 2026, Kleiner Perkins led another $80M at a $1.3B post-money valuation; in August 2026, The Information reported ~$50M annualized revenue, nearly quadrupling the level Patil disclosed in November, and talks of a round near $3B led by Elad Gil.
How it ended up
As of August 2026 the round was not finalized; Applied Compute had ~$50M annualized revenue and was scaling deployments across Fortune 500 clients. Still live and expanding.
Background
Applied Compute was founded in 2025 by three former OpenAI researchers — Yash Patil, Rhythm Garg, and Linden Li — who had worked on Codex and the reinforcement-learning methods behind o1. Their thesis: frontier models are general, but enterprises need "Specific Intelligence" trained on their own workflows and data.
The company emerged from stealth in October 2025 with $80M from Benchmark, Sequoia, Lux Capital, Elad Gil, and others, with Cognition, DoorDash, and Mercor as early adopters. Rather than outsourcing model work, it embedded engineers with client teams and built its own training stack and agent platform.
In April 2026, Kleiner Perkins led another $80M round at a $1.3B post-money valuation. By August 2026, The Information reported about $50M in annualized revenue — nearly quadruple the level CEO Yash Patil disclosed in November — and negotiations for a round near $3B led by Elad Gil.
The growth rode a shift as enterprises tried to cut inference costs and single-vendor dependence by moving from closed frontier APIs toward tailored in-house open-source models. As of September 2026, the $3B talks had not closed, and Applied Compute was expanding across Fortune 500 customers.
What has to be true
- Open-weight models matured enough that fine-tuning beat frontier APIs on cost and accuracy for specific enterprise tasks.
- Data-control worries and vendor risk pushed companies to prefer models they run and own themselves.
- Its founders' reinforcement-learning pedigree let Applied Compute build the full training stack rather than reselling another lab's tools.
- An embedded-engineer model created switching costs and direct feedback loops that a pure software product would lack.
What can be applied
Model vendors are not the only bottleneck; selling the context layer around open models lets a startup outflank frontier labs on trust and price.
Aftermath
As of the August 2026 reports, Applied Compute was generating about $50M in annualized revenue and negotiating a round that could value it near $3B, with terms still unfinalized. It was deploying agents for clients including DoorDash and working across financial services and law, positioning itself as the infrastructure layer for enterprises that want to own their AI rather than rent it from frontier labs.
Sources
- Applied Compute In Talks For US$3 Billion Valuation
- Applied Compute Launches with $80 Million To Build Specific Intelligence For Enterprise AI Agents
- AI startup Applied Compute is in talks for new funding, with a valuation expected to double to $3 billion
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card