Applied Compute
Category: AI Infrastructure
AI supercloud for training, inferencing, and continuously improving custom open-source models on enterprise data. Applied Compute was founded in 2025. The company is led by Yash Patil. Based in San Francisco, United States. Team size: 11-50. Total funding raised: $160M. Latest round: Series B. Key investors include Kleiner Perkins, Elad Gil, Lux Capital, Greenoaks, Neo, Hanabi Capital, Benchmark, Sequoia Capital, Conviction, Victor Lazarte, Omri Casspi, Definition.
- Founded
- 2025
- Headquarters
- San Francisco, United States
- Team size
- 11-50
- Total funding
- $160M
Value proposition
Helps enterprises run and customize open-source AI models with their own data, training proprietary "specific intelligence" agents that continuously improve from production usage — all within the customer's secure environment.
Products and solutions
Applied Compute Agent Cloud (one platform to train, run, and improve owned models), training infrastructure for RL post-training and long-horizon agents, production-grade inference/serving, self-distillation and online RL for continuous model improvement, Bring Your Own Harness (BYOH) to AC2.
Unique value
"The Best AI is Built, Not Bought" — frontier-lab-veteran research team embedded with customers to train custom models on their own data, with full data sovereignty (SOC 2, ISO 42001, VPC deployment) and continuous improvement from production signals.
Target customer
Enterprises and AI-native companies (Fortune 500 and startups) that want to own and customize their own AI models/agents rather than rely on general-purpose frontier models.
Industries served
Enterprise AI, legal (Harvey), food delivery (DoorDash), coding assistants (Cognition), AI training data (Mercor), biology (Latch Bio), customer support (Bridge).
Technology advantage
Deep RL post-training expertise from OpenAI alumni (o1 reasoning, Codex); trains on clusters of several thousand GPUs; model-flexible stack supporting open-weight models from 1B to 1T+ parameters; online RL and self-distillation for continual learning; single control plane across serverless and customer VPC.
How they differentiate
Unlike pure inference platforms (Together, Baseten, Fireworks), Applied Compute embeds a frontier-lab research team with customers to co-design evals and train custom models on proprietary data, combining training + serving + continuous improvement in one platform with full data sovereignty.
Main competitors
Together AI, Baseten, Fireworks AI
Key partnerships
NVIDIA (post-training partner for NVIDIA Nemotron), Microsoft/Satya Nadella engagement on enterprise model ownership.
Notable customers
DoorDash, Cognition AI, Mercor, Harvey, Bridge, Latch Bio
Major milestones
Founded early 2025 by three ex-OpenAI researchers, $20M seed led by Benchmark (June 2025), Emerged from stealth with ~$80M total capital / Series A ~$60M (Oct 2025), $80M Series B at $1.3B valuation led by Kleiner Perkins (April 2026), In talks for ~$3B valuation round led by Elad Gil (Aug 2026)
Growth metrics
Total funding $160M; valuation $1.3B post-money (April 2026); in talks for ~$3B valuation round led by Elad Gil.
Market positioning
Fast-scaling AI infrastructure startup valued at $1.3B (April 2026), in talks to double to ~$3B on strong open-source customization and enterprise data-sovereignty demand.
Geographic focus
United States (San Francisco); global enterprise customers.
About Yash Patil
Ex-OpenAI Member of Technical Staff (Codex programming assistant team); Stanford University. 23-year-old founder who helped build OpenAI's agentic coding systems.
Latest news about Applied Compute
More AI Infrastructure companies
Official website: https://www.appliedcompute.com/
