Pomegra
Applied Compute, a custom enterprise AI startup founded by ex-OpenAI researchers, raised $350M at a $3.25B valuation, led by Elad Gil, as its run rate passed $100M.
Key Takeaways
- Applied Compute raised $350M at a $3.25B valuation in a round led by Elad Gil.
- Annualized run rate passed $100M in under 18 months, up from $13M in November 2025.
- The valuation is 2.5 times the $1.3B set in April 2026.
Lead
Applied Compute announced on November 2, 2026 that it raised $350M at a $3.25B valuation, led by Elad Gil. The San Francisco company builds and runs custom AI models for enterprises. It said annualized revenue has crossed $100M, less than 18 months after founding.
Kleiner Perkins, Lux Capital, Hanabi Capital, NVIDIA, Blackstone, Shopify Ventures and HubSpot Ventures joined the round. Diffusion, Definition, Timeless, Layer Global, Neo, Abstract, Nova, Quantum Light, First Harmonic and Swish also participated. The company plans to spend the money on compute capacity, hiring and enterprise deployment.
What Does Applied Compute Actually Sell?
Applied Compute sells a managed platform that trains, serves and continuously improves models tuned to a customer's own data and workflows. The product is AC2, the Applied Compute Agent Cloud, which combines training, inference and feedback from production use. Its inference service processes more than 1 trillion tokens a day.
Customers include Microsoft, Shopify, Mercor, Harvey, DoorDash, Cognition, Clay and Decagon. The company says 40% of customers use it for both training and inference, and that it has completed more than 1,700 training runs on AC2 since July. Its pitch is that companies should own the AI they run on, rather than rent a general-purpose model that every competitor can also rent.
The founders are Yash Patil (CEO), Rhythm Garg (CTO) and Linden Li (Chief Architect). All three previously worked at OpenAI. The company's stated focus is reinforcement learning on proprietary data, aimed at sectors including life sciences, cybersecurity, hardware design, legal, financial services and logistics.
How Fast Has the Valuation Moved?
The valuation has risen about 2.5 times in roughly seven months. Applied Compute raised $80M at a $1.3B post-money valuation in April 2026, in a round led by Kleiner Perkins. Elad Gil, Lux, Greenoaks, Neo and Hanabi took part. That followed a seed-stage financing in October 2025 at a reported $500M valuation, backed by Benchmark, Sequoia Capital, Lux and Gil.
Revenue has moved on a similar slope. The run rate was about $13M in November 2025 and about $50M by August 2026, before passing $100M. At the new price, the company is valued at roughly 32 times its current run rate. That multiple is high for a services-heavy business, though low next to some AI infrastructure peers.
Gil has now led or joined every announced round, which makes this a follow-on by an insider rather than a fresh outside price check. The presence of NVIDIA and Blackstone adds a compute supplier and a large capital allocator to the cap table. Both have reasons to back a company that consumes a lot of GPU time.
Why Does Custom Enterprise AI Attract This Much Capital?
Investors are paying for a bet that enterprises will want models tuned to their own data rather than off-the-shelf frontier models. Applied Compute's customer list suggests the demand exists among AI-native software companies. These buyers have engineering depth, large volumes of proprietary data and a direct interest in cutting inference costs.
The risk is concentration and substitution. Many of the named customers are venture-backed technology firms, a base that can shrink quickly if funding tightens. Frontier labs such as OpenAI and Anthropic also sell fine-tuning and enterprise deployment tools, and they can narrow the quality gap that custom training is meant to exploit. Applied Compute has not disclosed gross margin, customer retention or what share of revenue is recurring versus project-based.
Headcount offers one signal about the work involved. The team has tripled since January, and 85% of staff are researchers and engineers. That mix points to a business that still depends on skilled people for each deployment, which can limit how quickly margins improve.
What Comes Next for Applied Compute?
The next test is whether the run rate can keep compounding while the company spends heavily on compute. Fresh capital gives it room to buy capacity and hire against labs and other model-customization startups for scarce researchers. A repeat of the last two rounds would imply a valuation above $8B within a year, which would require revenue to scale well past $100M.
Two scenarios are plausible. In one, enterprises standardize on owned models and Applied Compute becomes the default training and serving layer for them. In the other, frontier labs bundle comparable tools and margins compress. The share of customers buying both training and inference will be the clearest early indicator.
Outlook
Applied Compute has turned a $13M run rate into more than $100M in about a year and has tripled its valuation twice in less than 13 months. The round confirms strong investor appetite for enterprise model customization, but it also leaves open questions on margins, customer concentration and competition from frontier labs. The coming quarters will show whether revenue growth can keep pace with the valuation it now carries.
