Key points
- Palo Alto Networks agrees to buy Applied Compute
- Applied Compute's founders came from OpenAI
- Applied Compute's work with Chinese AI models
Palo Alto Networks (PANW) agreed on Nov. 18 to buy Applied Compute, a startup that helps companies train and run their own versions of open-weight AI models, in an all-stock deal valued at $4.5 billion. Three former OpenAI researchers started Applied Compute in 2025, and the startup said this month that it had reached a $100 million revenue run rate in less than 18 months.
The $4.5 billion deal value, which is subject to adjustments, is about 38% above the $3.25 billion valuation Applied Compute announced on Nov. 2. It's more than three times the $1.3 billion valuation Applied Compute announced in April, and about 45 times its $100 million revenue run rate. Palo Alto announced the deal on the same day it beat estimates for its fiscal first quarter and raised its full-year forecast. The shares closed up 9.9% at $339.92 on Nov. 18, before either announcement came out.
What does Applied Compute do?
Palo Alto's deal announcement describes Applied Compute as a platform for post-training and inference of open-weight AI models. Open-weight models publish their underlying settings, called weights, so other companies can download and adapt them, as we explained when Reflection AI released one in October. Post-training is additional training that tailors an already-trained model to a specific job. Inference is running that model to answer requests.
The company works mainly with open models such as Alibaba's Qwen and Moonshot's Kimi, and its engineers train them on a client's internal data before running them on Applied Compute's servers, according to Forbes. For DoorDash, Applied Compute built a model that fixes errors in restaurant menus when new merchants sign up. In August, Applied Compute opened a private beta of AC2, which it calls "the training and inference cloud for open weights," so customers can use its training tools themselves.
Microsoft, Shopify, Harvey, DoorDash, Cognition, and Clay are among the companies using Applied Compute. A revenue run rate takes current revenue and projects it over a full year, so it isn't the same as revenue Applied Compute has already collected.
Who founded Applied Compute?
Yash Patil, Rhythm Garg, and Linden Li were friends at Stanford before they became colleagues at OpenAI. Patil, now Applied Compute's chief executive, worked on OpenAI's Codex coding assistant. Garg was a core contributor to OpenAI's o1, which the founders describe as the first reasoning model trained with reinforcement learning, and Li worked on the systems and infrastructure for that training, the founders wrote in their October 2025 launch post.
The three were on a ski trip together in Lake Tahoe when Chinese startup DeepSeek released its R1 model in January 2025. They left OpenAI about five months later to build a company around open models. "Our worldview was just very different than a lot of our OpenAI colleagues," Patil said. Their ages ranged from 23 to 25 in September.
How much money has Applied Compute raised?
Applied Compute has announced three funding rounds, starting with its public launch in October 2025.
| Announced | Amount | Valuation | Led by |
|---|---|---|---|
| Oct. 29, 2025 | $80 million | Not disclosed | Not disclosed |
| April 8, 2026 | $80 million | $1.3 billion | Kleiner Perkins |
| Nov. 2, 2026 | $350 million | $3.25 billion | Elad Gil |
Sources: Applied Compute's October 2025, April 2026, and November 2026 announcements.
Benchmark, Sequoia, and Lux backed the October 2025 round. The April round brought total funding to $160 million, Applied Compute said at the time. Nvidia, Blackstone, Kleiner Perkins, and Lux Capital were among the investors in the November round. Patil said most of the new money would go toward chips to build out the company's cloud.
Applied Compute has also made a purchase of its own. On Nov. 12, Applied Compute said it had acquired Delegate, a startup founded in late 2025 by two people who had spent years at Scale AI. The company didn't disclose terms.
What does Palo Alto want with Applied Compute?
Palo Alto plans to let businesses train, adapt, and deploy open-weight models "with security built in," in a public cloud or in their own sovereign environments. Applied Compute's team will also help train Palo Alto's own AI models, and Applied Compute will keep serving its current customers.
"Every enterprise is going to run a mix of open and closed models, and every one of those models needs to be secured," Palo Alto Chairman and Chief Executive Nikesh Arora said. He added that Palo Alto saw Applied Compute's work firsthand as a prospective customer.
Security was already part of Applied Compute's pitch. In a September post, Applied Compute argued that both U.S. and Chinese models "can be deployed safely in high-stakes enterprise applications when properly sandboxed and adequately evaluated."
Palo Alto's deal announcement doesn't mention Chinese models. Reuters reported on Nov. 18, the day of the announcement, that the Trump administration is weighing a rule that would require US companies to disclose their use of Chinese AI models. DoorDash, one of Applied Compute's customers, is among the companies that received letters from Reps. Andrew Garbarino and John Moolenaar, who chair two House committees and are investigating US companies' use of Chinese AI. DoorDash co-founder Andy Fang has said the company began using a Kimi model for lower-level work. Applied Compute's case study on its DoorDash work doesn't say which base model the company used.
Palo Alto competes with CrowdStrike, Zscaler, and Fortinet, as we laid out in our cybersecurity stocks guide.
When will the deal close?
Palo Alto expects to complete the purchase in its fiscal third quarter of 2027, which runs from February through April 2027, subject to regulatory clearance and other customary conditions. Patil will become Palo Alto's co-chief AI officer once the deal closes.












