Palo Alto Networks Buys AI Startup Applied Compute for $4.5 Billion in Stock
Palo Alto Networks is buying Applied Compute, a startup founded by former OpenAI researchers that helps companies train and run their own versions of open AI models. The $4.5 billion deal will be paid entirely in stock, and the startup, founded last year, has already reached $100 million in annual revenue. Palo Alto wants to sell the platform to its customers alongside security. For businesses, the signal is clear: securing open models is becoming a market of its own.
AI-processed from TNW; edited by Hamidun News
The largest cybersecurity company is buying an AI startup for $4.5 billion and says it is moving beyond its usual business. Palo Alto Networks has agreed to acquire Applied Compute and will pay only in stock. For developers and product managers, the significance of the deal lies less in the price than in the problem it points to: companies are rapidly moving to open models, and no one yet knows how to protect them properly.
What Palo Alto Is Buying
Three former OpenAI researchers founded Applied Compute last year. The startup helps companies train and run their own versions of open-weight models. These are models whose weights, the numbers that define how a neural network behaves, can be downloaded and modified. Put simply, instead of renting someone else's closed model through an API, a company takes an open one, fine-tunes it on its own data, and runs it in-house.
The startup's financials are impressive. According to The Wall Street Journal, it raised $350 million at a $3.25 billion valuation earlier this month, and its annual revenue, at the current run rate, has reached $100 million. The acquisition carries a premium over the last funding round: $4.5 billion versus $3.25 billion. CEO and co-founder Yash Patil will become co-Chief AI Officer at Palo Alto after the deal closes. The transaction still requires regulatory approval, and the companies plan to close it in the third fiscal quarter of 2027.
Applied Compute will continue serving its existing customers, while Palo Alto will sell the platform to its own customer base. The team will keep its offices and, according to company CEO Nikesh Arora, will largely remain independent. He also noted that employees have "millions of reasons" to stay for another three to four years, referring to the Palo Alto shares they receive in the deal.
Why This Is Becoming a Security Market
Arora explains the logic this way: every company will use a mix of open and closed models, and each one will need protection. In his words, no major provider today offers open models in a secure form. Palo Alto wants to let companies train, adapt, and run them in a public cloud or in their own "sovereign" environments, meaning under the customer's full control.
He backed the argument with a figure: on an analyst call, Arora said open models now account for more than 60% of token traffic, up from below 5% a year ago. These are the words of the executive of the company buying the business, not independent statistics, so they should be treated with caution. But the direction is clear: the more teams host models themselves, the more they worry about data leaks, weight tampering, and access control, all the things that used to be the responsibility of closed API providers.
The plans are ambitious. Palo Alto expects the deal to add more than $1 billion in annual recurring revenue by fiscal 2029. To do so, the company plans to build more than 100 megawatts of inference capacity, meaning the capacity to run models. Capital expenditures for this, it promises, will not exceed $3 billion through fiscal 2030. Arora said that if everything works out, the new business could eventually rival cybersecurity in scale.
Context: A Strong Quarter and Caveats
The deal was announced alongside results for the first fiscal quarter. Revenue rose 39% to $3.44 billion, next-generation security annual recurring revenue grew 66% to $9.73 billion, and net income came in at $482 million, up from $334 million a year earlier. The company raised its annual forecast: revenue of $14.52–14.57 billion and adjusted earnings of $4.26–4.30 per share. Arora linked part of the demand to a recent series of outages, partly caused by IT teams rushing to deploy AI.
There are also obvious risks. The deal requires regulatory approval. A company that grew entirely on network security is entering the model training infrastructure market, where its competitors are cloud giants and specialized startups. Integrating an independent team into a large corporation is always risky, and 100 MW of capital investment is new territory for a company of this profile.
What This Means in Practice
If you build products on open models, expect security vendors to start offering ready-made tooling: weight control, fine-tuning audits, and data isolation. If you are choosing between a closed API and self-hosting, security will increasingly enter the cost comparison. And for founders of AI infrastructure startups, the deal shows that large players are willing to pay handsomely for teams that know how to turn open models into an enterprise product.
- Nov 21, 2026Apple Is Building an AI Agent and Making It Harder for Outsiders to Get Access: Permission Now Takes 14 Steps, Not 4
- Nov 21, 2026Developer Skips Pinecone: Searching 72,450 Code Chunks in 50 ms
- Nov 21, 2026Gemini 4 Neon shrank the TS Rust binary from 47.44 to 7.97 MB in under 10 hours
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
Key Questions Answered
What Palo Alto Is Buying?
Three former OpenAI researchers founded Applied Compute last year. The startup helps companies train and run their own versions of open-weight models. These are models whose weights, the numbers that define how a neural network behaves, can be downloaded and modified. Put simply, instead of renting someone else's closed model through an API, a company takes an open one, fine-tunes it on its own data, and runs it in-house. The startup's financials are impressive.
Why This Is Becoming a Security Market?
Arora explains the logic this way: every company will use a mix of open and closed models, and each one will need protection. In his words, no major provider today offers open models in a secure form. Palo Alto wants to let companies train, adapt, and run them in a public cloud or in their own "sovereign" environments, meaning under the customer's full control.
What are the key details on Context: A Strong Quarter and Caveats?
The deal was announced alongside results for the first fiscal quarter. Revenue rose 39% to $3.44 billion, next-generation security annual recurring revenue grew 66% to $9.73 billion, and net income came in at $482 million, up from $334 million a year earlier. The company raised its annual forecast: revenue of $14.52–14.57 billion and adjusted earnings of $4.26–4.30 per share. Arora linked part of the demand to a recent series of outages, partly caused by IT teams rushing to deploy AI. There are also obvious risks.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.