Palo Alto Networks to buy Applied Compute for $4.5 billion in all-stock deal
The cybersecurity firm is paying $4.5 billion in stock for Applied Compute, a startup co-founded by former OpenAI researchers that helps companies customise and run open-weight AI models securely.

Key takeaways
- Palo Alto Networks will pay $4.5 billion in stock for Applied Compute, founded in 2025 by three former OpenAI researchers.
- Applied Compute grew from no revenue to over $100 million annually in under 18 months, according to EE Times.
- The startup builds infrastructure for post-training and inference of open-weight AI models, which carry security risks such as data leakage and tampering.
- The deal awaits regulatory approval and is expected to close in the third quarter of 2027, with Yash Patil becoming co-chief AI officer.
In this article
What the deal covers
Palo Alto Networks says it will buy Applied Compute, a startup that helps companies adapt and run open-weight AI models, in a deal worth $4.5 billion paid entirely in shares. EE Times reported the news on 20 November 2026, noting that the price and the target's distance from traditional cybersecurity caught many observers by surprise.
Applied Compute was started in early 2025 by three researchers who had previously worked at OpenAI. According to EE Times, the company went from no revenue to more than $100 million a year in under 18 months, serving major AI businesses.
Palo Alto Networks chairman and chief executive Nikesh Arora said the company would deliver secure, open-weight models to customers to enable secure AI adoption in the enterprise.
Why an AI infrastructure startup interests a security firm

Open-weight models are AI systems whose trained parameters are published openly, so any company can download, adjust and deploy them. That flexibility creates security risks: attackers may tamper with a model, data can leak, and the software can be misused.
EE Times notes that the data used to adapt a model is often a company's most sensitive material, including support tickets, contracts and customer records, and the resulting model weights can retain traces of it. Applied Compute's platform handles post-training, meaning the customisation of a base model, and inference, the stage when the model is actually run.
Tracing also matters. EE Times reports that adapted models can carry the classification of their training data, so they can be tracked and retired if that source is withdrawn.
Strategy and market reaction
The purchase fits a wider trend of large technology firms building their own AI capabilities instead of depending only on model providers such as OpenAI and Anthropic. EE Times reports that Palo Alto Networks assessed nine providers before choosing Applied Compute, in part because AI talent at big security firms is scarce and often sits in small startups.
Analysts quoted in the report saw consolidation ahead. Catharine Trebnick of Rosenblatt Securities told Reuters that AI tools are exposing vulnerabilities faster than companies can fix them, which pushes firms towards fewer, broader platforms. Sanchit Vir Gogia of Greyhound Research said the deal adds the models themselves to a company already present in firewalls, chips and identity controls.
The timing also matters. Palo Alto Networks' pending acquisition of CyberArk is expected to close in the third quarter of 2027, and EE Times suggests the combined position could let the firm embed security where enterprises train, govern and run custom models.
What comes next
The transaction still needs regulatory approval and other standard conditions. EE Times reports that it should complete in the third quarter of 2027, and that Applied Compute chief executive Yash Patil will become co-chief AI officer at Palo Alto Networks afterwards.
The longer-term plan is a platform that lets organisations train and deploy open-weight models securely, including in sovereign cloud environments. EE Times points out that Palo Alto Networks already holds credentials for securing data, such as customer-held keys, but those arrangements do not yet cover AI accelerators.
The company says it also hopes to build security-specialised models of its own. For readers, the deal marks a shift: Palo Alto Networks is aiming not just to secure AI use, but to supply some of the technology needed to train and run the models themselves.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



