Palo Alto Networks will pay $4.5 billion in stock for Applied Compute, a startup that helps companies run and customize their own AI models, the cybersecurity vendor said on Wednesday alongside fiscal first-quarter results that beat estimates. Shares rose more than 7% on the combination.
The deal is a step outside the company's core business. Applied Compute works on open-weight models, the kind a customer downloads and runs on its own infrastructure, and Palo Alto's stated reason for buying it is that securing those deployments is becoming part of what a security vendor sells. "If it works, I think this will dwarf the ability that we have in cybersecurity," chief executive Nikesh Arora said on the earnings call. "This could be bigger than what we do."
The numbers behind the quarter
Annual recurring revenue for the next-generation security business rose 66% from a year earlier to $9.73 billion, which Arora described as one of the largest beats in the company's history. Net income came to $482 million, or 57 cents a share, against $334 million, or 47 cents, a year earlier. For fiscal 2027 the company guided to adjusted earnings of $4.26 to $4.30 a share on revenue of $14.52 billion to $14.57 billion, both above the $4.18 and $14.16 billion analysts expected.
The stock has climbed about 85% this year, a run that tracks a year of intrusions and outages that made security budgets easier to defend in board meetings. Executives tied the demand to fears of AI-powered attacks, pointing to incidents such as the Hugging Face breach, which raised the profile of intrusions that arrive through data and model pipelines rather than through a stolen password. A week of failures at Apple, banks and air traffic systems did that work even though investigators found no attacker behind them, and the FBI's takedown of the FaceStealer campaign gave buyers a concrete example of what a sophisticated intrusion costs.
The patching business is the growth business
Arora's argument is that the volume of repair work is about to become the product. He said the company expects customers to need tens of thousands of virtual patches by the end of the fiscal year, applied to systems that were never designed to be updated quickly. "Whatever they do today is going to take three to six months, maybe nine months, depending on how complex and how tired their environment is," he said.
That timeline is the commercial opportunity and the risk at once. Security vendors benefit when the installed base is old and slow to fix, but a customer that spends nine months patching is also a customer exposed for nine months. Applied Compute, in that framing, is less about protecting AI models than about owning the layer where companies decide how their models run, and charging for the confidence that they run safely.
The deal is all stock, which keeps cash on the balance sheet and ties the price to Palo Alto's own valuation. It still needs to close, and the company has not said how much revenue Applied Compute generates. The first real test will come in the fiscal second quarter, when Arora's forecast has to hold and the acquisition starts appearing in the numbers.