Palo Alto Networks Bets $4.5 Billion On AI Training — Its CEO Sees A Bigger Business

The planned Applied Compute purchase would bring model training into the cybersecurity company and temporarily cut operating margins.

Share
Palo Alto Networks Bets $4.5 Billion On AI Training — Its CEO Sees A Bigger Business

Palo Alto Networks, which sells cybersecurity platforms for enterprises, plans to buy Applied Compute for $4.5 billion in stock. Chairman and Chief Executive Officer Nikesh Arora said on Wednesday's earnings call that the AI-training expansion could become bigger than its cybersecurity business.

Training The AI That Watches AI

Applied Compute customizes open-weight AI models and runs them for customers. Arora's proposed next step is to train an enterprise's application model alongside another model that supervises it, giving the supervisor an understanding of how the application should behave.

“If it works, I think this will dwarf the ability that we have in cybersecurity,” Arora said. “This could be bigger than what we do today.”

The idea grew from Palo Alto Networks' work on Prisma Airs, which discovers AI assets, inspects their traffic and applies security controls. Arora said the Hugging Face incident persuaded management that fixed rules could not cover every action or deception an AI agent might attempt.

That led the company to seek expertise in training models capable of interpreting and stopping dangerous AI behavior. Arora said management examined nine providers before choosing Applied Compute: “we have to get in the training business to be able to build a good supervision business.”

The expansion would extend a business built around network defense, security operations, cloud protection and identity security into helping customers build and operate AI applications. Prisma Airs already has nearly $170 million in annual recurring revenue across more than 1,000 customers, Arora said.

A Partner Competition Question

A Citi analyst asked whether selling AI infrastructure could put Palo Alto Networks in competition with customers and partners, weakening its position as a neutral cybersecurity provider.

Arora defended developing an integrated offering first: “let's make sure your Apple product works well.” He argued that effective supervision requires closely connected AI discovery, traffic enforcement and an understanding of model behavior.

The acquired team's skills would also improve Palo Alto Networks' own products, he said. The company is training more than 35 specialized security use cases and expects its in-house team of frontier AI researchers to reach 100 by year-end.

Chief Product and Technology Officer Lee Klarich said most enterprise AI uses beyond coding remain experimental. Security and accuracy are obstacles to production deployment, and he expects Applied Compute to help address both.

Compute Spending And Lower Margins

Chief Financial Officer Dipak Golechha said Applied Compute reached $100 million in annualized revenue within 18 months of its founding. Management expects it to contribute $1 billion in annual recurring revenue by fiscal 2029. The purchase includes approximately $500 million in cash and prepaid computing commitments held by Applied Compute.

Palo Alto Networks plans to use cloud partners and gradually add dedicated capacity as demand becomes visible. Golechha expects cumulative capital spending below $3 billion through fiscal 2030, primarily for GPUs and servers; Arora described an ambition for 100 to 150 megawatts of computing capacity.

Assuming the expected closing in the third quarter of fiscal 2027, Golechha projects operating margins would fall by up to 1 percentage point that year and 2 percentage points in fiscal 2028. He expects a return to 29% to 30% in fiscal 2029 and free-cash-flow margins of at least 38%, excluding data-center capital spending, from fiscal 2028 onward.

The company raised its fiscal 2027 revenue outlook to $14.52 billion to $14.57 billion, representing 26% to 27% growth. Golechha emphasized that the forecast excludes the pending acquisition.

Management plans to give investors more detailed financial projections after the Applied Compute purchase closes.

Sources

More on Palo Alto Networks

All Palo Alto Networks (PANW) coverage →