Palo Alto Networks Pays $4.5B for Applied Compute and Targets $1B in New ARR
Applied Compute's CEO says a few dozen people took the AI startup from zero revenue to more than $100 million in annualized run rate in just over 18 months.

TL;DR
- Palo Alto Networks agreed on November 18 to buy AI post-training startup Applied Compute for $4.5 billion in stock.
- The price, at most 45 times a company-stated run rate, rests on Palo Alto's sales force multiplying that revenue.
- Palo Alto expects Applied Compute to add more than $1 billion in annual recurring revenue by fiscal 2029.
Palo Alto Networks, the cybersecurity company, signed a definitive agreement, announced on November 18, to buy Applied Compute, which helps companies train and run their own AI models, for $4.5 billion in stock. Applied Compute was founded in 2025. The price is subject to adjustments and antitrust clearance, Palo Alto's 8-K says. Applied Compute CEO Yash Patil posted that a few dozen people took the company from no revenue to more than $100 million in annualized run rate in just over 18 months. Palo Alto executives repeated that figure on the earnings call.
That puts the price at up to 45 times the run rate Patil posted. Take out the roughly $500 million of cash and prepaid compute that Palo Alto says comes with the company, and it's about 40 times. Palo Alto CEO Nikesh Arora told analysts on November 18 to expect Applied Compute to add more than $1 billion to next-generation security ARR by fiscal 2029, by pairing it with Palo Alto's "global go-to-market reach." That's about 10 times its current run rate.
Researchers who sit with the customer's engineers
Patil, Rhythm Garg and Linden Li were researchers at OpenAI before founding Applied Compute, SiliconANGLE reported. Patil wrote that the founding belief was that every company should own its intelligence. Joining Palo Alto will "accelerate our traction with the world's largest organizations," Patil said in the release. The company's researchers work beside a customer's engineers to train models on that customer's data and quality standards. Its site sells this as Embedded and Managed partnerships.
Applied Compute's case studies report these results:
- Clay cut inference costs for its Claygent research agent by up to 47% at the same quality with a router Applied Compute trained, according to Clay head of AI Jeff Barg (October 27).
- DoorDash cut the share of low-quality menus by about 30% against its baseline in an A/B test, then rolled the model out to all US menu traffic (February 17).
- Harvey got a Review Table model with better answer and citation quality at less than half the inference cost of a frontier alternative (November 2).
- Cognition got bug detection 10 times faster than its frontier alternative, now in production in Windsurf (May 11).
Microsoft, Shopify, Mercor and Decagon are also customers, the company said on November 2.
Palo Alto wants to sell open-weight models securely
Patil becomes co-chief AI officer at Palo Alto when the deal closes. The release pitches the deal as secure open-weight AI for large companies. It says, "Today, no large provider is delivering those models securely." Applied Compute will keep serving its current customers, and Palo Alto will bring the platform to its own global customer base, the release says. Patil wrote that customers will still own and control their models.
Arora told analysts Palo Alto is training models for more than 35 specialized security use cases and plans more than 100 megawatts of inference capacity over three to five years, on capital spending of less than $3 billion through fiscal 2030. CFO Dipak Golechha expects the deal to cut operating margin by up to one percentage point in fiscal 2027 and two in fiscal 2028, according to MarketBeat. Arora said Palo Alto first looked at Applied Compute as a prospective customer and evaluated nine providers before choosing it, Calcalist reported.
"If it works, I think this will dwarf the ability that we have in cybersecurity," Arora said on the call, according to CNBC.
Palo Alto's first-quarter results, released the same day:
- Revenue of $3.44 billion, up 39%.
- Next-generation security ARR of $9.73 billion, up 66%.
- Prisma AIRS, its AI security product, at nearly $170 million in ARR and more than 1,000 customers within five quarters of general availability, Arora said.
Its shares rose more than 7% after the report.
The closest recent deal is smaller and of a different kind. CrowdStrike agreed on November 3 to buy Fiddler AI and Dreadnode, which monitor and test AI agents, for about $495 million combined. The Wall Street Journal reported that most of the price is cash.
Cash and a research team are part of the price
On November 2, Applied Compute said it had raised $350 million at a $3.25 billion valuation in a round led by Elad Gil, with Nvidia and Shopify's and HubSpot's venture arms joining. In April it raised $80 million at $1.3 billion, led by Kleiner Perkins. Palo Alto's price, 16 days after the November round, is about 38% above it. Arora has said Palo Alto re-vests acquired founders for three or four years because "you're buying a team," in an interview with David Senra that Calcalist reported on November 16.
Applied Compute's run rate is company-stated, and neither company has published retention or customer concentration. Without those, Palo Alto's price buys a team and its cash as much as a revenue line.
Palo Alto's sales force takes over the growth at closing
Palo Alto expects the deal to close in its fiscal third quarter of 2027, according to its 10-Q. The fiscal 2027 guidance it gave on November 18 excludes Applied Compute.
If it works, I think this will dwarf the ability that we have in cybersecurity.
Palo Alto's CFO expects the deal to cut operating margin by up to two percentage points in fiscal 2028, before it returns to 29% to 30% in fiscal 2029.



