Palo Alto to Acquire Applied Compute for $4.5 Billion
Announced with Q1 fiscal earnings, the purchase elevates Palo Alto Networks to the post-training layer for open-weight models, paying $1.25 billion above two-week-old valuation.

Palo Alto Networks finalized a deal on Wednesday (18) to acquire Applied Compute for $4.5 billion, fully paid in stock and subject to adjustments, according to the company’s announcement. The San Francisco startup, founded by three former OpenAI researchers, offers a post-training and inference platform for open-weight models: the client takes a public model, fine-tunes it with their own data, and runs it in their controlled infrastructure. The closing is expected in the third fiscal quarter of 2027 for the buyer.
The announcement coincided with the release of the first quarter fiscal results for 2027, ended October 31. Revenue rose 39% to approximately $3.44 billion, and the ARR for Next-Generation Security, the metric the company uses to measure customer migration to the platform, grew 66% year-over-year. Palo Alto had projected 63% in September, within a range of $9.54 billion to $9.56 billion. Shares jumped over 7% following the announcement, and the company raised its fiscal year projections.
The Price of a Year of Life
Applied Compute announced a $350 million funding round led by Elad Gil at the beginning of November, with a post-money valuation of $3.25 billion. Two weeks later, Palo Alto is paying $1.25 billion above that mark. In April, the same company was valued at $1.3 billion following an $80 million investment led by Kleiner Perkins. The startup claims to have surpassed $100 million in annualized revenue in less than 18 months, but has not disclosed full annual revenue or profit.
The implied multiple, at a maximum of 45 times the reported annualized revenue, places the deal within the range of AI acquisitions currently accepted by the market without immediate penalty. Some analysts quoted by the American press consider the price high even by this standard; others view the value as a ticket to enter a market that still has no owner, that of protecting open models within the company. Both interpretations depend on a premise that no financial statement can yet confirm: that demand for customized open models will grow at the pace that the startup has demonstrated thus far.
Arora's Thesis
"Every company will run a mix of open and closed models, and each of these models needs to be protected," said Nikesh Arora, chairman and CEO of Palo Alto, in the statement. This sentence summarizes the bet: the firewall and SOC provider now has a presence at the point where the model is tuned and served, not just at the network edge. Yash Patil, co-founder and CEO of Applied Compute, will assume the role of co-chief AI officer at Palo Alto following the closing. "Customers will remain owners and in control of their models," Patil stated. The startup's research team will also train the acquirer's proprietary models.
Payment in shares, rather than in cash, preserves the buyer's balance sheet and transfers part of the execution risk to the startup's shareholders. For those acquiring technology, the effect is contractual: those already under platform contract with Palo Alto may gain a layer of open models within the same package and lose bargaining power if that layer is embedded and not separable.
What Changes for Clients in Multiple Markets
In the United States, the most immediate effect falls on banks and insurers already testing open models to keep regulated data out of third-party APIs. A post-training offer sold by the same firewall provider shortens the approval cycle with the CISO, which is typically the bottleneck in these projects.
In India, where integrators such as TCS, Infosys, and Wipro maintain large managed security practices for global clients, the acquisition changes the design of service contracts: adjusting and operating open models for clients becomes a managed revenue stream, requiring engineers who understand both security and post-training, a scarce profile in the delivery centers of Bangalore and Hyderabad.
In Brazil, the read includes the LGPD and Central Bank rules on outsourcing processing. Financial institutions hesitant to send client data to models hosted outside the country theoretically gain a path to train open models in local cloud under already approved security contracts. The point of friction is the same across the three markets: the package becomes more convenient and harder to dismantle.
The real test will come in the first earnings report post-closing, when Palo Alto will need to demonstrate how much of the platform's ARR comes from clients who purchased the model layer and not just from those already paying for the firewall.
Sources
- bloomberg.comhttps://www.bloomberg.com/news/articles/2026-11-18/palo-alto-networks-to-buy-applied-compute-in-4-5-billion-deal
- cnbc.comhttps://cnbc.com/2026/11/18/palo-alto-networks-panw-q1-2027-earnings-.html
- investors.paloaltonetworks.comhttps://investors.paloaltonetworks.com/news-releases/news-release-details/palo-alto-networks-reports-fiscal-first-quarter-2027-financial
- investing.comhttps://www.investing.com/news/company-news/palo-alto-networks-to-acquire-applied-compute-for-45-billion-93CH-5037020
- siliconangle.comhttps://siliconangle.com/2026/11/03/ai-cloud-startup-applied-compute-raises-350m-after-reaching-key-revenue-milestone/