Markets
Business & Deals

Palo Alto's $4.5 Billion AI Deal Lands Two Weeks After a $3.25 Billion Valuation

Palo Alto is paying $4.5 billion in stock for Applied Compute, about 38% above the startup's private price two weeks earlier and roughly 45 times its $100 million run rate. The deal bets security moves into the model itself.

Rows of server racks in a modern enterprise data center aisle lit by blue and white indicator lights

Palo Alto Networks signed a definitive agreement on 18 November to acquire Applied Compute, a startup founded in 2025 that helps companies train, serve and improve open-weight AI models, in an all-stock transaction valued at $4.5 billion, subject to adjustments. The deal is expected to close in Palo Alto's fiscal third quarter of 2027, and Applied Compute co-founder and chief executive Yash Patil will become a Co-Chief AI Officer of the combined company.

The timing and the price tell most of the story. Two weeks earlier, Applied Compute had raised $350 million at a $3.25 billion valuation, a round led by Elad Gil with NVIDIA, Blackstone, Kleiner Perkins and Lux Capital among the participants. Palo Alto's price is roughly 38% above that mark, and roughly 45 times the $100 million annualized revenue run rate the startup said it had reached from zero in under 18 months.

The currency matters as much as the number. Palo Alto reported fiscal first-quarter results the same day: revenue of $3.445 billion, up 39%, and its shares jumped more than 7%, leaving them up 85% this year. It is buying with stock that just re-rated, at a moment when the gap between what it reports on a GAAP basis and what it asks investors to focus on is widening.

What the deal actually buys

The strategic case rests on a claim about how enterprises will run AI. Palo Alto argues the future is hybrid: a mix of open and closed models, with a large share of companies tuning open-weight models to their own data, cost and control needs. "Every enterprise is going to run a mix of open and closed models, and every one of those models needs to be secured," chief executive Nikesh Arora said in the announcement, which asserts that no large provider is delivering those models securely today. The control point, in other words, moves from the network perimeter into the model itself.

Applied Compute gives it a running start. Customers already include Microsoft, Shopify, DoorDash, Harvey and Decagon. For Harvey it trained a model that improved answer and citation quality at less than half the inference cost of a frontier alternative, and for Clay a custom router cut production inference costs by up to 47% without sacrificing quality. Its AC2 agent cloud, introduced a few months before the fundraise, has hosted more than 1,700 training runs since July, and its inference service moves more than 1 trillion tokens a day, with 40% of customers using it for both training and inference.

The price, and what it is paid in

The quarter behind the currency is strong on the metrics Palo Alto emphasizes. Next-Generation Security ARR reached $9.73 billion, up 66%, remaining performance obligations $21.0 billion, up 35%, and adjusted free cash flow margin 56.2%. On a GAAP basis, though, operating income fell to $96 million from $309 million even as non-GAAP operating income rose to $1.1 billion, and GAAP net income of $482 million was flattered by a $457 million tax benefit. Operating expenses grew faster than the old cost structure: research and development ran $798 million against $528 million a year earlier, sales and marketing $1,167 million against $820 million.

Dilution is already in the numbers. Weighted-average basic shares rose to 819 million from 679 million a year earlier, and the fiscal 2027 guidance that chief financial officer Dipak Golechha said the company is "materially raising" still projects a non-GAAP operating margin of 29.5% and an adjusted free cash flow margin of 38.0% on revenue of $14.52 billion to $14.57 billion. Each share handed over for Applied Compute carries the price the market just re-rated to.

What has to be true

For the price to work, several things have to hold at once. Enterprises have to keep running open-weight models in production at a scale that turns a $100 million run rate into something a $4.5 billion outlay can amortize. Palo Alto has to convert that platform into secured deployments, whether in the public cloud or sovereign environments, and management is already pointing to tens of thousands of virtual patches by the end of the fiscal year. Arora went further on the earnings call: "If it works, I think this will dwarf the ability that we have in cybersecurity. This could be bigger than what we do."

That "if" is the whole trade. The deal assumes the AI premium Palo Alto just paid, 38% over a price the private market set two weeks earlier, is the entry fee to a market where security sits inside the model rather than around it. If enterprises consolidate on closed models secured by their existing vendors, the buyer will have paid a premium for capability it could have built or partnered into more slowly. If the hybrid future arrives as described, the same price will read as the cost of standing where the perimeter used to be.