The deal. Last night Palo Alto Networks agreed to buy Applied Compute in an all-stock deal valued at $4.5 billion, subject to adjustments. Close is expected in its fiscal third quarter, February to April. Applied Compute runs a platform for post-training and serving open-weight AI models: models whose trained weights are public, so a company can tune and run them on its own hardware. CEO Yash Patil becomes Palo Alto’s co-chief AI officer.
The price, in plain math. CFO Dipak Golechha said Applied Compute reached $100 million in annualized revenue within 18 months of founding, and that the deal brings about $500 million in cash and prepaid compute. $4.5 billion over $100 million is 45 times annualized revenue. Net of that $500 million, about 40 times. Applied Compute’s revenue is Palo Alto’s figure. It hasn’t been independently reported. Palo Alto says the business adds more than $1 billion of Next-Generation Security ARR by fiscal 2029.
This is not a security acquisition. Palo Alto has bought security tools for a decade. This time it bought the thing being secured. Nikesh Arora said open-weight models grew from under 5% of token traffic to more than 60% in a year, without saying where that figure came from. His argument: enterprises want to tune models on their own data, keep them inside their own walls, and nobody large sells that securely yet. He also said this business “could be bigger than what we do today.”
The bill comes with hardware. Palo Alto plans more than 100 megawatts of inference capacity over three to five years, backed by capex of less than $3 billion through fiscal 2030. Golechha said most of it goes to GPUs and servers. A firewall company is now planning data centers.
The quarter that paid for it. Fiscal Q1 revenue rose 39% to $3.44 billion. NGS ARR rose 66% to $9.73 billion. Both growth rates include acquired businesses, notably CyberArk, which closed early this year. That helps explain why the full-year NGS ARR guide is 26% to 27%, not 66%. Full-year revenue guidance moved to $14.52 billion to $14.57 billion, from $14.10 billion to $14.20 billion. Shares rose about 9% after hours, per Reuters.
The number that matters more. Prisma AIRS, Palo Alto’s product for securing AI apps and agents, is near $170 million in ARR across more than 1,000 customers, five quarters after launch. Arora called it the fastest-growing product in company history. That is real AI security revenue, today, from customers using other people’s models. It is a better proof point than any $4.5 billion bet on what comes next.
The platform pitch, turned up. Palo Alto logged about 170 new platform deals in the quarter, a record, and says platform customers now make up two-thirds of NGS ARR. Arora’s case for owning models goes further: he told JPMorgan’s Brian Essex he “cannot go in and intercept and understand somebody else’s application,” so buyers should train a supervisory model alongside their own. In plain words, the safest AI is the AI we helped build.
The best-of-breed reply. Greyhound Research’s Sanchit Vir Gogia, quoted by CSO Online, put it sharply: “Palo Alto is asking enterprises to let the guard build the vault it guards.” A supervisor trained by the same hands as the system it watches, he said, “is a colleague, not a check.” That is the opening for independent vendors. Today’s two Cyber 60 picks, Blink Ops and Cogent, both sell neutral layers that sit across a buyer’s tools. That pitch gets louder every time a platform swallows a new layer of the stack.
The CyberMerge view. As M&A, this is an option on enterprise AI infrastructure, priced like one and paid in stock. As security, it is a bet that trust in AI will be sold by whoever trains the model, not whoever inspects it. Microsoft and Google already sell both the model and the security. Palo Alto doesn’t want to be the big security vendor that can’t.
What to watch: whether Applied Compute keeps its AI-native customers once it is owned by a security vendor, named sovereign regions and operators for the GPU buildout, and capex against that 38% free cash flow margin guide. If Prisma AIRS keeps compounding without the model shop, the market will ask what the $4.5 billion bought.
Bottom line. Palo Alto just told the market it wants to sell the AI and the lock. Buyers who want those from different vendors now have one more reason to say so in writing.
