Cybersecurity vendors have spent the past two years trying to figure out how to guard artificial intelligence without getting left behind by it. On November 18, 2026, Palo Alto Networks decided that guarding the perimeter is no longer enough.
The Santa Clara company announced a $4.5 billion all-stock agreement to acquire Applied Compute, an 18-month-old startup specializing in post-training and inference for open-weight AI models.
Founded in May 2025 by 23-year-old Yash Patil alongside two former OpenAI colleagues, Applied Compute has already crossed $100 million in annualized revenue. By absorbing the startup, Palo Alto Networks is stepping directly into the enterprise AI infrastructure market.
How Secure Open-Weight AI Infrastructure Actually Works
Most large corporations do not want to hand their proprietary data over to a single closed frontier model. Instead, they are moving toward a hybrid setup mixing closed models with open-weight alternatives that can be fine-tuned to their own cost, latency, and compliance requirements.
The catch is that open-weight models come with thin security controls out of the box. Applied Compute solves the operational side of this problem by handling post-training, customization, and inference so companies can adapt and serve proprietary models using their own data.
Palo Alto Networks plans to fuse that engine directly with its existing security stack and Unit 42 threat intelligence, letting enterprises train and run models with built-in guardrails across either public clouds or sovereign environments.
According to Paloalto networks, Palo Alto Networks actually discovered Applied Compute while evaluating nine different providers as a prospective customer itself, before its Unit 42 Frontier AI Assessment validated the startup’s security foundations.
That hands-on vetting matters because enterprise buyers have been hesitant to trust sensitive corporate data to young AI startups, even ones with early clients like Microsoft, DoorDash, and Harvey. Once the deal closes in fiscal Q3 2027, Patil will step in as Co-Chief AI Officer, and his research team will also help train Palo Alto Networks’ own internal large and small AI models across its broader product line.
The Financial Bet and New Competitive Battleground
CEO Nikesh Arora made it clear during the company’s earnings call that this is not a defensive tuck-in acquisition, telling investors that the Applied Compute business could eventually grow larger than Palo Alto Networks’ core cybersecurity operations. Management expects the new unit to generate more than $1 billion in annual recurring revenue within its Next-Generation Security segment by fiscal 2029.
Getting there will require real capital and some near-term financial discipline. Running GPU-heavy training and inference carries a very different cost structure than traditional subscription software.
Palo Alto Networks expects the deal to trim its non-GAAP operating margin by up to one percentage point in fiscal 2027 and two percentage points in fiscal 2028 before returning to its 29% to 30% range in 2029, while keeping compute capital expenditures under $3 billion through fiscal 2030.
The bigger strategic test lies in who Palo Alto Networks now has to fight. Selling post-training and inference puts the company head-to-head with hyperscalers like AWS, Microsoft Azure, and Google Cloud, as well as data platforms like Databricks and specialized inference providers such as Together AI and Fireworks AI.
Most enterprises still default to their cloud provider’s native tools, which means Palo Alto Networks will have to win on hybrid flexibility and strict data sovereignty. If the integration holds up under real enterprise workloads, the company will have successfully shifted from being a vendor that merely secures AI to one that builds and runs it.




