Palo Alto Networks has agreed to buy Applied Compute, a startup that helps companies customize and run open-weight AI models, in an all-stock deal valued at $4.5 billion. The company announced the deal on November 18, 2026, the same day it reported fiscal first-quarter results that beat Wall Street expectations.

The deal is a large bet on a specific idea. Big companies won't standardize on one AI provider. They will run a mix of models, some bought from frontier labs and some they host themselves, and every one of those models will need security controls. "Every enterprise is going to run a mix of open and closed models, and every one of those models needs to be secured," Palo Alto Networks chairman and CEO Nikesh Arora said in the company's press release, which was filed with the SEC as an exhibit to an 8-K.

The Deal in Brief

According to the filing, the two companies signed a definitive agreement. The $4.5 billion all-stock price is subject to adjustments, and closing depends on customary conditions, including regulatory clearance. Palo Alto Networks expects the deal to close in its fiscal third quarter of 2027. After closing, Applied Compute co-founder and CEO Yash Patil will become Co-Chief AI Officer at Palo Alto Networks.

Arora also said in the release that Palo Alto Networks saw Applied Compute's platform firsthand as a prospective customer, and that its Unit 42 Frontier AI Assessment "confirmed their strong security foundations." The release does not publish the assessment itself or describe what it covered.

What Applied Compute Actually Does

Help Net Security describes Applied Compute as a platform for post-training and inference of open-weight models. Here is what those terms mean.

  • Open-weight models are AI models whose trained parameters, the "weights," are published so anyone can download them and run them on their own hardware. A closed model, by contrast, is only available through the vendor's API.
  • Post-training is the work done after a base model is built to make it useful for a specific job, such as fine-tuning it on a company's own data or shaping how it behaves.
  • Inference means running the model to produce answers. Each prompt sent to a chatbot is an inference request.

According to Help Net Security, which reproduces the company's announcement, enterprises will be able to train, adapt and deploy open-weight models with security built in, in the public cloud or in sovereign environments. Sovereign environments are infrastructure kept under a specific jurisdiction's or organization's control. That matters to governments, regulated industries and anyone who can't send sensitive data to a third-party model API. Help Net Security also reports that Applied Compute's research team will help Palo Alto Networks embed AI across its offerings.

A Young Company With a Steep Valuation Curve

Applied Compute is barely a year old. SiliconANGLE reports it was founded in 2025 by former OpenAI researchers Yash Patil, Rhythm Garg and Linden Li. Earlier this month the startup raised $350 million at a $3.25 billion valuation in a round led by Elad Gil, with Nvidia, Shopify Ventures and HubSpot Ventures participating. SiliconANGLE says the company has raised more than $500 million in total, said at the time of that round that its annualized recurring revenue had topped $100 million, and counts Microsoft and DoorDash among its customers.

Palo Alto Networks' $4.5 billion price is therefore about $1.25 billion above the valuation investors set only weeks ago, and it is being paid in stock, not cash.

The Earnings Backdrop

The announcement came with a strong quarter. SiliconANGLE reports that for its fiscal first quarter, which ended October 31, Palo Alto Networks posted adjusted earnings of $1.05 per share against an expected 97 cents. Revenue was $3.44 billion, up 39% year over year and ahead of the $3.31 billion analysts expected. Next-generation security annual recurring revenue, the company's measure of its newer platform and subscription business, reached $9.73 billion, up 66%. Shares rose more than 8% in late trading, after a gain of nearly 10% during the regular session.

Those results give Palo Alto Networks room to make a big acquisition in stock, and investors at least initially did not punish the purchase.

Why a Security Company Wants an AI Platform

The press release presents the deal as filling a gap. Palo Alto Networks argues that enterprise AI will be hybrid, mixing open and closed models, and says no large provider is currently delivering open-weight models securely. Whether that claim holds up is debatable, but the logic behind it is clear. When a company calls a closed model through an API, the vendor handles much of the hosting, patching and abuse monitoring. When a company downloads open weights, fine-tunes them and serves them from its own infrastructure, all of that becomes the company's job. The model becomes another production workload with its own supply chain, data pipelines and attack surface.

By owning the platform where that post-training and inference happen, Palo Alto Networks says it can offer security built in rather than bolted on afterward. Whether it delivers on that is something customers will have to judge once products ship.

The deal also moves Palo Alto Networks beyond selling tools that protect AI and toward providing a platform for building and running it. That is a different business from its traditional firewall and endpoint products, and it is an aggressive move that will take years to judge.

What It Means for Defenders

This is an acquisition, not a vulnerability disclosure. Nothing needs patching and no credentials need rotating because of it. The sources list no affected versions, fixed releases or mitigations. Security teams still have a few practical points to consider.

  • Nothing changes before close. The deal still needs regulatory clearance and is expected to close in Palo Alto Networks' fiscal Q3 2027. Until then, the two remain separate companies. Applied Compute customers should not expect immediate changes to contracts, data handling or product roadmaps, but they should ask their account teams how those will be handled after closing.
  • Inventory your open-weight models. Palo Alto Networks' pitch assumes many enterprises will run self-hosted models tuned to their own data. Whatever you think of the sales pitch, list which open-weight models your organization runs, where they run, who fine-tuned them, on what data, and who can reach their inference endpoints. If you can't answer those questions, you have an unmanaged workload.
  • Treat model hosting like any other production service. Self-hosted inference servers need the same access control, logging, network segmentation and change management as other internet-facing or internally sensitive services. A sovereign or on-prem deployment makes you responsible for those controls.
  • Watch for vendor consolidation. If you run Palo Alto Networks gear, expect AI features from this acquisition to appear in its products over time, since the company says the research team will help embed AI across its offerings. Evaluate them on what they do and how data flows, not on the announcement. Federal contractors working through CMMC should check where any AI-related telemetry or training data would be processed before turning such features on.
  • Ask for the assessment. Palo Alto Networks says its Unit 42 Frontier AI Assessment confirmed Applied Compute's strong security foundations, but it has not published the assessment. Customers of either company who are evaluating the combined offering are entitled to ask what that review covered.

Sources