Palo Alto Networks Buys Applied Compute to Build, Not Just Guard, AI

Palo Alto Networks Buys Applied Compute to Build, Not Just Guard, AI

Analyst: Fernando Montenegro
Publication Date: November 19, 2026

Palo Alto Networks will pay $4.5 billion in stock for Applied Compute, a platform for post-training and serving open-weight AI models. The deal moves the company from securing models to building and running them, which brings a new buyer and a new set of competitors.

What is Covered in this Article

  • The deal: An all-stock, $4.5 billion purchase of an 18-month-old company founded by former OpenAI researchers, whose CEO will become Co-Chief AI Officer.
  • The motive: Palo Alto Networks argues enterprises need open-weight AI delivered securely; we read it as a capability and talent purchase as well.
  • The seam: The platform’s natural buyer is the machine learning (ML) or platform team, not the chief information security officer (CISO), and customers on other platforms will want assurances on neutrality.
  • The new competitors: Hyperscalers, data platforms, AI clouds, model makers, and enterprise infrastructure stacks, not only security peers.

The News: Palo Alto Networks announced on November 18, 2026, a definitive agreement to acquire Applied Compute, which builds a platform for post-training and serving open-weight AI models, in an all-stock transaction valued at $4.5 billion, subject to adjustments. The deal requires regulatory clearance and is expected to close in Palo Alto Networks’ fiscal third quarter of 2027. Applied Compute’s team will join Palo Alto Networks, its co-founder and CEO will become Co-Chief AI Officer, and the platform will continue to serve existing customers. The announcement came alongside the company’s fiscal first-quarter 2027 results.

Applied Compute was founded in 2025 in San Francisco by three former OpenAI researchers. Its AC2 cloud trains and serves open-weight models using asynchronous reinforcement learning, and the company pairs it with a model router and a research team that works with customers on post-training. In early November, it reported annualized recurring revenue above $100 million and raised $350 million at a $3.25 billion valuation, in a round led by Elad Gil with Nvidia among the participants.

Palo Alto Networks, based in Santa Clara, California, reports 70,000+ customers across network, cloud, security operations, AI, and identity security. Its acquisitions over the past two years include CyberArk (about $25 billion, closed February 2026), Chronosphere ($3.35 billion, closed January 2026), Protect AI (2025), and Koi, Portkey, Embrace, and Console (all 2026).

Palo Alto Networks Buys Applied Compute to Build, Not Just Guard, AI

Analyst Take: Palo Alto Networks’ argument for Applied Compute, in its own terms, is that enterprise AI will be hybrid: many organizations will run open-weight models tuned to their own data, cost, and control needs next to closed frontier models, and no large provider delivers the open-weight half securely today. It intends to be that provider. That places the company underneath the models, shaping how they are trained and served, while most of the security industry’s AI deals have landed beside them, inspecting what goes in and out.

We see a second motive running alongside that one. Applied Compute is built around its research team, its CEO will become a Co-Chief AI Officer, and the release says the team will help train Palo Alto Networks’ own large and small models, so this is as much a capability and talent purchase as a platform one. The price, nearly 40% above the valuation set in a funding round two weeks earlier, fits that read. The two motives are compatible, and either way this is the company’s first acquisition whose core product is AI infrastructure rather than security or observability.

Open Weights Are Where the Controls Are Thinnest

Organizations that tune open-weight models take on work that a closed-model application programming interface (API) hides from them: curating training data, deciding what a model may learn from production prompts, securing the serving endpoints, and proving later what changed between versions. Applied Compute’s platform refines weights from live prompts, which makes provenance and change control part of the product, and also gives poisoned or sensitive prompts a path into the weights. Its founders had already listed “the systems that run, monitor and secure” models on their roadmap, so the security angle predates the deal.

Buyers describe the same gap. In the Futurum 1H 2026 Cybersecurity Decision-Makers Survey (N=897), 31% of respondents placed a shortage of engineers who can secure continuous large language model (LLM) fine-tuning among their top three frictions in deploying internal generative AI. Every option in that question drew between 31% and 39%, so the friction is spread widely.

A Security Vendor That Also Ships the Model

The seam we would watch most closely is the buying center. A post-training and inference platform is bought by ML, data science, or platform engineering teams, while Palo Alto Networks sells mostly to security leaders. Those groups do not always agree on who owns AI risk: in the same survey, 34% of respondents placed friction between data, application security (AppSec), and identity teams over AI security ownership in their top three. A combined offering could help bridge that divide, or it could land in the middle of it.

Neutrality is the second question. Prisma AIRS, which absorbed Protect AI, secures models wherever they run, including on hyperscaler and rival platforms. Once Palo Alto Networks also trains and serves models, customers elsewhere may reasonably ask whether its security roadmap treats every platform equally. The Unit 42 frontier AI review of Applied Compute, carried out while Palo Alto Networks evaluated it as a customer, points at a related tension: the assessor and the supplier now sit in one company.

Then there are the economics. Training and inference consume graphics processing unit (GPU) capacity, a cost profile unlike subscription software, and we expect customers and investors to ask how much of that capacity Palo Alto Networks will run itself.

The Competition Now Includes Hyperscalers and Model Makers

Most security peers have bought controls that sit around models, including Check Point with Lakera, Zscaler with SPLX, and CrowdStrike with Fiddler and Dreadnode, which we covered earlier this month. Palo Alto Networks took that route itself with Portkey, as we discussed when it announced that deal. The closest precedent we can find is Rubrik’s 2025 purchase of Predibase, a fine-tuning and serving platform, at a fraction of this price.

The more consequential competition sits outside security. Selling training and serving for open-weight models puts Palo Alto Networks up against, among others:

  • the hyperscalers (AWS, Microsoft, and Google), whose AI platforms are the default for most enterprises;
  • data and AI platforms such as Databricks, which tune and serve models next to the data they already govern;
  • open-model training and inference specialists such as Together AI, Fireworks AI, Baseten, and Thinking Machines Lab;
  • model makers selling private and sovereign deployments, such as Mistral AI and Cohere;
  • enterprise infrastructure stacks such as Red Hat AI, often paired with Nvidia, which is also an investor in Applied Compute.

Indirectly, it also competes with the closed frontier labs, since the hybrid thesis is partly an argument for running fewer workloads through their APIs. Most of these vendors win the ML team’s budget on cost per token, GPU efficiency, and model quality, and few lead with security. That gives Palo Alto Networks a distinct pitch, made to a buyer it does not usually call on.

The hyperscalers are the incumbent to beat. In the same survey (N=904), 37% of respondents placed reliance on the built-in controls of AWS, Microsoft Azure, and Google Cloud among their top three approaches to AI security for the next 12 to 18 months. Sovereign and hybrid deployments are where the Palo Alto Networks pitch departs most clearly from that default.

Settle Who Owns the Tuned Model Before the Proof of Concept

Before a proof of concept, ask who holds the tuned weights, the training-data lineage, and the version history if the platform contract ends, and whether a model can move to another inference provider without retraining.

At contract time, ask how Palo Alto Networks security controls will apply to models running on other platforms, and get that commitment in the agreement, not only on the roadmap.

The hurdle for Palo Alto Networks is to keep AC2 open to any security tooling while showing that its own security stack works as well on other platforms as it does on AC2.

For more information, read the full announcement from Palo Alto Networks.

What to Watch

  • Does Applied Compute’s customer base hold through close? AI-native companies may hesitate to tune models on a platform owned by a security vendor; churn before the fiscal Q3 2027 close would be an early signal.
  • When does AC2 appear in Prisma AIRS packaging? A combined stock-keeping unit (SKU), or Prisma AIRS controls running natively on AC2 endpoints, would show how quickly the integration moves beyond the research team.
  • Which sovereign cloud partners sign on? Named regional or national cloud partners would show where the infrastructure runs and who carries the GPU cost.

Sources

1. Palo Alto Networks Buys Applied Compute to Build, Not Just Guard, AI


Declaration of generative AI and AI-assisted technologies in the writing process: This content has been generated with the support of artificial intelligence technologies. Due to the fast pace of content creation and the continuous evolution of data and information, The Futurum Group and its analysts strive to ensure the accuracy and factual integrity of the information presented. However, the opinions and interpretations expressed in this content reflect those of the individual author/analyst. The Futurum Group makes no guarantees regarding the completeness, accuracy, or reliability of any information contained herein. Readers are encouraged to verify facts independently and consult relevant sources for further clarification.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Read the full Futurum Group Disclosure.

Other Insights from Futurum:

Qodo Builds the Governance Layer Agentic SDLC Has Been Missing RSM Bets $1B on Outcome-First Delivery for the Middle Market Dropbox Bets on AI Infrastructure With Tigris Data Acquisition

Author Information

Fernando Montenegro

Fernando Montenegro serves as the Vice President & Practice Lead for Cybersecurity & Resilience at The Futurum Group. In this role, he leads the development and execution of the Cybersecurity research agenda, working closely with the team to drive the practice's growth. His research focuses on addressing critical topics in modern cybersecurity. These include the multifaceted role of AI in cybersecurity, strategies for managing an ever-expanding attack surface, and the evolution of cybersecurity architectures toward more platform-oriented solutions.

Before joining The Futurum Group, Fernando held senior industry analyst roles at Omdia, S&P Global, and 451 Research. His career also includes diverse roles in customer support, security, IT operations, professional services, and sales engineering. He has worked with pioneering Internet Service Providers, established security vendors, and startups across North and South America.

Fernando holds a Bachelor’s degree in Computer Science from Universidade Federal do Rio Grande do Sul in Brazil and various industry certifications. Although he is originally from Brazil, he has been based in Toronto, Canada, for many years.

Related Insights
Ingram Micro ONE 2026: When the Channel's Next Interface Is an Agent
November 19, 2026

Ingram Micro ONE 2026: When the Channel’s Next Interface Is an Agent

Fernando Montenegro, Vice President & Practice Lead, Cybersecurity & Resilience at Futurum, Ingram Micro ONE 2026 positions AI agents as the critical interface for channel partners, addressing governance, security, and...
Autonomous Offensive Frameworks Are the Third Wave of AI-Driven Attacks
November 19, 2026

Autonomous Offensive Frameworks Are the Third Wave of AI-Driven Attacks

Fernando Montenegro, Vice President & Practice Lead, Cybersecurity & Resilience at Futurum, Autonomous offensive frameworks conduct real-world intrusions, exploiting forgotten credentials to achieve privileged compromise in hours....
Cato's AgentSense Moves SASE Toward Controlling What AI Agents Do
November 19, 2026

Cato’s AgentSense Moves SASE Toward Controlling What AI Agents Do

Fernando Montenegro, Vice President & Practice Lead, Cybersecurity & Resilience at Futurum, Cato Networks' AgentSense blocks untrusted content from steering autonomous AI agent actions across endpoints and networks, addressing critical...
LM Config Closes the ITOps Visibility Gap for AI-Ready Infrastructure
November 19, 2026

LM Config Closes the ITOps Visibility Gap for AI-Ready Infrastructure

LogicMonitor's new LM Config solution closes critical visibility gaps in IT operations by natively integrating configuration management with infrastructure monitoring, enabling enterprises to deploy trustworthy agentic AI in their environments....
T-Rex Adds Former Border Patrol Chief to Sharpen Federal AI Credibility
November 19, 2026

T-Rex Adds Former Border Patrol Chief to Sharpen Federal AI Credibility

T-Rex Solutions bolsters federal credibility by appointing former U.S. Border Patrol Chief Michael Fisher to its National Security Board of Advisors, addressing security and compliance barriers to AI adoption in...

Book a Demo

Welcome

The vision behind everything in Futurum’s Custom Research practice is this: research should show you what is happening, what comes next, and what to do about it. It should be personal to each audience, easy for people to grasp, and structured so LLMs can reason over it accurately. And it should be fast and turnkey; you want answers now, not another project to carry for quarters.

Whether you are defining business, channel, or go-to-market strategy; evaluating vendors or justifying ROI; or commissioning research to fill an emerging market need, we have your back, with a program that answers your questions with the objectivity and credibility to drive real decisions.

To do it, we bring unmatched data to bear: Futurum research, surveys, and market projections; validated market feeds; ETR’s 15 years of insight from 10,000 technology decision-makers; G2’s buyer and user data; and what our analysts hear every day. Add leading primary collection, from AI-moderated voice interviews to surveys and analyst-led interviews, all turnkey, and every project comes out credible, nuanced, and actionable.

And we don’t just drop the results in your lap. For internal work, we provide analyst-led sessions, interactive dashboards, and a range of formats. For market-facing work, Futurum delivers turnkey activation and amplification that actually gets seen, by people and by LLMs, through our media and share of voice. This is research that moves decisions and markets.

We will meet you wherever you are, from a fast-turn brief to a multi-year program, and shape the work to your goals, timeline, and budget. The right program for your moment.

If any of this is useful, I would love to talk.

Benjamin Brown, VP Custom Research, Futurum Research

Benjamin Brown

VP, Custom Research · The Futurum Group

Newsletter Sign-up Form

Get important insights straight to your inbox, receive first looks at eBooks, exclusive event invitations, custom content, and more. We promise not to spam you or sell your name to anyone. You can always unsubscribe at any time.

All fields are required






Thank you, we received your request, a member of our team will be in contact with you.