wasl.tech

Palo Alto Networks Buys Applied Compute for $4.5 Billion to Secure Open-Weight AI

Imagine the chief security officer of a big bank, staring at a dashboard late on a Wednesday evening. Her team wants to run a powerful AI model on the bank’s own data, inside its own walls, with help from a startup called Applied Compute.

The model is free to download, and the savings look enormous. But one question keeps her awake: who guards a model that lives on your own servers?

That question is now a $4.5 billion business opportunity. Palo Alto Networks announced on November 18 that it has agreed to buy Applied Compute, a startup that helps companies adapt and run open-weight AI models, in an all-stock deal valued at $4.5 billion.

The announcement landed alongside Palo Alto’s quarterly results, and the market liked both pieces of news. Shares jumped after the company reported revenue up 39 percent and raised its full-year forecast. But the acquisition is the bigger story.

It moves the world’s largest pure-play cybersecurity company out of familiar territory — protecting networks and apps — and into the layer where AI models themselves are trained, tuned and served.

What Applied Compute Brings to Palo Alto Networks

Applied Compute is barely eighteen months old, yet it has already become one of the most talked-about names in enterprise AI infrastructure. Its platform does two things that companies struggle to do on their own.

First, it takes open-weight models — powerful AI models whose weights are publicly available to download — and adapts them to a customer’s own data through post-training. A generic model becomes a specialist that knows the business, not just the internet.

Second, it runs those adapted models in production, serving their answers to real users at scale. Its AC Inference service now handles more than one trillion tokens a day, according to the company.

Customers have run more than 1,700 training runs on AC2, the company’s agent cloud, since July 2026. Four in ten customers use both training and inference. Companies are not just experimenting — they are building on the platform.

The economics are the hook. Applied Compute says a model it trained for Harvey, the legal AI company, improved answer and citation quality for its Review Table product. It did so while costing less than half as much to run as a frontier alternative.

A custom router built for Clay, the sales data platform, cut production inference costs by up to 47 percent without reducing quality. Frontier-level results at a fraction of the frontier price, running on infrastructure the customer controls.

A $4.5 Billion Bet on Open-Weight AI

The price tag tells its own story. Just sixteen days before the acquisition, Applied Compute disclosed a $350 million funding round at a $3.25 billion valuation, led by the investor Elad Gil. It also reported a $100 million annualised revenue run rate.

Palo Alto is paying roughly $4.5 billion in stock, subject to adjustments — a premium of nearly 40 percent over that fresh valuation. The deal is expected to close in Palo Alto’s fiscal third quarter of 2027, and it still needs regulatory clearance.

Applied Compute co-founder and chief executive Yash Patil is set to become Palo Alto’s co-chief AI officer once the deal closes. The startup’s team will join the cybersecurity company, and its research group will help build AI into Palo Alto’s products.

That last detail matters. Palo Alto is not just buying a product. It is buying a research team. The company says its headcount has tripled since January 2026, and 85 percent of its people are researchers and engineers.

Why a Security Company Wants the Model Layer

Palo Alto’s chairman and chief executive, Nikesh Arora, put the strategy in a single line. “Every enterprise is going to run a mix of open and closed models, and every one of those models needs to be secured,” he said, according to Reuters.

Until now, Palo Alto’s AI security work has mostly focused on the edges: watching how employees use AI tools, scanning what goes into prompts, and guarding the data that flows out. The Applied Compute deal pushes the company deeper, into the machinery itself.

Under the announced plan, security would be embedded in model training, in customisation, and in deployment. Customers would get options for public clouds and sovereign environments, built for strict compliance requirements.

There is a commercial logic behind the move, too. Applied Compute will keep serving its existing customers — AI-native companies, digital natives and large enterprises — while Palo Alto distributes the platform to its own global customer base.

Palo Alto sells to most of the world’s largest companies already. If even a fraction of them buy the platform through Palo Alto, the deal pays for itself faster than any standalone sales effort could.

Interestingly, Palo Alto did not buy blind. Arora said the company had evaluated Applied Compute as a prospective customer first. Its Unit 42 threat research team ran a Frontier AI Assessment that confirmed the startup’s security foundations.

The Numbers Behind the Deal

The acquisition arrived with strong quarterly results. For its fiscal first quarter of 2027, ended October 31, Palo Alto reported revenue of $3.44 billion, up 39 percent from a year earlier.

Its next-generation security annual recurring revenue — the subscription metric investors watch most closely — reached $9.73 billion, up 66 percent. The company also raised its forecast for the full year.

It now expects fiscal 2027 revenue of between $14.52 billion and $14.57 billion, up from its previous expectation of $14.10 billion to $14.20 billion. The raised guidance, the earnings beat and the acquisition together sent the shares sharply higher.

Context helps explain the urgency. The open-weight ecosystem has exploded in the past month alone. Meituan released its giant LongCat-2.5 model under an MIT licence, Alibaba previewed Qwen4-Max, and MiniMax shipped a 4K image model with open weights.

Every one of those releases gives enterprises another reason to run models themselves — and another surface for attackers to probe. That risk is not theoretical.

Microsoft researchers recently showed how fewer than ten reprogrammed neurons can turn an AI agent rogue, and the breach at Fireworks AI exposed customer datasets and model weights. When the model itself becomes infrastructure, securing it becomes a core security job.

What Happens Next

The deal is an agreement, not a completed merger. Regulatory clearance is still pending, and the close is not expected until Palo Alto’s fiscal third quarter of 2027 — roughly nine months away. Until then, Applied Compute continues to operate as its own company.

The real test begins after the close. Palo Alto must stitch a fast-moving startup’s platform into a giant’s product line without slowing either one down. Applied Compute’s engineers will need to keep shipping at startup speed while their work is folded into Palo Alto’s security products.

And Palo Alto must convince its customers that a security company can also be a credible AI platform provider — a role its rivals, from the cloud giants to the model labs themselves, also want. There is also the question of neutrality: Applied Compute’s appeal rests partly on being model-agnostic.

The Bigger Picture: Security Is Moving Into AI Itself

Step back, and the acquisition fits a pattern that has been building all year. AI is no longer a tool that sits on top of a company’s systems. It is becoming the system — the thing that reads the documents, writes the code and answers the customers.

That shift changes what security means. A firewall cannot check whether a model’s training data was poisoned. An endpoint agent cannot tell whether a fine-tuned model has quietly learned to leak secrets.

The defences have to live inside the model pipeline, from the first training run to the millionth served answer. Palo Alto’s bet is that enterprises will soon pay for that kind of security the way they already pay for network security.

If the bet is right, the $4.5 billion will look cheap in hindsight. If it is wrong, the company will have bought a very expensive lesson in how fast AI markets move.

Either way, one thing is certain: the race to secure open-weight AI has just acquired its biggest player yet. The bank security chief from our opening scene may soon have an answer to her sleepless question — and a sales call to take.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top