Skip to content

AI·News & analysis

Palo Alto Networks to buy Applied Compute for $4.5 billion in stock

The cybersecurity giant announced the all-stock deal alongside a record quarter, with revenue up 39% to $3.44 billion and a sharply higher full-year forecast.

Dan KostBy Dan Kost aka Poseidan8 min readX
The three co-founders of Applied Compute smiling together in an office with plants on shelves behind them.

Tide

Ripple

Sci-fi

5/10

Reality

Shipping

A security company shopping for a model-training factory.How we rate

The Squeeze

Palo Alto Networks will buy Applied Compute, a startup that helps companies train and run their own AI models, for $4.5 billion in stock.

Applied Compute was valued at $3.25 billion two weeks ago and reports more than $100 million in yearly recurring revenue. The security company wants to sell open-weight AI models with protection built in. The deal is expected to close in its fiscal third quarter of 2027.

What to know

  1. Palo Alto Networks agreed to buy Applied Compute, a startup that helps companies train and run their own AI models, in an all-stock deal valued at $4.5 billion.
  2. The startup was valued at $3.25 billion in a funding round announced about two weeks earlier, and it says its yearly recurring revenue tops $100 million.
  3. Applied Compute's CEO, Yash Patil, will become Palo Alto Networks' Co-Chief AI Officer once the deal closes, which is expected in the fiscal third quarter of 2027.
  4. The company reported first-quarter revenue of $3.44 billion, up 39%, and raised its full-year revenue forecast to between $14.52 billion and $14.57 billion.

Palo Alto Networks said on Wednesday that it will buy Applied Compute for $4.5 billion in stock. It is a lot of money for a startup that is not yet two years old.

Now the fun part: why would a company that guards networks want a company that trains AI models?

First, what does Applied Compute do?

Applied Compute sells a platform for what the industry calls post-training and inference. Post-training means taking a finished AI model and teaching it extra skills. Inference means running the model to get an answer.

The goal is simple. A company wants an AI model that knows its own work, its own data and its own rules, and Applied Compute helps build and run it.

The startup works mostly with open-weight models. Open-weight means the model's inner settings, called weights, are published, so anyone can download the model, run it and change it.

In real life Think of a ready-made suit. Post-training is the tailor who adjusts the sleeves and the waist until it fits one person perfectly.

How did the startup grow so fast?

Applied Compute was founded in 2025 by Yash Patil, Rhythm Garg and Linden Li. It launched less than 18 months before its latest funding round.

On November 2, it announced $350 million in new funding at a $3.25 billion valuation. Investor Elad Gil led the round. Nvidia, Shopify Ventures and HubSpot Ventures joined in.

The company said its annualized recurring revenue had passed $100 million. That figure is what its customers pay in a year, if current contracts keep going.

  • Customers: Microsoft and DoorDash were named among its clients.
  • Product mix: a training platform, an inference engine and a router called Shift-1.
  • Router job: it sends each coding request to the model that can handle it at the lowest cost.

Who are the founders?

The three founders worked at OpenAI before they started the company. Their work there covered Codex, reasoning models and the systems used to train models with reinforcement learning.

Reinforcement learning is a way of teaching a model by trial and reward. The model tries a task, gets a score, and slowly learns what works.

The company came out of hiding in October 2025, backed by investors including Benchmark and Sequoia. In April 2026 it raised $80 million at a $1.3 billion valuation, led by Kleiner Perkins, which brought its total funding to $160 million.

That means the valuation roughly tripled between April and the November round. The sale price adds another jump on top.

What is special about its training method?

A normal training setup has two parts. One part makes many copies of the model and gives each a task, which creates training data. The other part uses that data to adjust the model's weights.

Usually the two parts take turns, so each waits for the other. Applied Compute uses asynchronous reinforcement learning, which lets both parts run at the same time and speeds up the work.

The platform also supports self-distillation, which removes the need for a separate teacher model. A teacher model is a second neural network that supervises the training of the first.

On the running side, an autoscaling engine adds or removes hardware as user traffic changes. A feature called online reinforcement learning uses the prompts a model receives to fine-tune it, which means making small adjustments to its weights, from time to time. That keeps it in line with what the customer needs.

What exactly was announced?

The deal is valued at $4.5 billion, subject to adjustments, and it is paid entirely in Palo Alto Networks stock. The company expects it to close in its fiscal third quarter of 2027.

Yash Patil will become Co-Chief AI Officer at Palo Alto Networks after the close. The acquisition brings the platform and its research team into the company.

By the numbers: $3.25 billion was the startup's value in early November. $4.5 billion is the price tag about two weeks later.

Why does a security company want this?

Palo Alto Networks says the future of enterprise AI is hybrid. In its view, big companies will run a mix of open-source models, open-weight models and closed frontier models, depending on the job.

Closed frontier models are the top systems from the big labs, which you can only use through the maker's service. Open-weight models can sit on a company's own servers.

The company wants to help firms train, adapt and run open-weight models with security built in. That can happen in the public cloud or in sovereign environments, which are setups kept inside a particular country's rules.

CEO Nikesh Arora said Applied Compute has built "the best post-training and inference platform we've seen." He added that its research team is exceptional.

Why it matters: Companies that handle sensitive data often cannot send everything to someone else's AI service. Owning the model and the security in one package is the pitch.

And how was the quarter?

The deal arrived with the company's first-quarter results for the period that ended October 31, 2026. Revenue grew 39% from a year earlier to $3.44 billion.

Arora called it a record first quarter, and said customers are turning to the company because of an unprecedented threat environment. His words were that AI is reshaping the threat landscape.

  • Next-Generation Security ARR: $9.73 billion, up 66%. ARR means annual recurring revenue.
  • Remaining performance obligations: $21.0 billion, up 35%. This is the value of signed contracts not yet delivered.
  • Adjusted free cash flow: $1.9 billion, a margin of 56.2%.
  • GAAP net income: $482 million, or $0.57 a share, up from $334 million a year earlier.

What did the company say about the rest of the year?

Chief financial officer Dipak Golechha said the company is materially raising its full-year guidance. Guidance is the forecast a company gives investors.

For the full fiscal year 2027, it now expects revenue of $14.52 billion to $14.57 billion, which is growth of 26% to 27%. It expects adjusted free cash flow margin of 38.0%.

For the second quarter, it expects revenue of $3.530 billion to $3.550 billion. That would be growth of 36% to 37%.

One detail is worth a look. GAAP operating income fell to $96 million from $309 million a year earlier, while the adjusted figure, which leaves out items such as stock pay and acquisition costs, rose to $1.1 billion from $746 million.

Is this a pattern for Palo Alto Networks?

It is, and the pattern is getting bigger. The company completed its roughly $25 billion purchase of CyberArk on February 11, adding identity security as a core pillar.

In January 2026 it completed the $3.35 billion purchase of Chronosphere, which helps companies watch over their cloud systems. The Applied Compute deal moves the shopping list toward AI itself.

Together, the three deals show a company that is buying its way into each new layer of the business software stack.

What it means for you

  • If you work in IT: expect security vendors to start selling AI model training and hosting as part of one package.
  • If you buy software: a company that tunes its own open-weight model may keep more control over its data.
  • If you invest: the deal is paid in stock, so it adds new shares, and the company is betting that growth pays for it.
  • If you are curious: the AI tools your bank or hospital use may soon come with security built into the model's training.

The bottom line

Palo Alto Networks is paying $4.5 billion in stock for a young startup that helps companies train and run their own AI models. It arrived with a strong quarter and a higher forecast.

The company says the future is a mix of open and closed models, and it wants to be the one that keeps them safe. The deal still has to close, which is expected in its fiscal third quarter of 2027.

Key facts

Deal value
$4.5 billion, all stock
Expected close
Palo Alto Networks fiscal Q3 2027
Quarterly revenue
$3.44 billion, up 39%
Next-Generation Security ARR
$9.73 billion, up 66%
Full-year revenue forecast
$14.52 billion to $14.57 billion

Got questions?

Quick answers, plain words

What is Palo Alto Networks buying?

It is buying Applied Compute, a startup that runs a platform for training and running open-weight AI models. The deal is valued at $4.5 billion and paid in stock.

What does Applied Compute do?

It helps companies build AI models that know their own business. Customers can train a model on their data, then run it, and the startup also offers a team of researchers who help with projects.

What is an open-weight model?

An open-weight model is an AI model whose inner settings, called weights, are published so anyone can download it, run it and adjust it. That is different from a closed model, which you can only use through the maker's service.

When will the deal close?

Palo Alto Networks expects the deal to close in its fiscal third quarter of 2027. Like most acquisitions, it still has to pass the usual conditions before it is final.

Who will lead the AI work after the deal?

Yash Patil, Applied Compute's co-founder and CEO, will become Palo Alto Networks' Co-Chief AI Officer after the deal closes.

How fast did Applied Compute grow?

The company launched less than 18 months before its November 2026 funding round. It said then that its annualized recurring revenue had passed $100 million.

How did Palo Alto Networks do last quarter?

Revenue for the quarter ended October 31, 2026 grew 39% to $3.44 billion. Net income under the standard accounting rules (GAAP) was $482 million, or $0.57 a share.

Did Palo Alto Networks change its forecast?

Yes. It now expects full-year revenue of $14.52 billion to $14.57 billion, which is growth of 26% to 27%. Its chief financial officer said it is materially raising its guidance.

Does this change anything for regular users?

Not right away. The deal is aimed at large companies that want to run their own AI models safely, so most people will only feel it through the products those companies build.

SourcesPalo Alto Networks
Topics and tagsCybersecurity, Data centers, Funding & deals, palo alto networks

Get the next one in your inbox.

AI, tech and robots in five minutes. Free, every morning.

Go deeper

More in brief