AI Momentum 110.7 +2.9%GitHub Pulse 5178 -6.1%Capital Flow $204.6B +12563.4%Hiring Heatmap 0.0 -100.0%
Advertisement
Markets & Financials

Palo Alto Networks pays $4.5 billion, about 45 times annual revenue, for an AI startup founded last year

Cybersecurity company Palo Alto Networks has agreed to buy Applied Compute for $4.5 billion, paid entirely in Palo Alto stock. The seller is a San Francisco startup founded in 2025 by three former OpenAI researchers.

The price is steep. Applied Compute reached about $100 million in annualized revenue within 18 months of its founding, Palo Alto said on its earnings call. By our arithmetic, that puts the price at about 45 times its yearly sales.

It is also well above what private investors paid weeks ago. Applied Compute was valued at $3.25 billion in a $350 million round led by investor Elad Gil. Palo Alto is paying nearly 40% more than that.

Investors did not object. Palo Alto announced the deal on Wednesday, November 18, alongside quarterly results that beat Wall Street forecasts. Its shares finished Thursday up about 6%, while the S&P 500 fell about 2% and most big AI stocks dropped.

What Palo Alto is buying

Applied Compute takes AI models whose trained settings are published for anyone to use, known as open-weight models, and retrains them on a company's own data. It then runs those custom models for the customer.

The pitch is cost. Chief executive Yash Patil, 23, told Forbes in September that in some cases such tailored models can be 10 times cheaper than top-tier models from the big labs. Customers named by Forbes include DoorDash, the legal AI firm Harvey and Nvidia.

Palo Alto chief executive Nikesh Arora argues that big companies will run a mix of open-weight models and closed models from the big AI labs, and that every one of them will need protecting. He told analysts that, until now, companies have lacked a trusted partner to adapt and run open-weight models safely.

Arora said Palo Alto looked at nine providers before choosing this one. He told analysts no other team matched Applied Compute's "technical pedigree" in retraining open models.

There is a second motive: talent. The startup's research team will join Palo Alto and help train AI models for its own security products. Patil becomes co-chief AI officer once the deal closes. Arora said Palo Alto expects to have 100 AI researchers by year-end.

The cost to shareholders

Palo Alto was candid about the bill. Chief financial officer Dipak Golechha said the deal will lower operating margin by up to one percentage point in fiscal 2027 and up to two points in fiscal 2028. He expects margins to recover to 29% to 30% in fiscal 2029.

The main reason is computing costs. Training and running AI models takes costly chips and data center space, unlike the subscription software Palo Alto mostly sells. The company expects to spend under $3 billion on equipment through fiscal 2030. It will keep buying computing capacity from cloud partners and add its own over time.

Applied Compute brings about $500 million of its own cash and prepaid computing commitments. Palo Alto has not said how many new shares it will issue, so the dilution for existing holders is not yet known.

In return, Palo Alto set a target: adding $1 billion in annual recurring revenue from the business by fiscal 2029, about ten times Applied Compute's current revenue by our arithmetic. That is the company's own goal. None of its published guidance includes Applied Compute, and the deal still needs regulatory approval. Palo Alto expects it to close in its fiscal third quarter of 2027.

Why the market cheered anyway

The deal landed on top of a strong quarter. Revenue rose 39% to $3.44 billion in the three months to October 31, ahead of the roughly $3.31 billion analysts expected. Adjusted earnings of $1.05 a share beat forecasts of $0.97.

Recurring revenue from Palo Alto's newer cloud and AI security products grew 66% to $9.73 billion. The company also raised its full-year revenue forecast to $14.52 billion to $14.57 billion, above Wall Street's estimate of about $14.2 billion.

Some analysts are wary of the acquisition itself. "We see this deal as risky," Morningstar analyst Malik Ahmed Khan wrote, according to Stocktwits. He warned that misjudging future customer demand could lead Palo Alto to overpay for acquisitions or overspend building the same technology itself.

Research firm Futurum flagged a sales problem. Buyers of model-training tools are usually engineering and data teams, not the security chiefs Palo Alto knows best. It also asked whether customers will trust a security vendor that now runs a rival AI platform to police other platforms fairly.

For investors, the deal is a test of a broader question. Security companies have mostly bought tools that guard AI from the outside. Palo Alto is betting it can make money building and running the models too, a business with thinner margins and heavier spending than its core software business.

Written by AI from the public sources linked below.