Palo Alto Networks agreed on November 18 to buy Applied Compute in an all-stock deal valued at $4.5 billion, and the Applied Compute Palo Alto $4.5 billion acquisition is the fastest multibillion-dollar founder flip of the AI cycle. The startup was founded 18 months ago by three former OpenAI researchers. Sixteen days earlier it raised $350 million at a $3.25 billion valuation. Palo Alto is paying about 38 percent more than that, in stock, for a company that has barely existed long enough to hire a second cohort of engineers.

The deal is not closed. It needs regulatory clearance and is expected to close in Palo Alto’s fiscal third quarter of 2027. But the price is set, and the price is the story.

Who built it

The three founders all left OpenAI in 2025. Yash Patil, 23, worked on Codex, the coding assistant. Rhythm Garg worked on o1, the reasoning model. Linden Li worked on the reinforcement learning systems and infrastructure that train models at frontier scale. They started Applied Compute in May 2025, weeks after a ski trip in Lake Tahoe, the same week DeepSeek released its R1 model and rewrote everyone’s assumptions about what open models could do.

What they built is deliberately unglamorous. Applied Compute takes open-weight AI models, trains them on a customer’s own data, and runs them in production. The customers reported so far include Microsoft, DoorDash, Shopify and Harvey, and Forbes reported in September that Nvidia used the company to post-train its Nemotron models. The company says it reached a $100 million-plus annualized revenue run rate in under 18 months. That is the company’s own claim, and it should be read as one, but the customer list is doing a lot of the persuading on its own.

Patil becomes Palo Alto’s co-Chief AI Officer once the deal closes. That title is the tell. This is not a product acquisition with a team attached. It is a team acquisition with a product attached.

Palo Alto bought the team, not the revenue

Palo Alto Networks is a cybersecurity company worth roughly $270 billion. Its chief executive, Nikesh Arora, said his team first evaluated Applied Compute as a prospective customer, then decided to buy it instead. Read that in order. Palo Alto tried the product, liked what the team could build, and concluded that owning the team was worth more than renting the product.

That is the new math of the AI talent war. Capital is not scarce. Compute is not scarce, or at least not scarce enough to stop anyone serious. What is scarce is a small group of researchers who have actually shipped frontier models, and Palo Alto just put a public price on one: $4.5 billion, paid in stock.

The market liked the day. Palo Alto shares rose about 10 percent on November 18, alongside a strong earnings report. Revenue for the fiscal first quarter hit $3.44 billion, up 39 percent from a year earlier, and the company raised its full-year revenue forecast. The deal and the earnings arrived together, so the share move belongs to both, not to the acquisition alone.

What founders should learn from the Applied Compute Palo Alto $4.5 billion acquisition

The lesson cuts both ways, and founders should hear both sides before they celebrate.

First, the opportunity is real. The fastest path to a multibillion-dollar outcome right now is a small, elite technical team solving one hard problem for big customers. Not a big headcount. Not a broad platform. One hard problem, solved well enough that the biggest companies in the world become your customers within a year. Applied Compute did that in 18 months, and the exit came to them. Build the capability first. The valuation follows, and sometimes the buyer comes to you.

Second, the warning is real too. When a security company pays $4.5 billion for an 18-month-old product, it is pricing in a future where those researchers are worth far more than the revenue they have produced so far. If the team leaves, or the models disappoint, or the regulators slow the deal down, the buyer has paid a premium for a story. The same deal that proves small teams can win also proves buyers can overpay for them.

This is the same tension running through the whole AI cycle. Ex-Anthropic founders raised $1 billion for Mirendil before it had customers or revenue, as we covered in our AI bubble reporting. The market keeps paying for the people first and asking about the business second. That works until it does not, and nobody knows which deal is the one where it stops working.

The question until 2027

Between now and the expected close, the question is simple: did Palo Alto buy the next core of its AI business, or the most expensive acqui-hire of the cycle? Either way, every founder with a small elite team just got a new comparable, and every acquirer just got a new price to beat.

The deal also says something quieter about where AI is going. The frontier labs are not the only ones who need frontier researchers anymore. A cybersecurity company just decided it does too, and it was willing to pay $4.5 billion in stock to get them. For founders, that is the real signal. The buyers are no longer only other AI companies. They are every large company that has realised its future depends on models it cannot build itself. For more on the companies shaping this moment, read the rest of our AI coverage.

Free downloads for founders

Get the weekly startup funding tracker, the founder press checklist, and the CEO Medium media kit. Enter your work email and they're yours.

Browse free resources

Follow CEO Medium

Leadership stories & founder interviews, wherever you are.








blank
Author

Daniel Hayes writes on leadership, operations, and the tools modern CEOs actually use.

Write A Comment