Open in interactive viewer → charts, metric popovers & call review

Palo Alto Buys a Training Company — And the Tape Isn't Buying It Yet

A record Q1, a raised guide, and a $4.5B all-stock detour into open-weight model training — struck from a stock 19% below its October high
PANW · Earnings Call · 2026-11-18

Palo Alto Buys a Training Company

On the surface, Palo Alto Networks' fiscal Q1 2027 was about as clean as enterprise software gets. NGS ARR of $9.73 billion, up 66% year over year. Total revenue of $3.44 billion, up 39%, accelerating from the prior quarter. Non-GAAP operating margin of 30.7%, a record, against a seasonally slow first quarter. Guidance was raised materially across every line. And yet the real content of the call was buried in the last third of Nikesh Arora's prepared remarks: a $4.5 billion all-stock acquisition of Applied Compute, a company barely 18 months old that post-trains open-weight models.

That is not a cybersecurity acquisition. It is a bet that AI security cannot be sold deterministically forever, and that to supervise an AI agent you must first be able to train one. It is, by several orders of magnitude, the strangest thing this company has ever bought.

The Beat, and the Bet

The numbers first, because they matter as the funding source. Revenue of $3.44 billion and 39% growth, with all three reported segments accelerating: network and AI security +20%, Cortex +53%, and Idira +23% pro forma. Net new platformizations hit roughly 170, more than the prior two first quarters combined. The virtual patching service went from a product footnote to a demand driver, with Lee Klarich noting the service holds "thousands of virtual patches" and expects "tens of thousands" by year-end — a direct response to Arora's observation that critical vulnerabilities are being discovered at nearly seven times last year's monthly average.

Then the deal. Dipak Golechha was precise: “The total consideration is $4.5 billion paid entirely in stock, and we expect to close in our 3rd quarter of fiscal '27.” — Dipak Golechha, Chief Financial Officer · 2026-11-18 Roughly $500 million of that arrives as cash and prepaid compute. The target has reached $100 million of annualized revenue in 18 months. Management's plan is to add north of $1 billion to NGS ARR by FY29, and to build 100+ megawatts of inference capacity for cumulative CapEx of under $3 billion through fiscal 2030 — about half a year of expected free cash flow.

The strategic logic Arora laid out is unusually coherent for an acquisition this far from home turf:

I cannot go in and intercept and understand somebody else's application because I don't know how it's supposed to behave. 'Cause I didn't train it. So imagine if you can train your application and have the same team train something that supervises your application. I think that's the future of AI security. So in that context, we have to get in the training business to be able to build a good supervision business.

Nikesh Arora, Chairman and Chief Executive Officer · 2026-11-18

And the ambition is not shy: “If it works, I think this will dwarf the abilities that we have in cybersecurity.” — Nikesh Arora, Chairman and Chief Executive Officer · 2026-11-18

Two Words That Were Everywhere

What is genuinely new in the company's keyword profile is the vocabulary of compute. Applied Compute arrives as the top-ranked company theme. Open weight came from nowhere to near the top of the list, alongside open weight model and the front-runner theme of frontier AI defense. Arora's framing — that open-weight models went from under 5% of token traffic to more than 60% in a year — is a market-level observation that also showed up in the global keyword set last quarter as Open weight model, so this is Palo Alto riding a wave rather than inventing one.

What is company-unique is token flow as a business model. Gabriela Borges asked the right question, and Arora gave the most quotable answer of the call: “Prisma AI is possibly going to be 2 to 5% of your spend over the long term of your token spend.” — Nikesh Arora, Chairman and Chief Executive Officer · 2026-11-18 Every segment, he argued, becomes traffic-sensitive — Cortex ingests it, observability samples it, SASE meters it, Prisma AI inspects it, and identity grows with the agents that need credentials. Notably, he admitted SASE will start metering agentic traffic in six to nine months, after a 9x surge that is still inside licensed capacity. That is a future pricing lever planted, not pulled.

What Went Quiet

The contrast worth flagging is what management stopped talking about. Chronosphere was a marquee theme two quarters running — the observability acquisition, the $200 million frontier-lab customer, the "multibillion-dollar ARR" aspiration. This quarter it appeared mainly as a rebrand: Cortex XCore, ARR up roughly 4x, with a nine-figure net new contribution from the same large frontier lab. The narrative expanded into compute and the onetime acquisition story receded. That is a legitimate tell: the observability story is now a data-ingestion argument inside Cortex, not a standalone growth engine.

More striking is the reversal on M&A discipline. Six quarters of prior calls had Arora insisting that “M and A is not a strategy. M&A is a consequence of stuff that we do from a product development perspective.” — Nikesh Arora, Chairman and Chief Executive Officer · 2026-09-01 In September he framed CyberArk as the proof point that would earn the right to do more: “if I can prove that, the market will give me the license to go in again.” — Nikesh Arora · 2026-06-02 Nine weeks later, he spent the license — on a training company, paid in stock, with a two-year margin drag of up to 200 basis points.

The Tape Disagrees

Here is where the story bites. Palo Alto's shares sit 19% below the October 6 peak of $419.91, down about 4% over the last 90 trading days — into a quarter that beat everywhere. The market is not paying for growth; it is discounting the capital allocation.

The stock trades at roughly 28x trailing revenue, up about 92% year over year — while GAAP operating income has swung to a loss of roughly $110 million, the cost of stock compensation that rose to about $460 million in the quarter plus M&A-related charges. And price to free cash flow sits near 139x. When Arora says this could "dwarf" cybersecurity, he is asking holders paying 28x sales to underwrite an inference build-out on top of it.

The tape's skepticism has company-external rhyme. Across the whole market over the last 30 days, demand for compute is one of the sharpest decliners, with AI-cloud and data-center-capacity themes selling off hard over 90 days. Palo Alto is buying into compute exactly as the compute trade de-rates. Meanwhile the security side of the tape is constructive — tariff-refund-exposed and enterprise-software names have led — so the drag appears deal-specific.

Two things to watch when the transaction closes in Q3 FY27. First, whether R&D spending — already up 55% year over year to about $779 million against $3.4 billion of revenue stays inside the 29.5% operating margin guide. Second, whether the gross margin story holds: Dipak cited 75.2% non-GAAP gross margin, while the GAAP series has been sliding toward 67.6% on hardware component costs and a heavier SaaS mix.

The honest read: this was an exceptional quarter attached to a genuinely bold, genuinely unproven bet. Arora has earned the benefit of the doubt before — but the stock is currently voting that a security company buying a training company is not obviously the same trade.