Palo Alto Networks is now a ~$265B market cap company. The stock is up roughly 100% over the past twelve months and hit an all-time high of $368.80 in July.
What makes it interesting isn’t the run. It’s how they got it. They spent about $29B on acquisitions in twelve months, took a GAAP loss, diluted shareholders by ~14%, cut EPS guidance, and watched the stock fall to $139 in February. Then the core business accelerated and the whole thing re-rated.
First, what they actually do
Palo Alto Networks was founded in 2005 by Nir Zuk and went public in 2012 selling next-generation firewalls: the appliance that sits between a company’s network and the internet and inspects every packet going in and out. That’s still the anchor. Network security is about 70% of total revenue.
Over the last eight years, under Nikesh Arora, it turned from a firewall company into a five-pillar security platform, mostly through 20+ acquisitions:
- Network security (Strata): firewalls in hardware, virtual machine, and cloud-delivered form, plus SASE for remote and hybrid workforces. SASE ARR is $1.6B, growing 40%.
- Security operations (Cortex / XSIAM): the AI-driven replacement for the security operations center. Ingests all of a company’s security telemetry, detects threats, and automates response. $600M+ ARR, growing 100%.
- Cloud security (Cortex Cloud): protecting workloads running in AWS, Azure, and GCP.
- Identity (CyberArk, acquired February 2026 for $25B): privileged access management. Controlling and auditing who, or what, is allowed to touch critical systems.
- Observability (Chronosphere, acquired January 2026 for $3.35B): monitoring whether infrastructure is actually healthy at AI-era data volumes.
Roughly 70,000 customers, about 16,000 employees, and a $11.4B revenue run rate. Hardware is only ~10% of revenue now. The rest is subscription and support.
Why AI is a tailwind and not a threat
Most software categories are getting asked whether AI compresses their value. Security is getting the opposite question, and Arora’s framing on the Q3 call was that AI has raised the terminal value of the entire cybersecurity industry. Four concrete mechanics, all of which show up in the numbers:
1. Agents create far more traffic to inspect. Conversational AI was one prompt and one response. An agent completing a workflow triggers hundreds of secondary machine-to-machine calls to tools and data. That’s a step-change in east-west traffic inside the data center, and it all needs inline inspection. This is why a “declining” hardware line just had its best quarter in ten years.
2. Attacks got faster than humans can respond. Palo Alto’s Unit 42 researchers simulated a full ransomware campaign, from initial entry to data exfiltration, in 25 minutes. The typical enterprise still takes days to identify a breach. Days-to-minutes is not a gap you close by hiring analysts. You close it with an automated platform, which is the pitch for XSIAM.
3. Every agent is a new identity. Enterprise identity used to mean securing a few hundred privileged human administrators. Autonomous agents with credentials multiply that by orders of magnitude, and every one of them can act on production systems at machine speed. That’s the entire strategic logic of paying $25B for CyberArk.
4. AI workloads generate telemetry as a byproduct of running. More compute means more logs, metrics, and traces, which means the observability bill scales with the customer’s GPU footprint rather than their headcount. That’s why Chronosphere nearly doubled ARR in two quarters.
There’s also a fifth thing, which is an entirely new category: securing the AI applications and agents themselves. Prisma AIRS is the fastest-scaling product in company history and didn’t exist eighteen months ago.
Net effect: AI increases the number of things to protect, the speed at which they must be protected, and the volume of data produced in protecting them. That is a rare position to be in right now.
The headline numbers
From Q3 FY26 (quarter ended April 30, reported June 2):
- Revenue: $3.0B, up 31%
- FY26 revenue guide: $11.42B, up 24%
- Next-Gen Security ARR: $8.13B, up 60% (up 28% organic)
- RPO: $18.4B, up 36% (up 22% organic)
- Adjusted free cash flow margin: 38.5% on a trailing 12-month basis
- Platformized customers: ~2,280, at 120% net revenue retention
- 70,000+ total customers, ~16,000 employees
5 Interesting Learnings:
1. The stock doubled. But it fell 25% first, and management bought the bottom.
The sequence matters more than the outcome.
July 2025: Palo Alto announces it’s buying CyberArk for $25B, the largest deal in the history of the security industry. The market hates it. The stock drops 12.5% in a month.
November 2025: they add Chronosphere for $3.35B, at roughly 21x ARR.
February 2026: the CyberArk deal closes. On the same earnings call, they cut FY26 adjusted EPS guidance from $3.80-$3.90 down to $3.65-$3.70 to absorb acquisition costs. The stock falls another 7%. It bottoms at $139.57 on February 24.
June 2026: Q3 beats every guided metric. Organic bookings accelerate. NGS ARR comes in at $8.13B against a raised bar. By mid-July the stock is at $368.80.
The detail that should stick with every founder and board member: in Q3, Palo Alto spent $1B buying back 6.8 million of its own shares at an average price of $147.69. Those shares are worth roughly $347 today. Management bought the bottom of their own dilution panic, five months before the market agreed with them.
The market prices large M&A on announcement-day dilution. It reprices on execution. The gap between the two can be 100%+, and it can take three quarters to close.
2. They paid 21x ARR for Chronosphere. Two quarters later the multiple had roughly halved.
Chronosphere was doing north of $160M in ARR when the deal was announced in November 2025. Palo Alto paid $3.35B. That’s about 21x, and it looked expensive.
In Q3, observability ARR surpassed $300M, up more than 50% sequentially from Q2 and nearly double what it was at announcement.
Why: an existing LLM customer ramped consumption as it migrated off the incumbent vendor. Two of the top five frontier labs are now on Chronosphere. One frontier AI lab alone is over $200M in ARR with Palo Alto for observability across its training and inference clusters, and that number is still growing as the migration completes.
This is the part most B2B operators underweight. Consumption-based revenue inside an AI-native customer base does not behave like seats. AI workloads generate telemetry as a byproduct of running, so the revenue scales with the customer’s compute, not their headcount. A price that looks like 21x against last quarter’s ARR can look like 10x two quarters later if the underlying consumption is compounding.
The corollary: the multiple you pay is only expensive relative to the growth you can put through the asset. CyberArk is tracking the same way. It beat internal targets in its first quarter post-close, and Palo Alto now says it’s 3 to 6 months ahead of schedule on converging CyberArk’s profitability with its own.
3. 2,280 customers do 120% NRR. The other 68,000 don’t.
Palo Alto has 70,000+ customers. Only about 2,280 are what they call “platformized,” meaning they’ve consolidated multiple security functions onto one architecture instead of buying point products.
That cohort does 120% net revenue retention with single-digit churn. They added 110 net new platformizations in Q3, including 20 from the CyberArk and Chronosphere integrations.
The target is 4,000 platformized customers by FY2030, and that alone is supposed to carry NGS ARR from $8.1B to $20B.
Supporting evidence that the mechanic is real:
- The installed base averages more than 4 subscriptions per firewall device, against 11 advanced subscriptions now on offer
- 80% of net new observability customers this year adopted multiple products
- Roughly 1,000 cross-organization sales engagements have been initiated between the core and CyberArk identity teams since close
Most B2B companies have a multi-product cohort and a single-product cohort with radically different economics, and never separate them in reporting. Palo Alto built its entire growth model, its M&A strategy, and its FY2030 target around the delta between the two. Worth asking what your own number looks like if you split it.
4. “Legacy” hardware just had its best quarter in a decade. Because of AI.
Next-gen firewall bookings grew nearly 40% year over year in Q3. It was the strongest hardware quarter in ten years, with record Q3 backlog.
Hardware is only about 10% of total revenue now, down from 20% in FY21. Everyone, including most of the sell side, had written it off as a melting ice cube.
The driver is agentic AI, and the logic is worth understanding because it generalizes. The first wave of enterprise AI was conversational: one prompt, one response, minimal traffic. Agents are different. An autonomous agent completing a workflow triggers hundreds of secondary machine-to-machine calls to tools and data. That’s a large increase in east-west traffic inside the data center, and east-west traffic needs inline inspection.
New buyers showed up too: sovereign infrastructure providers and AI labs, which is a category that didn’t buy firewalls three years ago. One of the largest deals in the quarter was an $80M transaction with a U.S. power producer sitting at the center of the AI infrastructure buildout.
Two related numbers: they took a 10% price increase on hardware in April, and 46% of trailing-12-month product revenue is now recurring software, up from 22% three years ago. The “hardware” line stopped being a hardware line a while ago.
The lesson: before you write off a declining line, ask whether the new wave increases or decreases the underlying unit of consumption. AI increased Palo Alto’s.
5. New products inside an incumbent are scaling faster than they used to. A lot faster.
Prisma AIRS, their AI security platform, went from 100 customers at the end of Q2 to over 300 in Q3. Management says they have clear visibility to $100M in ARR within the next couple of quarters for a product that was not in market a year ago. Their record deal for it was $20M+ with a global consulting firm running more than 2 trillion tokens per month on the platform.
XSIAM, launched 42 months ago, is now at $600M+ ARR, up 100% year over year, across 740 customers.
And Unit 42 Frontier AI Defense, launched in Q3, generated inbound requests from more than 1,200 customers, of which 800 meetings were completed in six weeks.
Zero to $100M in roughly five quarters is startup-caliber velocity happening inside a $11B company. The reason isn’t that Palo Alto builds faster than startups. It’s that in a category being created this quickly, an existing platform’s distribution advantage is larger than it used to be. The buyer already has a relationship, a contract vehicle, and a budget line. By the time a startup finishes its first go-to-market motion, the incumbent has already sold the same buyer.
That’s a genuinely harder environment for new security startups than it was in 2021, and founders in adjacent AI-infrastructure categories should assume the same dynamic.
And 5 bonus learnings
Bonus 1: They used frontier model early access to do a year of pen testing in under three weeks. Then they sold it. Palo Alto has partnerships with the leading frontier labs that give it early access to the most advanced models. They pointed those models at their own environments and compressed roughly twelve months of penetration testing into less than three weeks. That work became a product, Unit 42 Frontier AI Defense, which generated inbound meeting requests from more than 1,200 customers and 800 completed meetings in six weeks. Internal AI leverage turned into a category launch and a pipeline event in the same quarter. That’s the pattern worth copying.
Bonus 2: XSIAM’s headline metric is time, not seats. The majority of XSIAM customers now respond to threats in under 10 minutes, down from days or weeks. Nobody buys a security operations platform because of its ARR. They buy it because the number went from days to minutes. Palo Alto leads with the outcome metric and lets the $600M ARR follow. Most B2B companies report the reverse and wonder why the pitch lands flat.
Bonus 3: Secure Browser quietly hit 11 million licenses, a 4x increase. The browser is turning into a control point, because it’s where employees actually touch AI tools. Every SaaS app, every model, every agent interface runs through it. A distribution surface that goes 4x in a year inside an enterprise install base is the kind of thing that looks like a footnote for two years and then becomes a platform.
Bonus 4: Post-merger integration is a line-item grind, and they published the receipts. Over 40 new facilities came in with the acquisitions. They identified more than 300 IT vendors to consolidate and have already cut roughly 20% of them. They renegotiated cloud hosting economics for the acquired CyberArk business using combined scale. That unglamorous work is why the synergy targets landed 3 to 6 months early, and why they now expect CyberArk’s profitability to converge with the parent’s within 12 to 18 months. Integration isn’t a strategy deck. It’s a vendor list.
Bonus 5: Someone paid for the $25B, and it was existing shareholders. Q3 GAAP net loss was $177M against $684M in non-GAAP net income. Stock-based comp ran at 17% of revenue. Diluted share count went from 707M a year ago to 801M, roughly 13% dilution in twelve months. The company says SBC returns to pre-acquisition levels over time, and the $1B buyback at $147.69 offsets part of it. But $0.64 per share of SBC in a single quarter is not a rounding error, and any founder using stock as acquisition currency should model the dilution against the re-rating, not instead of it.
What this says about buying growth in 2026
The consensus view for the last few years has been that large B2B acquisitions destroy value, that acquired growth is lower quality than organic growth, and that acquirers should be discounted for it.
Palo Alto’s last twelve months are a counterargument, with conditions. The acquired assets are growing faster than the core, not slower. Chronosphere nearly doubled post-announcement. CyberArk beat internal targets in its first quarter and its synergy timeline pulled forward 3 to 6 months. Organic NGS ARR still grew 28% and organic RPO 22%, so the core didn’t stall while management was distracted.
The condition is that they bought into an accelerating demand curve rather than a decelerating one. Identity, observability, and AI runtime security are all expanding because AI is expanding. Buying an asset in a flat category to paper over your own deceleration is a completely different transaction, and the market is right to price that one lower.
Underneath all of it is Arora’s actual thesis. He was blunt on the call about the limits of frontier models in production security: error rates often reach 25%, forcing the manual intervention that destroys the speed advantage, and the models consistently fail at the last mile of complexity. In a mission-critical environment one wrong enforcement decision takes down a global production network.
So the model isn’t the moat. As frontier models become available to everyone, the advantage shifts to the data layer. Palo Alto’s version of that layer is 125 million sensors ingesting 17 petabytes of daily telemetry, and every additional sensor makes the platform smarter, which drives more deployments, which produces more data. Every acquisition on this list added sensors. That’s what they were actually buying.
If your AI product’s advantage is the model, you don’t have one. If it’s proprietary data generated by sitting inline with something that matters, you might.
Fiscal Q4 and full-year results land September 1, with a $3.345B to $3.355B revenue guide and NGS ARR guided to $8.9B to $8.95B.
