Key Takeaways
- CrowdStrike CEO George Kurtz said a general remediation task costs about $10 on the best frontier cyber model and about three cents on CrowdStrike’s own model.
- That cost gap lets CrowdStrike undercut the AI labs on price and still keep the roughly 75% gross margin it has held for the last three fiscal years.
- The main risk is that cheaper models push security prices down across the industry, including at Palo Alto Networks and Zscaler, as frontier labs move into cyber themselves.
- Analysts expect revenue to nearly double by the year ending January 2029, and the stock trades at 169x forward earnings, well above its three-year average of 103x.
Yesterday, details of two critical PHP bugs that let anyone run code on a server without logging in leaked before a patch was even out. Agencies from the US to Japan told people to take their systems offline.
Today, CrowdStrike (CRWD) CEO George Kurtz went on the No Priors podcast and explained what it costs to clean up that kind of mess with AI. He said that “for general remediation, if you use like, the highest caliber frontier cyber model probably cost you about $10. Ours is like three cents.”
That’s a huge edge, and it matters most for CrowdStrike’s margins.
TIKR tracks 54 analysts on CrowdStrike. How many say “Buy”?
See the full breakdown of Buy, Hold, and Sell ratings on CrowdStrike, along with Wall Street’s revenue and earnings estimates, all in one place. More than 500,000 investors use TIKR to research stocks like this one. See CrowdStrike’s Analyst Ratings
Three cents leaves plenty of margin
CrowdStrike keeps about three-quarters of every revenue dollar as gross profit. Its gross margin has stayed between 73% and 75% for five fiscal years.
Its subscription business did even better in the latest quarter. Subscription gross margin was 78% in the three months to July 31, 2026, up from 77% a year earlier.
Running AI agents on a frontier model at about $10 a task would cut into that margin fast. At three cents, the compute costs next to nothing against almost any price CrowdStrike charges. So it can undercut the labs and still keep most of each sale.
Kurtz says the cost edge comes from how CrowdStrike builds its models. It starts with NVIDIA’s open Nemotron models, post-trains them on more than a decade of its own security data, and runs them inside its own harness, which he says gets better results at a much lower cost.
Building those models isn’t free, of course. Research and development rose to 29% of revenue in the year ended January 2026, from 26% two years earlier, and that spending doesn’t show up in gross margin.
90 agents per employee
Kurtz says CrowdStrike can find the rogue AI agents running inside a company and trace everything each one does from the moment it starts: its sub-agents, its tool calls, the servers it reaches. Customers can then set rules to block what an agent shouldn’t be doing.
Then he made the business case: “they say you can have 90 different agents as an employee, well, that can be a much bigger opportunity than just EDR for us.” (He thinks that 90 figure is low.)
If every agent needs watching the way a laptop does, CrowdStrike’s market grows with the number of agents, which is far faster than headcount. Cheap models are what make watching all of them affordable for customers and profitable for CrowdStrike.
Analysts already expect revenue to go from $4.81 billion in the year ended January 2026 to $9.02 billion in the year ending January 2029.
That’s more than 20% growth a year. One interview won’t move those estimates, but agent security showing up in CrowdStrike’s results would.
Cheap cuts both ways
Of course, cheap models are coming for everyone. OpenAI and Anthropic already give vetted defenders access to their models through programs like Daybreak and Mythos Access. If the labs’ models get cheap enough, customers will expect to pay less for security across the board. Palo Alto Networks (PANW) and Zscaler (ZS) carry that same risk. This week Zscaler joined CrowdStrike, which expanded its AI agent security deal with NVIDIA, in supporting NVIDIA’s new architecture for securing agents.
Kurtz’s answer is the data: “there is no Reddit of CrowdStrike security data.” He says synthetic data can’t replace the attack chains CrowdStrike and its managed detection and response service have collected for more than a decade. A lab can cut its price, but it would have to build that history from scratch.
Plus: Kurtz wants the labs’ models on CrowdStrike’s platform for customers who prefer them. So CrowdStrike can sell either one.
The market is already paying up
Investors have noticed. CrowdStrike traded at 169x forward earnings on Oct. 21. That’s well above its three-year average of 103x and not far below its 198x peak on Oct. 6.
My take: at three cents a task, CrowdStrike can compete on price for agent security and keep its margins. That’s exactly what a stock at 169x needs. But at that multiple, I’d want to see agent revenue in the numbers before paying up.
Of course, three cents is Kurtz’s own figure. The next checks are CrowdStrike’s fiscal third quarter, which ends Oct. 31, and whether Palo Alto Networks and Zscaler start pricing their own AI the same way.
So what is CrowdStrike stock actually worth?
TIKR lets you forecast the future price of any stock in less than a minute. Just enter a few assumptions into TIKR’s valuation model and see what CrowdStrike could be worth. Start from Wall Street consensus estimates, or adjust the inputs to reflect your own view of the business. It’s free to use.
Disclaimer:
Please note that the articles on TIKR are not intended to serve as investment or financial advice from TIKR or our content team, nor are they recommendations to buy or sell any stocks. We create our content based on TIKR Terminal’s investment data and analysts’ estimates. Our analysis might not include recent company news or important updates. TIKR has no position in any of the stocks mentioned. Thank you for reading, and happy investing!
