CrowdStrike

IThe business

CrowdStrike does one thing. It installs a small piece of software, a sensor, on a company’s laptops, servers and cloud workloads, sends what the sensor sees to its own cloud, and stops attacks. It sells 33 modules on top of that one sensor.⁴ Fiscal years end on 31 January, so FY2026 ended 31 January 2026. TTM on this page means the four quarters to 31 July 2026, because the October quarter reports in December. All share and per-share figures are after the four-for-one split of 1 July 2026, including ones that the older filings give on the old basis.²

Revenue in the twelve months to July was $5.40B, up 24% on a year earlier C (calculated). Of that, 95% was subscription. The other 5% is incident response and consulting, which management treats mainly as a way to find new subscribers.²²⁰⁴ The company sells to more than 88,000 organizations, counting those served through security resellers. It stopped giving a count of direct subscription customers after 23,019 at January 2023. No single end customer matters. 65% of revenue is American.⁴¹⁵²

How it makes money

A customer signs a contract of one to three years, usually paying up front. CrowdStrike books the revenue evenly over the term. ARR, annual recurring revenue, is the yearly value of the contracts live on a given day. It is the number management steers by. It reached $5.84B at 31 July, of which $333M was added in the quarter, a record and 51% more than a year earlier.²³

What it sells now

Endpoint protection built the company. Management now breaks out three newer lines, and they are the reason the business can keep compounding. Cloud security was $905M of ARR, up 29%. Next-Gen SIEM, which stores and searches security logs, was over $695M, up about 60%. Identity was over $585M, up 34%. Together that is $2.18B, 37% of ARR C (calculated).⁵ These are call figures, and the transcript has at least one garbled number, so I treat them as company claims.

The newest item is AIDR, AI detection and response, which watches what AI agents do on a machine. Kurtz says its ARR nearly tripled from the April quarter to the July quarter and that it is a separate paid module. He has not given the dollar figure.⁵ He thinks it could be bigger than the endpoint business, because there will be far more agents than people.⁵ I cannot test that. It may be true. Without the base, nearly tripling means little.

Falcon Flex

Flex is the way most of this is now sold. A customer commits a dollar amount for a period and draws it down across any modules it likes. CFO Burt Podbere calls it a commitment model and not a consumption model.⁶ Accounts on Flex carried $2.29B of ARR at July, up 101%, and that is 39% of the total C (calculated).³ Careful with that figure. It is all the ARR of accounts that have adopted Flex, and most of it was there before. The better numbers are these: customers moving to Flex show about 40% more ARR, and customers who renew and expand a Flex show 25% more again, about eight months in on average.⁵

Flex has a cost. Remaining performance obligations, the contracted revenue not yet booked, rose 49% to $10.7B C (calculated). But the share due within twelve months fell from 52% to 46%. So the near-term part rose about 31%, to about $4.9B C (calculated).²²⁰ Contracts are getting longer. That is good for visibility and it flatters the headline.

The year of the Mythos moment

Nothing on this page makes sense without the last eight months. In February the market feared AI would eat software, and the stock closed at $87.56 on 24 February, its low for the past year.¹ In April Anthropic launched Project Glasswing around its Mythos model, which finds software flaws. Reuters reports that investors moved from fearing AI as a substitute to seeing it as a reason to spend. Kurtz described a deluge of inquiries.³⁰ The company now calls this the Mythos moment, and uses it in its results release.³

Look at what changed and what did not. The price is up 186% from that low, which added about $171B of market value C (calculated). Over the same months management raised its FY2027 net new ARR guide by $116M.¹⁵ The results are real. They added a small amount of new business and a large amount of belief.

A good subscription business that got much better in a year, and whose price got better faster.

IIThe moat

The question is what stops a customer from leaving and a rival from copying. There are three answers, and one open question.

The sensor is the toll booth

Once the sensor is on every machine, each new product is a licence switch. Kurtz says every customer is already enabled for Next-Gen SIEM, since the data is in the platform, and that AIDR needs no new software.⁵ A rival has to get a new agent through a security team and an IT team. CrowdStrike has to send an invoice. Removing it means replacing a tool that sits inside the operating system on every machine, with the risk of being unprotected during the swap.

The evidence is in the customer numbers. 51% of subscription customers use six or more modules, 35% seven or more and 26% eight or more.³ Net retention, which measures what last year’s customers spend this year, was 115% at January 2026. Net and gross retention both rose in the quarter to July.⁴²⁵ Gross retention counts only customers lost and ignores expansion. It was 96% and 98% in the two years before the 2019 prospectus, and the proxy lists 97% in each quarter of FY2026.¹⁹⁷ The 10-Qs give only the word high. I would like the number in every release.

Data

Management’s case is that 15 years of accumulated attack data, and about 7 trillion security events a day, cannot be bought, and that the company creates new data daily, so it trains better security models than a general lab can.⁶⁵ Its chief AI officer says the cost of a remediation fell from about $10 to about $0.03, against a frontier model with a harness.⁶ These are company claims and I cannot test them. What I can say is that data advantages are real when data is scarce and labels are expensive, and that this one has not yet been tested by a lab with a stronger model and a willingness to lose money.

Trust

Security buyers purchase on fear. Incident response, the service line, puts CrowdStrike in the room on the worst day of a customer’s year, and law firms and insurers refer that work.⁵⁴ A small independent buyer survey by Qualitate, published by theCUBE Research, found none of 217 customers polled since July voicing an intention to leave. That publisher discloses clients in this sector, so I give it little weight.³³

The stress test already ran

On 19 July 2024 a bad content update crashed Windows machines around the world.⁴ The stock fell 36% in eleven trading days, from $85.76 on 18 July to $54.47 on 2 August C (calculated).¹ That was as hard a test as a security vendor can get: the protection itself became the outage. Net retention fell from 119% to 112% in FY2025. ARR still grew 23%, and it reached 115% and 24% a year later.¹³⁴

Two cautions. CrowdStrike held customers partly by giving things away. The filings call these customer commitment packages: discounts, extra modules, services, longer terms. They say the packages caused more contraction and smaller upsells.² And the test ran against rivals who sell similar tools. The new test is different.

The open question

Everyone now asks the same thing about software. If a frontier model can find flaws and fix them, how much of a security tool is left? On the Q2 call an analyst asked why CrowdStrike is safe if labs ship their own security products. Kurtz’s answer was data, sovereignty and trust.⁵ TIKR reports OpenAI has Codex Security and that AWS made Continuum generally available in October.³² Kurtz says no model sits in line to stop an attack, which is true of a language model today.⁶ Whether it is true in 2032 I do not know, and neither does he.

It is wide today against rivals. Against the companies that may change what security is, it is untested, and it depends on labs that are also partners.

IIIThe competitive advantage

The advantage, in plain terms, is a single platform built in the cloud from the start, at a size that funds research that smaller rivals cannot match. R&D was $1.56B in the TTM, more than SentinelOne’s whole ARR C (calculated).²⁴²³

That table shows both sides of the argument.^2422223 ServiceNow is a superb software business and trades at about 8 times revenue. The market has decided that some software is at risk from AI and that security is the beneficiary. CrowdStrike and Palo Alto sit far above everything else in the table, and CrowdStrike about twice as high as Palo Alto.

Palo Alto Networks

Palo Alto is the serious rival. It has the firewall and it bought CyberArk for identity. Its Next-Generation Security ARR was $9.10B at July, up 63% with the deal in the number, and it guides 22% to 23% for the next year. It earned a 29.2% non-GAAP operating margin and a 38.4% adjusted free cash flow margin in its fiscal 2026, against 24% and 28% for CrowdStrike over the TTM C (calculated).²²² Its target model is what CrowdStrike is hoping to be.

CrowdStrike’s reply is that a stitched-together portfolio of bought products loses to one sensor and one data layer. The call cites a power company that replaced a legacy acquired SIEM and chose CrowdStrike over a firewall-first product.⁵ I believe the architecture matters. Palo Alto’s numbers say it matters less than a founder’s pitch suggests.

Microsoft

Microsoft is the competitor that can bundle. It does not publish security ARR. CrowdStrike’s 10-K lists Microsoft only as a partner, and from FY2027 it sells through the Microsoft Marketplace.⁴ A rival that gives away a good-enough product inside the licence is the classic way a premium tool loses its middle market. So far the record is the other way: management says marketplaces passed $600M of deal value in the quarter, up 30%.⁵

SentinelOne and the rest

SentinelOne’s ARR was $1.22B, up 22%.²³ It is growing almost as fast from a fifth of the base, without the same profit. Kroll, on the call, moved its customers off a competing next-gen endpoint tool, 450,000 endpoints so far.⁵ Scale is starting to tell at the small end.

Who keeps the money

Here is the part of the advantage I cannot yet see. Gross margin was 75% in the TTM, and 81% on subscriptions before stock pay.²³ Those are good and they are lower than the best software. Sales and marketing took 36% of revenue. R&D took 29%. A business with a deep advantage over ten years usually shows it in less selling per dollar. Sales and marketing was 40% of revenue in FY2023 and 36% now, so that is coming C (calculated).¹⁵² Slowly.

The FY2027 guide has each added dollar of revenue bringing about 38 cents of non-GAAP operating income, against a margin of 22% in FY2026 C (calculated).³⁴ That is what a real advantage looks like. About 8 of those cents are an accounting change, which I come to in Financials.

IVManagement

George Kurtz founded CrowdStrike in 2011, after being chief technology officer of McAfee, and has run it since. He sold an earlier security company, Foundstone, to McAfee in 2004.⁷ Burt Podbere has been CFO since September 2015.¹⁹ Michael Sentonas is President.⁷¹⁰ Dr Bartley Richardson joined from NVIDIA as chief AI and autonomous systems officer in what Kurtz calls a long-planned change of chief technology officer.⁵

How they are really paid

Follow the incentive. The yearly cash bonus pays on non-GAAP operating income, net new ARR and net retention. The yearly performance stock pays on revenue and non-GAAP EPS.⁷ All of those metrics leave stock pay out. So the people who grant the stock are never charged for it in the numbers that pay them. That is common in software. It still means that the largest cost in the business does not enter the scoreboard.

The targets were close together. FY2026 revenue needed $4,743.5M to pay 50% and $4,805.5M to pay 200%, a range of 1.3%, and revenue came in at $4,812.1M. EPS came in far over. Both pay the cap, 200%.⁷ I do not call that rigged. It does mean that a small beat doubles the stock.

The pay ratio was 1,391 to 1 against a median employee at $178,015.⁷ The figure is inflated by the one-off award. Without it Kurtz’s stock award for the year was about $50M at target.⁷

That one-off award is the better part of the plan. Its target is 1.2M shares after the split, up to 2.4M, paying only on three-year total return against the S&P 500 members from 22 December 2025. Below the 25th percentile it pays nothing, and it takes the 90th to pay the maximum.⁹⁷² It has no price target, so a rising market alone does not pay. The start price was $120.79 after the split, and the stock has more than doubled since, which is arithmetic and not a ranking C (calculated).⁹¹ Sentonas got the same kind of award, 400,000 shares at target, in April.¹⁰ Kurtz’s 2021 award was granted at $70.50 after the split, with four price hurdles up to $122.50, and all four were met C (calculated).⁷

How much of their own money is in

A lot, and less than it was. Kurtz held about 7.38M shares directly after his last filing in October, worth $1.85B at the close, about 0.7% of the company C (calculated).¹²¹ The guideline asks him for six times salary.⁷ He sold about 1.9M shares for about $342M in the twelve months to October, mostly under trading plans adopted in January, and Podbere sold about 714,000 for about $158M, most of that in September, and Sentonas about 684,000 for about $127M C (calculated).¹² Directors have sold too.¹²

Trading plans are set in advance and these sales say little about any one day’s view. Founders diversify. But I note the direction. The people closest to this business were steady sellers through a tripling, and none I checked was buying.

What they did with capital

I have not seen a full cycle here, because there has been only one real down year for the shares, and that was 2022. So I judge what I can. The deals have been small and mostly good. Humio, one of the FY2022 deals, was a log-management business, and logs are the base of what is now a $695M line.⁵¹⁶ They total $2.34B since FY2018, 11.8% of revenue from FY2020 to July C (calculated).²¹² SGNL at $628M is the largest and it is too soon to judge.² XM Cyber closed on 8 September, after the quarter, and CrowdStrike issued 2.12M shares for it, worth about $0.44B at that day’s close of $210.02.³⁹ That is 0.2% of the share count and I have not added it to the 1,048M. The buyback was the single best use of cash I can see: $175.6M spent at about $91, then nothing at all in the quarter to July when the price had doubled.²¹¹ That is what an owner does.

Against that, stock pay totalled $4.4B from FY2018 to July, nearly twice the deals C (calculated).²¹² Weighted shares rose 12% from FY2022 to the July quarter on a basic count C (calculated). The company has never paid a dividend. Its $750M of notes at 3.00% due in February 2029 are the only debt, and they were cheap money.²

Governance

At the June meeting Denis O’Leary was re-elected with 61% of votes cast and Johanna Flower with 77% C (calculated). Holders voted about 85% against ratifying the supermajority rules, and 85% for limiting officer liability C (calculated).⁸ A director at 61% is a message.

A founder who built the category, with real money in. The pay works on the numbers that leave out the largest cost, and only the one-off award pays on what owners actually receive.

VFinancials

Ask what CrowdStrike earned and you get four answers. I start there, because the gap between them is the whole page.

The table is calculated from the 10-K, the 10-Qs and the releases.⁴²³ Most of the difference is stock pay. Management tells you to look at the top of the table. The cost at the bottom is real.

Margins

Gross margin was 75.2% in the TTM, from 70.6% in FY2020. Sales and marketing was 36.0% of revenue, R&D 29.0% and general costs 12.6%, a total of 77.7% against 100.9% in FY2020 C (calculated).¹⁸²⁴ Non-GAAP operating margin was 24% in the first half and management’s FY2029 target is 28% to 32%. Subscription gross margin was 81% against a target of 82% to 85%.⁶³ The targets are secondhand, from a transcript of the Fal.Con briefing, as no 8-K carried them.

Stock pay

Stock pay was $1.24B in the TTM, 23.1% of revenue C (calculated). In the July quarter it was $377M, up 35% on a year earlier and faster than revenue C (calculated).²⁴ With employer payroll tax on top, the July quarter was $399M.³ As a share of revenue it was 16.6% in FY2020, 23.5% in FY2023 and 22.8% in FY2026 C (calculated). It has not gone down.²¹

Unrecognized expense on unvested stock units was $2.3B at July, plus $310M on performance units, and that is before the special awards, which can issue up to 3.8M shares.² Grants to the top team are set in dollars and turned into shares at the month’s average price.⁷ So a higher price means fewer shares and not a lower cost, and if the stock halves the same dollars take twice the shares.

Some treat stock pay as non-cash. Every share handed to an employee is a share an owner did not buy, and the buyback to offset $1.24B a year would use most of the free cash flow. I count it as a cost, in full, every year.

What an owner can really take out

This is the number that matters. An owner gets the cash left after stock pay and after the capital needed to keep the business where it is.

So the honest figure is that an owner can take out $0.1B to $0.3B today. That is above GAAP net income of $0.05B and far below non-GAAP net income of $1.14B and free cash flow of $1.52B. The market value is $262B. I strip interest out because I add the cash to the value at the end, and counting both would be double.²⁴

Now the part I cannot judge. The right-hand column is a sketch. Prepayments vanish without growth, which takes off two thirds of free cash flow. But a stopped business would also fire much of the sales force, and no one outside the company knows what share of selling is renewal and what share is growth. I guessed half. The stand-still figure could be slightly negative or could be several hundred million dollars higher. Either way it is very small against the price, and the value lies almost wholly in what the business becomes.

Accounting worth a look

In February 2026 CrowdStrike lengthened the period over which it spreads commissions on new business from four years to five. It says this adds $85M to $95M to FY2027 operating income.⁴ That is about 1.5 points of margin on $6.0B of revenue C (calculated), and it explains why amortization of commissions fell slightly in the first half while capitalized commissions rose.² The change may be fair, since customers stay a long time. But FY2027’s non-GAAP margin rises about 3 points on the guide and about half of that is this C (calculated).³⁴ Commissions paid were $751M in the TTM and $447M went through the income statement C (calculated).⁴²

The company also corrected an immaterial error in the timing of stock pay on certain FY2022 and FY2023 awards, and revised prior periods.² The amounts are small. I use the revised figures.

Prepayments

Deferred revenue was $4.84B at July, up $1.01B in a year C (calculated).²²⁰ This is the best feature of the model: customers fund the business. It is also why free cash flow is above non-GAAP income. It only stays positive while the company grows, and in my model I credit it at 68 cents for each added dollar of revenue, deferred revenue less deferred commissions, both as a share of revenue C (calculated).

Capital spending and commitments

Software is meant to be light. It is getting heavier. Capital spending and capitalized software were 9.1% of revenue in the TTM, from 7.7% in FY2026 C (calculated). Podbere told investors it would go to 11% to 12% next year.⁶ Non-cancelable purchase commitments rose from $2.8B in January to $4.1B in July, and the company added another $2.9B, running to FY2034, after the quarter. The notes list data center capacity first among what these buy.⁴² That is $7.0B of promises, more than a year’s revenue C (calculated), in a business that says AI inference is its next product. The filings do not say what share is AI capacity.

Tax

CrowdStrike paid $51M of income tax in cash in FY2026.⁴ It has $2.6B of federal losses carried forward and a full valuation allowance.⁴² So today’s 28% free cash flow margin has almost no tax in it. Non-GAAP net income uses a 21% rate that it does not pay. At maturity it will pay. My cases charge 21% on GAAP operating profit from the start. The losses would be worth about $0.50 a share at that rate, and I leave them out.

Balance sheet

Cash was $5.01B and debt $750M at July, so net cash was $4.26B C (calculated).² Goodwill rose to $2.25B from $1.36B in six months on the two deals.² Nothing here worries me. A company with this balance sheet can pay for a mistake.

VIValuation

What would ruin it

Before any numbers, I ask how an owner gets badly hurt here. There are four ways, and the last has nothing to do with the company.

1. The product causes another outage. The sensor sits deep in the operating system on millions of machines. It happened on 19 July 2024. As customers consolidate on one vendor and let agents act on their own, the blast from one error grows. Delta’s suit is in discovery, seeking unspecified damages.² A second event would test whether trust can be spent twice.
2. The number the market prices on is questioned. The 10-Q says the company has had requests for information from the Justice Department and the SEC about its recognition of revenue and reporting of ARR for transactions with certain customers, as well as the July 19 incident. It is cooperating and cannot estimate any loss.² I have no evidence of wrongdoing. The securities class action was dismissed and derivative suits closed.⁴² The company counts a contract in ARR while it is negotiating a renewal after expiry, until the customer says it is leaving.⁷ But this page’s value rests on ARR, and at 45 times ARR, doubt about that number would cost more than any fine.
3. AI changes what security is. A frontier lab, Microsoft or a cloud provider makes detection and fixing a cheap feature, or a free one, and the premium goes. The same force that drove this year’s demand could turn. Labs are partners today.⁶
4. The owner pays too much. The business does everything I expect and the investor still loses money, because the price counted it already. On my numbers this is by far the most likely of the four.

The first three sit in the bear case. The fourth is what the rest of this section measures.

What the price implies

At $250.43 CrowdStrike is worth about $262B on the 1,048M diluted shares management guides for Q3, or $258B with net cash taken off C (calculated).¹³² That is 49 times TTM revenue, 45 times ARR and 146 times the $1.8B of free cash flow the FY2027 guide implies C (calculated). The cover showed 1,023.9M basic shares on 20 August.² On Nasdaq’s consensus the price is 1,138 times FY2028 earnings of $0.22 and 491 times FY2029 earnings of $0.51 from four estimates, and against the guided non-GAAP EPS of $1.25 to $1.26 for FY2027 it is about 200 times C (calculated).²⁹³ Reuters had it at 138 times forward earnings in June, with Palo Alto at 69.³⁰

A growth rate from $0.28B of owner cash would mean nothing, because the base is nearly zero. So I solve the question the other way. Hold my base margins, my 10% rate and everything after year ten, start from the $6.6B of ARR guided for January 2027, and ask what ARR growth the price needs for the next ten years.

The answer is 36.3% a year for ten years, to $146B of ARR in FY2037 C (calculated). ARR grew 25% in the last year.² Management’s own goals are $10B within FY2030 and $20B within FY2035, which take about 15% a year from here.⁶ The price needs about $79B by FY2035, four times the goal, and 14% of the $565B market management sizes for 2034 C (calculated). Even with my bull margins, it needs 33.4% E (estimate).

History helps here. From about this size, ServiceNow grew revenue 24.1% a year for five years and Salesforce 20.1% a year for ten, with very large deals. Palo Alto grew 21.9% a year for five, also with deals C (calculated).²⁴ I know of no software company that grew 36% a year for ten years from $6B of recurring revenue.

The price asks for growth above what the largest software companies have ever kept up from this size, for twice as long as they kept it up.

The discount rate

I use 10.0%, as on Snowflake, Palantir and AMD. The capital asset pricing model agrees. With a 4.94% ten-year Treasury, Damodaran’s 4.20% premium and a month-end beta of 1.26 since listing, it gives 10.2%. A five-year beta of 1.19 gives 9.9% C (calculated).²⁶²⁷²⁸ Volatility in the filings’ own option inputs runs at 41% to 54%.² This is not a low-risk stock.

The price is above every cell. The base equals the price at a 4.8% rate E (estimate), under the yield on a ten-year Treasury. Put simply, the market would have to treat these cash flows as safer than the US government. That cannot be what it believes. What it believes is that my cash flows are far too low. So the honest question is how far.

What I cannot judge

Four inputs drive the value, and I cannot know any of them. I show ranges for that reason.

• How long the Mythos demand lasts. Net new ARR up 51% in a quarter could be the start of a decade or a pull-forward of two years of budget. Kurtz says it is durable, and he has a reason to say it.⁵ A 20% change in growth every year moves the base from $42 to $75 E (estimate).
• What share of security frontier labs take for themselves. I have no way to size it.
• Where stock pay settles. I take it from 23% of revenue to 10%. ServiceNow is at 15%, Palo Alto at 15%, Fortinet at 4% C (calculated).²⁴ Three points either way moves the base between $50 and $62 E (estimate).
• What the data center commitments earn. The token business has no revenue figure yet, and Podbere says it will be priced so costs are covered.⁶ Two more points of capital spending every year takes the base to $53 E (estimate).

Three cases

Each case runs fifteen years from FY2028 to FY2042, because a business this young is worth little on ten-year numbers. Ending ARR drives revenue. Then come non-GAAP operating margin, stock pay, 21% tax on what is left, capital spending less depreciation, prepayments at 68 cents per added revenue dollar, and deals at a share of revenue C (calculated). I charge deals because the company buys what it does not build, 11.8% of revenue since FY2020 C (calculated), and uses 4% in the base. After FY2042 the rest of the value grows 3%. Because stock pay is charged as a cost, I hold shares at 1,048M and do not add dilution on top.

The base is a very good outcome, and I want that clear. ARR reaches $10B in FY2029 and $20B in FY2034, each a year ahead of management’s goals. Its FY2029 margin of 29.5% sits inside the 28% to 32% target. Add stock pay and deals back and its free cash flow margin is about 36%, the middle of the 34% to 38% target.⁶ Stock pay falls from 23% to 10% of revenue by FY2042. Owner cash goes from $0.28B to $5.7B in ten years, twenty times over C (calculated). It is worth $56 a share.

The bear has the demand fade by FY2029 as AI security turns out to be a budget shift, a frontier lab or Microsoft takes the new categories, and stock pay stays high. Net new ARR settles near $1B a year. CrowdStrike remains a large and healthy company with $16B of ARR in FY2037. An owner at $250.43 loses 11.5% a year E (estimate).

The bull has AI security become as large as Kurtz says. ARR compounds at 22% for ten years to $48B, with $37B in FY2035, nearly twice the goal, and 41% non-GAAP margin and stock pay at 10.5%. It is worth $119, less than half the price.

Weighted 25%, 50% and 25%, the cases give $62 a share E (estimate). At 20%, 50% and 30% the figure is $67, and at 15%, 45% and 40% it is $75 E (estimate). No sensible weighting reaches the price. Against the price the skew is −93% to my bear and −53% to my bull E (estimate).

A simpler way to see it

A DCF hides in its last line. So here is the plain version: a purchase at $250.43, held ten years and sold at the end of FY2037 at a multiple of that year’s owner cash. I have used 25, 35 and 50 times, and a stock at 50 times owner cash in 2037 would be a very rich stock E (estimate).

In the base, after ten years in which CrowdStrike beats every goal it has set, at 35 times owner cash, the equity is worth less than today’s market value, before the cash built up along the way. Only the bull at 50 times clears 10% a year. I do not need a discount rate to see that.

The same dollar, elsewhere

The next best use of the money is the fair test. A ten-year Treasury pays 4.94%.²⁶ Cash after stock pay yields about 0.1% here C (calculated). Palo Alto, the nearest thing, has a 26 times revenue multiple and earned $4.41B of adjusted free cash flow less about $1.77B of stock pay, a yield near 0.9% C (calculated), eight times CrowdStrike’s, with ARR growing about 22% on its own guide.²²²⁴ On this site, the fairest comparison is Snowflake, another high-stock-pay platform that is well above its value. My base there returns 6.6% a year.³⁶

What closes the gap

Is my base too harsh? I tested that on purpose, and on growth and margin it is not. Its FY2029 revenue is $9.47B against $9.02B consensus, and it reaches management’s ARR goals a year early.³²⁶ Its 37% non-GAAP margin is above the 28% to 32% target and above Palo Alto’s 29%, and taking stock pay from 23% to 10% of revenue is generous.⁶²² The discount rate is where reasonable people differ: the base is $67 at 9% and $83 at 8%. Stack every friendlier change in the table at once and the value is $95 at my 10%, $114 at 9% and $141 at 8% E (estimate). That is still 44% under the price.

Nothing operating closes it, even stacked. Only the bull case at about a 6.8% rate reaches the price E (estimate). The market may have a better reason than I can see. Security has a claim on AI spending that did not exist in 2024, and Gartner’s figures for AI spending, as quoted at the briefing, run to trillions of dollars.⁶ If AI budgets carry 3% to 5% for security for good, my bull is too low. I cannot show that from what is published.

How good the business is and how good the price is are separate questions. The business is among the best in its field, run by its founder, with customers who stay. This looks like a great business priced for something greater.

How the odds shrink

The bull needs many things at once, and these are my own judgments of each.

That comes to about 9% E (estimate). The five are not independent, as strong demand helps most of them, so the true figure is higher. If I raise every line to 85%, it is 44% C (calculated). And the bull, after all of that, is worth less than half the price.

The margin of safety

• Price$250.43Close, Q4 2026 [1]
• Base value$56My estimate E (estimate)
• Margin of safety below$39the base value less 30% C (calculated)

I set it at the base value less 30%, as on Palantir, because almost none of the value exists yet and half of it lies past FY2042. Below $39 my base returns about 13.8% a year and my bear about 2.1%, so the owner is paid even if the bear arrives E (estimate). That is 55% below the lowest close of the past year, and the stock last closed there in August 2023, so it describes a crash and not an entry.¹ At today’s price there is no margin of safety anywhere in my range. The market may be right and I may be wrong, and the reason would be that security’s share of AI spending is far larger and far more lasting than anything yet in the filings.

Where this could be in 5 and 10 years

I take my model’s value at the end of FY2032 and FY2037, which are January 2032 and January 2037, about five and ten years out.

For the base to come true at five years, ARR has to reach $17B from the $6.6B guided for January 2027, and the 10% rate has to be right. At ten years it needs $29B, a 37% non-GAAP margin and stock pay down from 23% to 12.5% of revenue. That makes owner cash $5.7B in FY2037 E (estimate). Even then the base is worth $94 in January 2037, about 37% of today’s price C (calculated).

The bear and the bull differ on growth and margin. The bear has ARR at $16B in FY2037, a 29% margin and stock pay stuck at 15%. The bull has $48B, 41% and 10.5%. The return is lower than the −2.6% under Three cases, because that one counts every cash flow to FY2042 and this one values the years after 2037 at my 10% rate.

The most likely way the base is wrong is growth. It needs demand to last past FY2028, and net new ARR up 51% in a quarter may be a pull-forward. Growth 20% lower every year takes the base to $42 E (estimate).

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

VIIThe case for the price

What most people miss

The $10B goal is easier than it sounds. From the $6.6B guide and the FY2028 floor of $1.63B of net new ARR, reaching $10B by January 2030 needs only about $0.88B a year after that C (calculated).³⁶ The company added $1.0B in FY2026 and guides $1.35B this year. Either management is leaving itself room, or it expects net new ARR to fall. My base assumes the first.

The Microsoft Marketplace opened only this fiscal year, and each of the three hyperscaler marketplaces had a record quarter, with $600M of deal value in Q2.⁴⁵ The vendor most likely to bundle it away now also collects a toll on it.

And the buyback tells you something. Management bought 1.92M shares at about $91 and stopped when the price rose.² People who are careless with stock do not behave like that.

The bull case

Kurtz makes it himself. Every enterprise will run on AI, securing it is the largest market in the company’s history, and AIDR could outgrow EDR because each person may have about 90 agents.³⁵ On a podcast in October he said the threat is only the beginning, TIKR reports.³² Michael Sentonas says $20B of ARR is only about 3.5% of a $565B market in 2034.⁶³⁸ These are company claims.

Reuters quoted Morgan Stanley in June seeing room for more multiple expansion.³⁰ TIKR reports targets after Fal.Con of $265 at Scotiabank, $260 at RBC, $250 at Wedbush and $245 at DA Davidson, and a mean near $234, which is below the close. Its own model has about $393 by January 2031.³¹ ARK’s 13F listed 94,500 shares at 30 June, about 423,000 after the split at 31 March, so it cut about 78% into the rally C (calculated). That is a position, and I found no stated view.³⁴

Against reality

If CrowdStrike adds $2.05B of net new ARR in FY2028 and $2.5B in FY2029, reaches 41% non-GAAP margin, takes stock pay to about 10% of revenue, and the market accepts about 6.8% as the right rate, the price is possible E (estimate). Today’s numbers are far from that. Net new ARR is guided at $1.36B. The non-GAAP margin is 25% against that 41%. Stock pay is 23% and rising in dollars. And 6.8% is 3.2 points under my 10%, for a stock with a beta of 1.26 C (calculated).

Consensus FY2029 revenue, per TIKR, is $9.02B, and my base has $9.47B.³² So I am above the Street and still at $56. We part on what that revenue is worth. On that revenue and a 36% margin, the $234 mean target is about 76 times free cash flow before stock pay, two years out C (calculated).

What a 10x would take. A tenfold price is $2,504, about $2.62T of equity, 26% a year to FY2037 C (calculated). At 35 times owner cash that needs $75B of owner cash. At my base’s 20% owner margin it needs $375B of revenue, more than Microsoft’s $332B in its latest year and two thirds of the entire $565B security market management sizes for 2034 C (calculated).²⁴⁶ At the bull’s 29% it still needs $261B, 5.7 times my bull’s FY2037 revenue. The stock itself has gone up 17 times since its first close, and the company’s value about 22 times, from about $12B C (calculated).¹⁹ NVIDIA went from about $260B in July 2020 to $2.6T in May 2024, the one fast case I checked. It started at about 24 times its last full year’s revenue of $10.9B, and revenue reached $216B by its FY2026, as it sold the scarce good of its era.²⁵ Microsoft went from about $466B in October 2016 to about $3.9T now, about 8 times in ten years C (calculated).²⁴ Tenfold rises from a few hundred billion have come only from platforms that ended with $200B or more of revenue. CrowdStrike starts above 50 times revenue on the same measure. For a security software company it is close to impossible.

Where I come down

The acceleration moved me: my base is above management’s goals and above consensus revenue, which I did not expect to write. It did not move the price question, since the gap is about stock pay and the multiple. I would raise my values if stock pay falls to 20% of revenue by FY2029 while net new ARR stays above $2B; growth 20% above my base every year would make the base $75 E (estimate).

VIIIRisks, and what would change my view

1. Outage. A bad update from a sensor in the kernel of customers’ machines. Delta’s suit is in discovery and the amount is unspecified.²⁴
2. The inquiries. The Justice Department and the SEC asked about revenue recognition and ARR reporting on certain customer transactions.²
3. AI substitution. OpenAI and AWS have products, TIKR reports, and frontier labs are partners today.³²⁶
4. A demand pull-forward. FY2027 net new ARR guidance rose 9.4% since March C (calculated). If that was urgency, FY2028’s guide of 20% or more growth is at risk.⁵⁶
5. Flex at renewal. Flex is a commitment. If customers do not use what they bought, the 10-Q warns of contraction at renewal.²
6. Bundling by Microsoft or Palo Alto’s discounting after the CyberArk deal.²²⁴
7. Stock pay. It rose 35% on a year earlier in the July quarter. Falling shares would raise it further in share terms.²
8. Commitments. $7.0B of non-cancelable purchases, some running to FY2034.²
9. Accounting. Longer commission lives and any further estimate changes flatter margins for a while.⁴
10. Key people. A new chief AI officer, a change of chief technology officer, and heavy selling by the top three.⁵¹²
11. Valuation. The stock fell 36% in eleven trading days in 2024, at a far lower multiple. Today it is at 49 times revenue.¹

What would raise my base value: AIDR proving to be a real category with dollars disclosed, stock pay falling, and margins past the target. What would lower it: a second outage, a change in how ARR is counted, or a lab shipping detection for free. A lower price would not change my base. It would change the answer to the fourth way of getting hurt.