Cloudflare

IThe business

Cloudflare is a network. A customer points its website, its staff’s logins or its code at Cloudflare, and traffic then passes through Cloudflare’s servers before it reaches anything else. On the way it gets checked for attacks, sped up, routed and, more and more, run. Cloudflare owns its servers and rents space for them in other people’s buildings, mostly internet providers and colocation sites, in more than 335 cities and over 125 countries.⁵ It does not build data centers. The Delaware Court of Chancery, which has no reason to flatter it, put its reach this way: more than twenty percent of the internet uses its products.⁸

The fiscal year ends on 31 December. TTM on this page means the four quarters to 30 June 2026, because third-quarter results come on 5 November, after the date of this page.¹⁶ In those twelve months revenue was $2,512M, up 34% C (calculated). Gross margin was 72.6%. GAAP operating loss was $354M, which includes a $150.7M restructuring charge.²³ Half of revenue comes from the United States, 27% from Europe, the Middle East and Africa, 14% from Asia Pacific.⁵

Revenue grew from $84.8M in 2016 to $2,167.9M in 2025, 43.4% a year, with no large acquisition in it C (calculated).¹⁷⁶ That is the first thing to admire. Growth slowed to 27.8% in the second quarter of 2025 and reached 35.9% a year later C (calculated).²⁹³¹⁸ The guide for the third quarter is $736M to $737M, up 31%. The last guide proved low: for the second quarter it guided to $664M to $665M and reported $696.1M.¹⁹²

Who pays

There are three kinds of customer. Free customers pay nothing. They are the individuals, students and start-ups who put a site on Cloudflare because it costs nothing, and they are not charity. They bring traffic, data and developers. Pay-as-you-go customers pay by card on the website. Contracted customers sign for one to three years, and bigger ones increasingly buy a “pool of funds” they draw on across products.⁵

At the end of 2025 there were 332,466 paying customers. Cloudflare stopped reporting that number in 2026, saying it had become much less important.⁶⁵ The number it keeps is customers paying over $100,000 a year: 4,698 in June, up 27%, and 73% of revenue, up from 71%.⁵¹⁰ Dollar-based net retention, which compares what last year’s customers pay now with what they paid then, was 120%, up from 114% a year earlier and 111% at the low at the end of 2024.⁵⁶ No customer reached 10% of revenue in any of the last three years.⁶ Most contracts cannot be cancelled in their term.⁵ Remaining performance obligations, the contracted revenue not yet earned, were $2,732.0M in June, 64% of it due within twelve months.⁵

What it sells

Management tells the story in four acts. The labels are theirs, and the company reports one segment, so no filing gives revenue by act. That is a real limit on what anyone outside can know. Reports of the June Investor Day say annual recurring revenue grew 43% in 2025 for Cloudflare One and 137% for the developer platform, but I could not open the company’s own slides, and those growth rates come without the dollar base.²³²⁴

The argument for the stock sits in acts three and four, so it is worth saying what is new. Prince says the internet’s traffic is turning from people to machines. Non-human traffic passed human traffic on Cloudflare in May 2026, and on the second-quarter call management said that on current trends it could reach as much as 1,000 times human traffic in five years, with the caveat that it has called this trend wrong before.¹⁰ It had said in May that the crossing would come in 2027.¹⁹ So it has already been wrong once, in the good direction. Agent requests rose 1,700% from June 2025 to May 2026, by the Investor Day figures as reported.²⁴

In May, Prince gave the volume at “hundreds of billions of agentic requests per month”.¹⁹ That sounds vast, and for a network it is. But most of its revenue is subscription, not a fee per request, and free sites pay nothing. Whether machine traffic pays Cloudflare more than it costs Cloudflare is the central question of this page. I come back to it in Valuation, because I can’t answer it, and nor, so far, can anyone else.

The price chart belongs here too, since a reader needs to know what kind of stock this is. It listed at $15 in September 2019 and closed its first day at $18. It rose to $217.25 in November 2021, then fell 83% to $37.84 a year later, while revenue grew 49%. The record close is $360.93, on 9 October 2026, and three weeks later it stands 22% lower C (calculated).¹¹⁷

IIThe moat

A durable business keeps others from taking its profits. Cloudflare has a strange kind of test here, because it has no profits to take yet. So I look at what that protection is meant to produce: low cost, high retention and the ability to raise what customers spend. On those it scores well. On return to owners it has not been tested.

The core idea is an engineering decision made early. Every server runs every service, on the same software, and traffic can move to any of them. The 10-K calls this serverless, and it means the machines are cheap commodity boxes and spare capacity is always there.⁶ Older vendors built separate boxes or separate networks for each product. A new Cloudflare product costs little to add, because it runs on hardware already in place.

The free tier then does something a competitor can’t easily copy. Cloudflare’s idle capacity serves free customers. Free customers bring traffic. That traffic makes Cloudflare attractive to internet providers, who let it put servers inside their networks and exchange traffic at low or no cost.⁶⁵ Lower bandwidth cost then pays for more free capacity. The free users also show Cloudflare new attacks first. Since the company’s founding the cost of that traffic has sat in sales and marketing, which is honest: it is how the company finds customers.¹⁹

Is it real?

The numbers that this kind of protection ought to move have moved. Net retention rose nine points in eighteen months. Sales productivity has risen for nine quarters, and deals above $1M rose 73% in the first quarter.⁵¹⁹ Prince has said “we’re the fittest we’ve ever been”, and he may be right.¹⁹

The test it has not passed is in the fall in gross margin, from 77.3% in 2024 to 72.6% C (calculated).⁴² Some of that is geography. As free developers move to paid products, their traffic moves from sales and marketing to cost of revenue, and the CFO calls that “literally a wash”.¹⁹ But part is real. Developer products carry a lower gross margin, and he said so.¹⁹ Reports of the Investor Day put the long-term range at 70% to 77%, with the floor at 70% or better.²³²⁴ So management is telling owners that the best-margin years are behind and that operating margin is now the measure to use. That may well be right. It also means an advantage that was visible in one line now has to be taken on trust across several.

The other weakness is that the network’s advantage is largest where the traffic is small and scattered. A hyperscaler has its own global network and its own edge. It can bundle security and delivery with the compute a customer already buys. Cloudflare says it wins such contests, and it names two deals won from hyperscalers in the first quarter.¹⁹ Those are the wins it chooses to tell. Nobody publishes the losses. I read the protection as wide against Akamai and Fastly and narrower than it looks against Amazon, Microsoft and Google, whose customers already sit one click away.

IIIThe competitive advantage

Cloudflare’s rivals come in three kinds: the old hardware vendors, the point products such as Akamai, Fastly and Zscaler, and the public clouds.⁶ Against the first it is taking business, as all cloud security does. The second can be compared with numbers.

Read across the first row and the second. Akamai, the original content network, grew 5.4% and makes money after stock pay. Its delivery business fell 6% in the quarter to $396M, and security, at $604M, grew 10%.¹¹ Cloudflare grew nearly seven times as fast on 58% of Akamai’s revenue C (calculated). That is the competitive advantage in one line. The market pays 7.5 times as much for it C (calculated).¹⁵

Akamai isn’t standing still. Its cloud infrastructure line is only $99M a quarter, but it grew 39%, and the company says it has signed more than $2.8B of multi-year contracts for it this year.¹¹ The lesson I draw is that demand for distributed AI compute is real and that Cloudflare isn’t the only one getting it. Fastly grew 23% with net retention of 117%, which suggests the rise of machine traffic is lifting more than one boat.¹² Prince’s reply is that traffic is not all equal, and that a content network chasing video bytes is in a different business from one sitting in front of APIs.¹⁹ He has a point, and I think he has most of the point.

Against the clouds

Workers is priced to win compute from Amazon, Microsoft and Google, and the 10-K says so.⁶ The 10-K calls it a lightweight execution environment, which suits short tasks and, management says, agents. Cloudflare also runs GPUs across the network and bought Replicate in December 2025 for $57.4M to host models.⁵ It bought VoidZero, which builds developer toolchain technology, for $164.2M in June, plus $106.4M of stock pay for the people, to own more of the path a developer takes.⁵ These are small, sensible purchases.

What I can’t see is the margin. Developer revenue carries a lower gross margin, and no filing says how much lower. Management says that unit economics across products are close once you count the lower cost to book the sale.¹⁹ There is logic to that: a developer who signs up alone costs no salesman. But a claim about unit economics that outsiders can’t check is one I hold to a lower standard of proof than a reported number, and it will carry much of the value in acts three and four.

What it has that nobody else has

One thing is close to unique. Cloudflare sits between websites and AI companies, and it has a default setting. On 1 July it said it would split so-called mixed-use crawlers, which fetch for search and AI at once, and block them by default on ad pages from 15 September for new customers and all free customers.²⁵ That is real power over a part of the web. It also pits Cloudflare against Google, whose answer is that Google-Extended already lets sites opt out of training, though Googlebot still feeds AI Overviews.²⁵ Google is the one crawler no site can afford to block. So the place where Cloudflare’s position is strongest is also where it faces the strongest counterparty. The first partners for Pay Per Use are Ceramic.ai and You.com, not the largest buyers.²⁵

IVManagement

Matthew Prince co-founded Cloudflare and has been chief executive since July 2009. He trained as a lawyer at the University of Chicago and has spent his career on internet security. Michelle Zatlyn is the other active co-founder, and president; she has been co-chair with him since February 2025.¹⁴²⁶ A third founder, Lee Holloway, stepped away in 2016 because of illness.⁸ The first decision I’d credit to them is the free tier, and the second is to keep building products on one stack. Few founders of companies this size have stayed this technically close.

How they are paid, and what that pays them to do

Follow the incentive first. There is no annual cash bonus at Cloudflare for its top officers.¹⁴ The founders drew no equity awards in 2024. In February 2025 each got a salary raise to $550,000, the first since the listing, and stock.¹⁴ Prince’s reported pay for 2025 was $60.6M, 387 times the median employee’s $156,306, and $58.3M of it was stock.¹⁴ Each founder got 259,672 restricted units that vest over five years for staying, and 175,111 performance units that vest only on the share price.¹⁴

So what does that pay them to do? Raise the share price. Nothing in the plan rewards free cash flow per share, return on capital or stock pay coming down. The hurdles reward growth, and a high multiple on that growth, and a founder who keeps the share price high also keeps a valuable currency for paying staff. That lines up well with owners who bought low. It lines up less well with the question on this page, which is what an owner gets out in the end. I also note the size. Next to a stake worth about $7B, $60M a year is small. Prince’s main incentive is the stake and the control that goes with it.

How much of their own money is in

A great deal. At the June record date Prince held 25.7M Class B shares and 45,508 Class A, about $7.2B at the close, and Zatlyn about 9.1M, about $2.6B C (calculated).¹⁴ Few executives at any company have as much in. Both are also steady sellers, and have been for years. The court counted about $1.18B of sales by Prince and $629M by Zatlyn from 2020 through the spring of 2026.⁸ From March to early August 2026 Forms 4 show Prince selling 942,945 shares for $219.2M, and through 7 October Zatlyn selling 713,222 for $184.2M C (calculated).²⁷ I found no open-market purchase by any insider in those filings. Founders diversifying after seventeen years is normal. But the selling is what led to the next section.

The fight over control

This is the most important thing to know about governance at Cloudflare today. The founders’ share of votes peaked at about 60% in 2021 and fell to 50.3% by June 2026, as they sold and as stock pay issued new shares.⁸ At their historical rate of selling, a projection in the court record had them losing the majority by December 2026.⁸ The board formed a special committee in October 2025.¹⁴ The plan that came out of it would swap the founders’ ten-vote shares for one-vote shares plus a new preferred with nine votes and almost no economic value, and hand every holder one non-voting Class C share for each share held.¹⁴²⁹ New stock pay and deals would use the Class C. The founders could then sell far more without losing control. On the plaintiffs’ reading, the extra room was worth about $2.1B of further sale proceeds.⁸

Holders other than the founders voted against it. The proposals passed only because the founders held 50.3%. Between 15.6% and 17.7% of the voting power outside the founders voted in favour, and ISS and Glass Lewis both opposed it.⁸³⁰ Pension funds including CalPERS sued.⁸ On 16 October 2026 Vice Chancellor Laster stopped the closing until after trial.⁸ He ruled for the defendants on every contested reading of the new statute, and he did not hold that the plan was unlawful.²⁹ He found a reasonable probability that the committee had not acted in good faith, and said that without the credibility problems he might not have enjoined it.⁸

The record is what an owner needs to read. Directors and founders used Signal with auto-delete settings on after being told to expect litigation. The committee got no fairness opinion and never valued what the founders received. A member used an AI chatbot to find a supportive academic. Prince wrote to a director, “I’m tendering my resignation in 12 hours if this BS isn’t resolved.”⁸ The defendants dispute how seriously anyone took that, and the court left it for trial.⁸ The defendants had ten days to choose a short first trial on good faith, ideally within about ninety days. I found no SEC filing on their choice by the date of this page.⁸

What do I take from it? The dual-class share may well have helped this company: it let Prince spend for the long run, and the numbers in Section I are the result. But the same man who built that record threatened to quit over a committee’s proposal, and the people meant to stand up to him texted through disappearing chats. The founders are worth more to this business than to almost any other I cover, which is why the threat works. An owner of the Class A is a passenger. I put no number on this, but it is one reason I won’t pay up for growth beyond what the cash can show.

What they did with capital over a full cycle

The fairest test covers good years and bad, and this company has had both in its stock, though not yet in its revenue. From 2016 to 2025 it generated $146M of free cash flow in total, while paying $1,486M in stock C (calculated).¹⁷²⁰⁶ It has never paid a dividend or bought back stock in size, and its own issuance has raised the share count by about 17.8%, by the court’s count.⁶⁸ What it did was reinvest and finance itself very well.

On the May cut, Prince says that usage of AI tools inside Cloudflare rose more than 600% in three months, that all production code contributions are now reviewed by AI agents, and that he expects more staff in 2027 than at any point in 2026.¹⁹ So this is less a shrinking than a swap of support roles for engineers and sellers who use the tools. It happened at a moment of record bookings, which took some nerve. Anyone who says they know how this works out is guessing. The second-quarter restructuring of $150.7M is real money, and it is why GAAP operating margin read −29.6% that quarter.²

VFinancials

The record is short and simple. Revenue has grown every year, gross margin was high and is now easing, and GAAP operating losses have never closed. Cloudflare has been public for seven years and has not reported a full-year GAAP operating profit, or a GAAP net profit. Management says GAAP profitability will come by 2028 at the latest, and that it is ahead of that pace.²³¹⁰

What an owner could really take out

Cloudflare says free cash flow was $315M in the last twelve months, 12.5% of revenue. Non-GAAP net income was about $411M. GAAP says a net loss of $206M. Which is the owner’s number? None of them, as they stand.

Start with free cash flow and take out stock pay, which is a real cost: it is how Cloudflare pays its people, and it dilutes every owner. The cash flow statement adds back $508M, 20.2% of revenue. Free cash flow less stock pay is −$193M C (calculated). Take out the one-off cost of the May cut, and that is −$66M C (calculated). Then look at where the cash came from. $133M of the free cash flow was cash interest on the cash pile, 42% of it, and much of that pile was borrowed at 0% through convertible notes C (calculated). That is good finance, and it isn’t the business. Without the interest the business lost $200M on this measure C (calculated).

What capital does it need just to stand still? A fast grower spends much of its capital on growth, so it is fair to ask. Depreciation and amortization were $225M, against capital spending of $319M, so about $93M went on growth, on that rough reading. Using depreciation as the cost to stand still, and ignoring the one-off, the owner’s number comes to about −$106M E (estimate). I would give a range of −$250M to +$50M, because the servers’ five-year life is itself an estimate, GPUs may age faster, and the spending in a light year like this one isn’t a fair guide E (estimate). Non-GAAP profit, at about $411M, is more than $518M away from that. On any honest reading, owners are getting no cash yet.

Over the full record the gap is even clearer. Ten years to 2025: free cash flow of $146M, stock pay of $1,486M, so $-1,340M for owners on $7.9B of revenue C (calculated). I don’t hold that against a young company. It does mean the whole value lies ahead.

Quality of earnings

Stock pay fell as a share of revenue from a peak of 21.1% in 2023 to 20.2% in the last twelve months, or 19.1% without the restructuring piece, which is slow progress C (calculated).²²² The company also spent $48.3M in 2025 on tax for staff share settlements, which sits in financing and so never touches free cash flow.³¹ Capitalized sales commissions ran ahead of amortization by $47.3M in 2025, which helps reported margin C (calculated).³ These are normal for the industry, and small. The non-GAAP gap is not small: non-GAAP operating income of $345M against a GAAP loss of $354M C (calculated).²³

Revenue quality is good. It is mostly subscription, spread over many customers, and billed ahead. Deferred revenue rose $223.8M in 2025 on $498.3M of added revenue.³ After commissions and leases, working capital brought in about $71M in 2025, 14% of added revenue, and I use 10% in the model C (calculated).³ The shift Seifert describes, away from a purely ratable model toward pools of funds and consumption, makes revenue less predictable and may make it larger.¹⁰ Reported net retention in the coming quarters will carry more noise.

The balance sheet

Cash and securities were $4,162.8M at June, with $3,293.8M of 0% notes.²⁵ Since then Cloudflare raised $2,462.3M net from the 2031 notes, spent $259.5M on capped calls, and, on its election in May, paid the $1,293.8M principal of the 2026 notes in cash and the excess in stock.⁷⁵ I make that about $5,072M of cash against $4,500M of debt, net cash of $572M, or $1.58 a share C (calculated). The notes pay no interest and fall due in 2030 and 2031. The balance sheet can’t hurt this company soon, and the credit line was amended in August to count $2.0B of cash in its leverage test.⁷ At $5B of cash and a share price this high, Cloudflare could buy almost anything it wanted with stock. That is a real strategic asset, and it lasts as long as the price does.

VIValuation

First, how this goes badly wrong

Before any numbers, I ask what would most likely ruin this investment, and work back from there. There are three ways, and only one of them needs the business to fail.

One: nothing goes wrong at the company, and the price falls anyway. At about 40 times trailing revenue the stock has no floor from cash. It has done this before: from November 2021 to November 2022 it fell 83% while revenue grew 49% C (calculated).¹²¹ It fell 23.6% in a single day in May 2026, the day after it reported a good quarter and announced the job cuts.¹¹⁹ It fell 9.5% on the last trading day of October with no company news that I could find, on a day when the S&P 500 fell about 1.8% C (calculated).¹³² Say that by the end of 2028 growth has slowed to about 20% and the stock trades at 10 times revenue, close to Zscaler today. On my bear revenue that is about $123 a share, down 56% E (estimate). The business could be fine and the owner badly hurt.

Two: machine traffic costs more than it pays. Cloudflare is betting that a world of agents sends far more traffic through its network. If most of that traffic comes from free sites or flat subscriptions, volume rises and revenue doesn’t. If compute moves to where the models already run, inside the clouds, Workers gets the small jobs and the clouds get the large ones. Gross margin then falls through the 70% floor, stock pay stays near 20% to keep engineers, and owner cash never arrives.

Three: a failure of trust. The product is a promise that the internet will keep working. On 18 November 2025 a database permissions change doubled the size of one file, the proxy software crashed, and core traffic failed for about three hours. Cloudflare called it its worst outage since 2019.³³ More than a fifth of the web sits behind that promise.⁸ Six months later the company cut about 20% of its staff, and it now has agents reviewing all production code. Prince expects that to make things more reliable.¹⁹ He may well be right. But one bad error at the wrong moment, or a founder walking out, would hit growth and the multiple together.

Working back, the cases below have to answer three things: how long growth lasts, whether stock pay falls, and what a share is worth when the growth ends. I treat those as the real assumptions. The rest is arithmetic.

What the price implies

At $280.70 and 360.8M shares, the market value is $101.3B, and net of $572M of cash, $100.7B C (calculated). That is 40 times TTM revenue and 35 times this year’s guide. There is no earnings multiple to quote: GAAP earnings are negative. On the guide of $1.25 to $1.26 of non-GAAP earnings per share, the price is 224 times, before stock pay C (calculated).² The owner free cash flow yield is below zero, against 4.94% on the ten-year Treasury.³⁴

The ordinary reverse DCF starts from today’s owner cash and solves for growth. Here today’s owner cash is negative, so it can’t. I solve for revenue instead, holding my margins. Each case runs fifteen years and then grows at a steady rate.

On my bull margins the price needs about 32% a year for ten years, to $47.2B. For scale, Salesforce reported $41.5B for its FY2026, as my Snowflake page notes, after two decades and many large acquisitions. Cloudflare would have to go past that in a decade, from $2.9B, without slowing much, and earn a far better margin after stock pay than it has ever shown.

The discount rate

I use 10.5%. The ten-year Treasury closed October at 4.94%, and Damodaran’s implied equity risk premium was 4.20% on 1 October.³⁴³⁵ Cloudflare’s 60-month beta against SPY is 1.37, which gives 10.7%; since listing it is 1.16, which gives 9.8%; over 36 months it is 1.71, which gives 12.1% C (calculated).¹³² I take 10.5% rather than 10.7% for one plain reason: it is the rate on my Snowflake page, the closest comparison on this site, and the two ought to be read side by side. A higher rate would only widen the gap. One point on the rate moves the base to $97 at 9.5% and $67 at 11.5%, up 22% and down 16% E (estimate).

Three cases

Each case runs from the guide midpoint for 2026, $2,867M, through ten years of stated growth, and then fades in a straight line over five years to a steady rate.² Each year I take revenue times a non-GAAP operating margin, subtract stock pay in full, tax what is left at 23% after using $1,859M of federal losses, which can shelter 80% of income in a year, add depreciation of about 10.5% of revenue, subtract capital spending, and add 10% of each added revenue dollar for working capital E (estimate).⁶ Interest on the cash is left out and the cash is added once. That is the same structure as my Snowflake page.

The base is ServiceNow’s path. ServiceNow had $2.6B of revenue in 2018 and $13.3B seven years later, 26.2% a year.¹³ My base has Cloudflare at $2.9B this year and $13.6B seven years on, a little more. That is one of the best records in software at this size, so I don’t think of it as cautious. Growth is 31% in 2027 and fades to 13% by 2036, at $20.7B. The non-GAAP margin climbs to 32%, past the 30% target management gave in June as reported, and stock pay falls to 9%.²³²⁴ Capital spending falls to 11%. The owner margin in 2036 is 18.4%, a little above the 15.7% ServiceNow earns now C (calculated).¹³ I would not expect a network that buys servers to beat a pure software company on this measure, so I think this is generous.

The bull is management’s plan, and then some. Growth starts at 36% and is still 16% in 2036, at $30.1B. The non-GAAP margin reaches 38%, stock pay 7%, and free cash flow before stock pay 31.7% of revenue, inside the 30% to 35% target as reported.²³²⁴ This is the case where agents pay, Workers scales and Act 4 becomes a business. It is worth $169.

The bear isn’t a disaster. Growth is 27% next year, then fades to 8%, and revenue still reaches $11.2B, about 2.6 times what Akamai earns today C (calculated). The non-GAAP margin stops at 25% and stock pay at 12%, so the owner margin is 9.3%. That is a good company that turned out to be a good security and delivery vendor, and not the operating system of the agent internet. It is worth $22.

I left out the second half of 2026 from the flows, because free cash flow and stock pay in that half about cancel on my numbers E (estimate). Weighted 25%, 50% and 25%, the cases give $88 a share E (estimate). The weights are judgments. They hardly matter here, because the price is above all three.

The obvious objection

A reasonable reader will say that a discounted cash flow always undervalues a business like this. The steady-state ending gives my base an implied value of about 16 times owner cash at the end of 2036, while it is still growing 13%. ServiceNow, growing 20%, trades at about 45 times owner cash today C (calculated).¹⁵¹³ So let me give the objection its full weight. Here are my base and bull with the business sold in 2036 at a stated multiple of that year’s owner cash.

Even at 50 times owner cash, in a decade, my base falls short of the price at $210. It would need 68 times. My bull gets there at 33 times, a rich but thinkable multiple for a business still growing 16%. So here is the most generous fair statement I can make: the price is reachable if the bull happens and the market still pays over 30 times owner cash a decade from now. Both have to be true.

Morgan Stanley’s reported method makes the same point from the other side. TheStreet says its $305 target in June used 48 times enterprise value to free cash flow in 2034, discounted at 11%.²⁴ Run backward on about 374M shares, that needs about $5.5B of free cash flow in 2034, before stock pay E (estimate). My base has $4.2B in 2034 before stock pay, and $2.4B after it. The difference between that target and my base is mostly two choices: the 48 times, and counting stock pay as free.

Is the model too harsh?

The base is 252% below the price, so I tested my own model for harshness before leaning on it. On growth I don’t think it is. The third-quarter guide implies fourth-quarter growth of 29% C (calculated)²³, and my base starts 2027 at 31%. The 2027 goal reported from Investor Day is growth above 30% with about a 20% operating margin, and my base has 31% and 19% ²⁴. My 2036 non-GAAP margin of 32% is above the 30% long-term target as reported ²³. On the rate, 10.5% sits inside the 9.8% to 12.1% my betas give.

It is cautious in three smaller ways. Cloudflare has beaten the midpoint of its revenue guide by 3% to 5% in each of the last three quarters C (calculated). If the second half beats by the same margin, 2026 lands about 2% above the guide I start from. I count cash flows at year end, and mid-year would add about 5%. And I fade growth to a steady rate over five years after 2036, when a business growing 13% could take ten.

So yes, it leans cautious, by up to two fifths on the base once every choice goes the friendly way E (estimate). It does not change the answer. With all four changes the base is $113 and the bull $255, both under $280.70. Only the bull at a 9% rate gets past the price. I keep $80 because the same structure sits under my Snowflake page, and I would rather not pick the kindest convention for each input.

How the base moves

Growth does most of the work, as you would expect. Stock pay matters less on its own than people think, and capital spending hardly at all. Even all the bull inputs together close only 45% of the gap.

What I can’t judge

Three inputs are beyond what anyone outside, and maybe inside, can know, and I would rather say so than print a false number.

• What agent traffic is worth. No filing gives revenue from agents or Act 4. Prince says what the internet’s business model changes to is “still an open question”.¹⁹ It could be a toll on trillions of small payments, or it could be nothing. I model it as a part of bull growth, not as a line. If it is worth a real toll, say $5B of 80% margin revenue by 2036, it adds about $41 a share to my base at a steady-state multiple E (estimate). So even a business of that size doesn’t bridge the gap.
• The gross margin of Workers. It isn’t disclosed. I show a range of owner margin instead, 9% to 25% in 2036.
• What 20% fewer people does. It may be the start of a much leaner company, in which case stock pay falls faster than even my bull has it. Or it may show up as an outage, or slower selling, in eighteen months. The third-quarter results will be the first full look.

There is also the recap. If it goes through after trial, each Class A share becomes one Class A and one Class C. The company expects the price to halve. All per-share values on this page are per Class A share as it stands today, so they would halve too.⁵

The margin of safety

My margin-of-safety price is $56: the base less 30%. I use the base, not the bear, because the bear still assumes a good company. I use 30% because almost all the value lies in cash flows that have not started, 55% of the base lies past 2041, and the stock has already fallen 83% once. On $56 the price would be about 7 times this year’s revenue, about where ServiceNow trades on the table above. That isn’t cheap by any older standard. It is what I’d need before saying the risk had been paid for.

The price against the range

$280.70 is 252% above my base value of $80, and 66% above my bull value of $169. On my base cash flows the price returns 5.8% a year, on the bull 8.3%, on the bear 3.4% E (estimate). I require 10.5%.

Against the next best use of the same dollar

The first alternative is a ten-year Treasury at 4.94%, which pays about what my base pays here, with no risk to the principal and nothing to go right.³⁴ An investor taking equity risk ought to want more than 0.8 points for it.

The best comparison on this site is Snowflake, which is growing as fast, 37% on product revenue, has about the same market value and also has negative owner cash, at more than twice Cloudflare’s revenue.¹⁵¹³ My page has it 140% above base. Cloudflare is further above its own. Microsoft at 44% above base and Broadcom at 25% above base each pay owners real cash today. Outside the pages I cover, ServiceNow is the sharpest test. For about the same $109B, a buyer gets $15.4B of revenue growing 20% and $2.4B of owner cash, against $2.5B growing 36% and none C (calculated).¹⁵¹³ Cloudflare has to grow into ServiceNow just to justify the same value today, and the buyer earns nothing for the years of waiting.

Two separate questions

How good is the business? Very good, I think, and maybe great. It has a cost advantage that took fifteen years to build, net retention rising, founders who stay close to the product, and a position in front of the machine traffic that everyone wants. How good is the price? Those are different questions, and here they have opposite answers. This looks like a great business priced for more than I can see it delivering even when it goes right.

Where this could be in 5 and 10 years

Value in year 5 is the present value, at October 2031, of the owner free cash flow after that date and of everything after year 15, at 10.5%. Year 10 is the same at October 2036. It is per share, on today’s 360.8M shares, in that year’s dollars. I leave out today’s cash and the notes.E (estimate)

For the base to be right at year 5, revenue has to reach $9.5B, 3.3 times this year’s guide,C (calculated) with growth still 23%, a non-GAAP margin of 28% and stock pay at 14%. That is an owner margin of 10.5%. At year 10 it needs $20.7B, a 32% margin and 9% stock pay, an owner margin of 18.4%. The base is then worth about $165 a share, and $38 of cash has been paid out on the way.

The bear and the bull differ on growth and on owner margin. In year 10 the bear has growth of 8% and an owner margin of 9.3%. The bull has 16% and 24.7%. Year-10 revenue is $11.2B against $30.1B. All three use 10.5%, and one point on it moves today’s base by 16% to 22%.E (estimate)

The most likely way the base is wrong is growth. If it runs three points below the base every year, my year-10 value is about $122, not $165.C (calculated) At $280.70 the base returns about −3% a year to year 10, and even the bull returns 4.7%.

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

Three things cut in Cloudflare’s favour and get less attention than they ought to. First, a good part of the fall in gross margin is good news in disguise. When a free developer starts paying, the traffic cost moves from sales and marketing to cost of revenue. The CFO calls that “literally a wash”, and Prince says it means the free tier is acting as the sales force for the developer platform.¹⁹ Second, the customer list is better than the count suggests. The CFO says more than 80% of major AI companies are customers, and in May Prince said most of the big labs are, while no single customer reaches 10% of revenue.¹⁰¹⁹⁶ Third, this company can borrow for nothing. In August it issued $2.5B at 0% due 2031, convertible 60% above the stock, and capped the dilution up to $854.12.⁷

The bull case

In plain words: the internet is being rebuilt for machines, and Cloudflare is the toll road and the construction site at once. Non-human traffic passed human traffic on its network in May, faster than management expected.¹⁰¹⁹ The developer base rose from about 3M to 7.4M in eighteen months, and developer recurring revenue rose 137% in 2025, as reported from the Investor Day.²⁴²³¹⁰ Management says it will reach a $5B revenue run rate before the end of 2028 and a 30%-plus operating margin in the long run, up from 20%-plus, as reported.²⁴²³

Prince is the case’s best advocate, and he says the business model of the internet is about to change over five years, with Cloudflare in the room as it does.¹⁹ On the Street, TheStreet reports that Morgan Stanley set a $305 target in June on 48 times 2034 free cash flow; I didn’t open the note.²⁴ Among the investors I checked, ARK has a position: 175,723 shares at 30 June, down from 210,090 at the end of 2025.³⁶ I did not find a published ARK model for Cloudflare. That date is four months old. The largest holders in the proxy were Capital World at 11.4% and Baillie Gifford at 5.3%, and their filings say nothing about why.¹⁴

Against reality

Today’s numbers support the first half of this. Growth is rising, net retention is 120%, and the agent traffic is really there. If Cloudflare does what it says, and hits $5B by 2028, reaches a 30% operating margin, and holds gross margin at 70%, my base case is roughly the result, and that is worth $80. So hitting the targets isn’t enough for this price.

The price needs my bull margins and about 32% growth for ten years, or my bull case and over 33 times owner cash in 2036. If they keep growing above 30% to the end of the decade, if Act 4 becomes a real line, and if stock pay halves, the price is possible. Today the business makes no owner cash, and it has no published revenue from agents. It helps to multiply it out.

Multiplied, that is 3% E (estimate). Good growth helps several of these at once, so I would allow perhaps 5% to 10% for all of them together. Each step looks reasonable. All six together do not.

What a 10x would take. Ten times $280.70 is $2,807, about $1.0T of market value. At 30 times owner cash that takes about $34B a year. At a 25% owner margin, my bull’s, that is about $135B of revenue by 2036, some 47 times this year’s, or 47% a year for ten years C (calculated). That is more than today’s revenue at every company in my rivals table combined, several times over. It has been done from about this size, but rarely. Nasdaq closes show NVIDIA first closed above $4.01 in July 2017 and hit $40.11 in May 2023, under six years, and Broadcom took nearly eight to do the same from early 2017.³⁷ Both already earned large cash profits at the start, and one had the greatest demand shock in the history of chips. For a company with no owner cash today it is close to impossible, and I wouldn’t build a case on it.

Where I come down

The pace of developer adds and the retention turnaround moved me: without them my base would start nearer 25% growth. The agent story did not, because no one has yet shown a dollar of it. A year of owner cash above zero, with stock pay falling and growth still over 30%, would raise my values most.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• Control and the courts. A trial on the recap is coming, with directors’ credibility at issue. If the founders win, the Class A holder’s voice gets weaker. If they lose, the founders may sell more slowly, or may lose the majority. Prince has threatened to leave before, and the record shows the board took it seriously.⁸
• Stock as currency. Staff are paid in stock. If the price falls hard, the same pay means more shares, or staff leave. In 2022 stock pay rose from 13.7% to 20.8% of revenue as the shares fell C (calculated).²¹ The amended 2019 plan passed in June with 212M votes against.³⁰
• Lumpier revenue. Pools of funds and consumption pricing make quarters harder to call, and the Q2 beat could reverse.¹⁰
• Interest income. $133M of the last twelve months’ free cash flow was interest on a $5B pile. A fall of one point in yields would take about $50M, a sixth of that free cash flow C (calculated).⁵
• Network cost. Servers and GPUs are the main capital cost, guided at 14% to 15% of revenue for 2026. The 10-Q warns that tariffs could raise what the equipment costs.⁵¹⁰
• Politics and content. Cloudflare is a chokepoint, and it gets pulled into disputes over what passes through it. The 10-K lists the risks of customers’ content and of regulators in many countries. Its stance on crawlers makes a new set of opponents.⁶²⁵
• Analyst dispersion. Yahoo Finance reports that Guggenheim’s target was $140 in June, against $305 at Morgan Stanley.²³²⁴ People who look at this for a living differ by a factor of two.