monday.com booked roughly four times as much revenue in 2025 as it did in 2021. Its stock closed on October 22 at a level 76 percent below the November 2021 high. Adobe’s revenue is up about 50 percent over the same stretch and its shares are down 61 percent. HubSpot’s revenue is up 2.4 times and its shares are down 69 percent. Wix has grown revenue by more than half and sits 77 percent below its February 2021 peak. None of these companies is shrinking, and all of them have spent the year being described as dead.
“SaaS is dead” has become a standard line in software coverage since the January selloff, and it has been used loosely. Customers have not stopped paying for Adobe, HubSpot or monday.com in any quantity that shows up in their results. What the market took away was the assumption that they would keep growing at the pace of 2020 and 2021, year after year, on top of customer bases that were already large. That assumption was a bubble in the ordinary sense, a price that required a future the arithmetic could not deliver. The bubble has popped, and what remains is a smaller price on a business that still exists.
History supplies both outcomes, and the difference can be checked. Groupon and Wish, neither of which sells software as a service, lost their share prices and then most of their revenue. Amazon and Cisco fell 94 percent and 89 percent in the dot-com crash and recovered fully, though Amazon needed almost ten years and Cisco more than twenty-five. The sorting question is whether the demand underneath was real and whether the price asked more of it than it could give. This editorial applies that test to software, and then to the American AI trade, which now carries many of the same assumptions on a larger scale.
What a 20.8x SaaS Multiple Assumed
Jamin Ball’s Clouded Judgement series, the most widely cited public tracker of cloud software valuations, put the median enterprise value at 20.8 times next-twelve-month revenue in mid-February 2021. On October 16, 2026 the same median stood at 4.5 times, a decline of 78 percent. Ball’s own pre-COVID benchmark for the median is about 8 times, so the group now trades roughly 44 percent below its pre-pandemic level and 78 percent below its pandemic peak.
A multiple of 20.8 times is a belief about how long a growth rate can last. Ball’s look-back at the third quarter of 2021 found median year-over-year revenue growth of 35 percent among public cloud companies, and the market was extrapolating something close to that for years. Compounding supplies the damage. Growth of 24 percent a year multiplies revenue 2.9 times over five years and 8.6 times over ten, by this publication’s arithmetic. Few companies of any kind have held that rate for a decade, and fewer still from a base measured in billions.
Individual companies show it more starkly. Using November 2021 closing prices and cover-page share counts, and ignoring net cash, this publication’s rough calculation puts Adobe near 21 times trailing revenue at its peak, HubSpot near 31 times and monday.com near 65 times. The same method today gives about 4.0, 3.6 and 3.7. A price of 65 times revenue is a claim about the middle of the next decade made in 2021.
Few people asked who would be paying for the next doubling while the multiples were still rising. Bain & Company’s 2025 technology report found that 43 percent of the value private equity created in software since 2010 came from multiple expansion, and described the era of buying a promising SaaS company and watching revenues and multiples climb together as drawing to a close. The question arrived late, after the prices had already fallen. This publication traced the same turn in how technology work is judged in its April analysis of the end of the easy tech era.
The Seat Count Was Always the Constraint
Software sold per seat grows when customers hire, when they add products to each seat, or when vendors raise prices. In the United States the first lever has barely moved. Bureau of Labor Statistics data show total nonfarm employment up 6.6 percent between November 2021 and September 2026, with the latest month preliminary. Professional and business services employs many of the white-collar workers these products are sold to. Its employment is up 2.1 percent and stands 1.4 percent below its November 2022 peak. The information sector is down 12.1 percent from its own peak.
The largest seat pool in software tells the same story with less drama. Microsoft reported Office 365 commercial seat growth of 17 percent in the June 2021 quarter. In each of its last two reported quarters it was 6 percent, on a base of about 450 million paid Microsoft 365 seats at the end of January. Growth now leans on price and on higher tiers, among them Copilot, which has passed 30 million paid seats, roughly 7 percent of the base.
Penetration is the quieter constraint: Bain put software adoption at about 95 percent in retail and manufacturing, and Zylo’s 2026 index reports that 36 percent of purchased licenses go unused.
Ball’s median expected growth for the following twelve months was 24 percent in January 2022. It was 17 percent by the end of that year, 14 percent in December 2023 and 12 percent in December 2024, and it stands at 13 percent this month. The compression was complete more than a year before the January 2026 selloff that is now attributed to AI. The BVP Nasdaq Emerging Cloud Index reached its high on November 9, 2021 and its low on November 9, 2022, down 64.7 percent, three weeks before ChatGPT launched. The first leg down was a repricing of growth.
Tomasz Tunguz has argued that the fastest-growing software companies stayed private after the IPO market closed in 2022, leaving an older, slower public sample, and KeyBanc and Sapphire’s 2025 survey expected private ARR growth to rise from 15 percent to 20 percent, a forward expectation that has yet to become a result. The deceleration among the public companies investors actually own is real. Whether it describes software as a whole is less clear.
Revenue Rose While the Stocks Fell
The survivors’ latest quarters show what a corrected bubble looks like from the inside. Adobe reported $6.76 billion of revenue for the quarter ended August 28, up 13 percent, with GAAP operating income of $2.35 billion and record operating cash flow of $2.52 billion, and it has cut its share count to 389 million from 472 million in January 2022. HubSpot’s second-quarter revenue rose 20 percent to $911.7 million with free cash flow of about $168 million, although its GAAP operating margin was only 4.8 percent. monday.com grew revenue 22 percent to $364.6 million, reported GAAP net income of $3.5 million and has repurchased $735 million of its shares this year.
Atlassian grew fiscal 2026 revenue 26 percent to $6.57 billion and produced $1.32 billion of free cash flow. Wix is the weakest of the five. It grew second-quarter revenue 15 percent to $563.1 million but posted a GAAP net loss of $76.4 million, ran a $1.62 billion tender offer and drew $500 million of credit in the same quarter. Its stock fell 27 percent on May 13 after a first-quarter earnings miss tied partly to compute costs at its Base44 AI unit.
Atlassian has guided fiscal 2027 revenue growth of about 13 percent, against 26 percent this year, and monday.com is guiding to 19 to 20 percent, against 27 percent in 2025. Even the best results are slowing, because slowing is what the arithmetic above predicts, and the lower multiple already assumes it. A stock that falls 76 percent as revenue quadruples is being repriced for what the 2021 price assumed the business would do.
The 90 Percent Club Is Smaller Than Advertised
The claim that many software stocks are down 90 percent or more needs a date and a list. Of sixteen publicly traded software names this publication checked against their all-time closing highs, one, Asana, is down more than 90 percent today, at 91.9 percent. Zoom is 83 percent below its high, Docusign 73 percent, Twilio 43 percent, ServiceNow 33 percent, Salesforce and Workday 32 percent, and Snowflake 22 percent. Datadog and Shopify set all-time closing highs earlier this month.
The 90 percent figure was closer to true in the spring. At the lows between February and June, Asana was down 96 percent, Wix 88.5 percent, Atlassian 87.5 percent, monday.com 87 percent and HubSpot 80 percent, and anyone who wrote the obituary then was describing real prices. Since the April bottom the software ETF IGV is up 46 percent and Atlassian is up 290 percent from its low. A handful of smaller software companies remain more than 90 percent below their highs on this publication’s price data. Squarespace, often named in this context, no longer trades. Permira took it private in October 2024 at $46.50 a share, about 27 percent below the roughly $64 closing high it reached in June 2021, according to a price-history aggregator.
The recovery has sorted the sector. Companies that sell data infrastructure and usage-priced platforms, such as Datadog and Snowflake, have regained most or all of their losses. Seat-priced applications such as HubSpot, monday.com, Wix and Adobe remain 61 to 77 percent below their highs. The market now treats software as two groups, paying up for products whose revenue does not depend on headcount and still discounting those whose revenue does. This publication’s March review of AI, SaaS and crypto in 2026 said markets would demand harder proof of returns and margins. The second-quarter results above show what that proof looks like when it arrives, and Wix shows what it looks like when it does not.
Where the ‘SaaS Is Dead’ Case Holds
Parts of the argument that AI agents will reduce human seats are visible in the data. Salesforce’s support organization shrank from about 9,000 people to about 5,000, as Marc Benioff described in an August 2025 interview, a change the company called a rebalancing. Challenger, Gray & Christmas attributed 120,136 announced U.S. job cuts to artificial intelligence in the first nine months of 2026, about 21 percent of the 573,195 total and the leading reason year to date, though only the fifth most cited in September alone.
The January and February selloff had identifiable triggers. According to market coverage, Anthropic’s Claude Cowork plugins, including a legal-sector tool released in early February, sent shares of RELX, Thomson Reuters and Wolters Kluwer sharply lower on February 3, and a widely circulated Citrini Research scenario on February 23 added to the pressure. IGV fell 18 percent in six sessions from January 28 and bottomed on April 10, 36.6 percent below its September 2025 closing high. Ball’s own reading of the first leg, written on January 30, was that there was no single trigger and that investors were questioning retention, terminal value and the discount rate.
The evidence on the other side is as concrete. Microsoft’s seat count is still growing, at 6 percent. Salesforce reversed course in December 2025 and offered AI agents under seat-based licenses because customers wanted predictable costs, and Gartner’s Jan Cook said at the time that customers “appear to be arming staff with AI rather than cutting headcount.”
The two sets of evidence fit together if AI is read as a change to the terminal value of a seat, with no measured loss of seats. Investors are lowering what they will pay for the last years of a forecast, because an agent might one day do the work a licensed user does today, while most existing seats remain in place. That matches this publication’s earlier analysis of AI deflation and SaaS inflation, which found customers pushing back on software rent as AI capability gets cheaper, and it matches the split between usage-priced and seat-priced names above. It leaves the arithmetic of the earlier sections untouched, because growth had halved before this year’s agent announcements arrived.
Groupon and Wish Lost the Business as Well as the Stock
Groupon and Wish are consumer marketplaces outside software, which is why they belong in this argument. Their stock charts look like the worst of software’s. Their businesses behaved differently, and the difference shows what a real death looks like.
Groupon priced its IPO at $20 in November 2011 at a valuation of $12.7 billion. The stock peaked on its first day of trading at $31.14, about $623 on the share count after the 1-for-20 reverse split of 2020, and it closed on October 22 at $16.24. Its market value is about $660 million, roughly 95 percent below the IPO valuation. Revenue was $3.19 billion in 2014 as originally reported and $498 million in 2025, down about 84 percent, and a restated series that peaks at $3.14 billion in 2016 gives the same answer. Active customers fell from 52.7 million in 2016 to 16.1 million in the second quarter of 2026.
Groupon still exists, and on its own measures it is stable: $69.3 million of adjusted EBITDA and $49.9 million of free cash flow in 2025, $296 million of cash at year-end, and 2026 revenue guidance of $513 million to $523 million. On a GAAP basis it lost $81.1 million from continuing operations. It survived by shrinking to the size its demand could support.
Wish priced its IPO at $24 in December 2020 at a valuation of roughly $14 billion to $17 billion and closed its first day below the offer price. After the 1-for-30 reverse split of April 2023, the stock’s January 28, 2021 intraday high of $32.85 corresponds to about $985 on today’s share count, and the $24 offer to $720. Revenue fell from $2.54 billion in 2020 to $287 million in 2023, down 89 percent. In April 2024 ContextLogic, Wish’s parent, sold the marketplace to Singapore’s Qoo10 for $173 million as announced, about $161 million after adjustments and roughly 1 percent of the IPO valuation. Whether the app still operates at scale is unclear.
The stock still trades, and the story turns odd. ContextLogic bought the salt producer US Salt for $907.5 million in a deal that closed in February and now reports its revenue from salt. Its shares trade over the counter at about $15, having sat near $7 to $8 from late 2024 into this year. A holder who kept the Wish stock now owns a salt company with $211.5 million of debt.
Stock charts alone cannot separate these cases from the ones that matter for software. Zoom fell 83 percent from its October 2020 closing high while its revenue rose from $4.10 billion in fiscal 2022 to $4.87 billion in fiscal 2026. Snap is down about 92 percent from its 2021 peak and still reports revenue growth of 19 percent. Teladoc is down about 98 percent with revenue roughly flat since 2023. Peloton, down about 97 percent, sells roughly 39 percent less than it did at its fiscal 2021 peak. The first three are multiple compression on businesses that kept operating at scale. Peloton, Groupon and Wish are demand that was pulled forward and never came back. The revenue lines of Adobe, HubSpot and monday.com put them in the first group.
Amazon and Cisco Took 10 and 26 Years to Recover
The usual defense of a corrected company is that the dot-com survivors came back. They did, on a timescale that rarely appears in the defense. Amazon closed at $106.69 on December 10, 1999 and at $5.97 on September 28, 2001, a decline of 94.4 percent. It did not close above its 1999 high until October 23, 2009, nearly ten years after the peak. Cisco closed at $80.06 on March 27, 2000, fell 89.3 percent to $8.60 by October 2002 and first closed above that peak on December 10, 2025, after more than twenty-five years. Those figures are price only. Cisco has paid a dividend since 2011, so a holder who reinvested recovered sooner.
Amazon is often explained by its warehouses. The record is messier. Its 2001 annual report recorded $181.6 million of restructuring-related expense, including impairments tied to closing a fulfillment center in McDonough, Georgia. Webvan, which committed to roughly $1 billion of automated warehouses, filed for bankruptcy in July 2001. What separated the two, according to Brad Stone’s account and the Commoncog case study built on it, was financing and demand. Amazon had sold $672 million of convertible bonds in early 2000, held about $997 million of cash and securities at the end of 2001 and had customers who kept coming back.
Faster recoveries exist. Facebook went public at $38 in May 2012, fell to $17.55 that September and was back above its offer price by August 2013 on revenue that kept compounding. Carvana fell 99 percent from its 2021 peak to its December 2022 low, regained the old high in July 2025 and now sits about 22 percent below it again.
Across these cases, the repricing took between four months and three years. The recovery, where it came, took between about one and twenty-six years, and it came for companies whose demand turned out to be real and whose balance sheets lasted long enough to prove it. For investors who bought software at 2021 multiples, growth below the pace the price assumed is enough to produce a shortfall, and median growth is there now.
What AI Valuations Assume
The same test applies to the AI trade, and the inputs are public enough to run it. Four hyperscalers have guided to roughly $725 billion to $745 billion of capital spending for 2026 between them: about $175 billion at Microsoft including finance leases, $195 billion to $205 billion at Alphabet, about $220 billion at Amazon and $135 billion to $145 billion at Meta. Microsoft’s figure reflects an accounting reclassification, and the company said its plans were unchanged. OpenAI and Anthropic, two of the largest customers for that capacity, have combined annualized revenue of roughly $115 billion to $135 billion by this publication’s sum of reported figures. Capital spending runs five and a half to six and a half times the labs’ combined run-rate. Definitions differ, so the ratio is an order of magnitude.
The valuations attached to that revenue sit in the range software occupied at its peak. OpenAI’s last closed round valued it at $852 billion in March, and it is reported to be negotiating a new round at more than $1.4 trillion before the new money goes in. OpenAI’s run-rate is about $50 billion, the figure the Financial Times reported on a basis comparable to Anthropic’s (other reports range from $40 billion to $70 billion). That puts the March price at 17 times revenue and the new one at 28 times. Anthropic’s $965 billion round in May was struck at about 15 times its $65 billion run-rate, and reported listing expectations of $1.5 trillion to $2 trillion would put it at 22 to 30 times. SpaceX, at roughly $1.9 trillion to $2.2 trillion depending on the share count, trades near 34 times annualized third-quarter revenue. The median public cloud software company sits at 4.5 times, against 20.8 at its February 2021 peak.
What would justify those multiples has been estimated several times. JPMorgan Asset Management’s January analysis put the revenue AI must generate at about $650 billion a year in perpetuity to earn a 10 percent return on the capital being spent, which it translated to roughly $35 a month from every iPhone user. Bain’s 2025 technology report estimated that AI companies would need about $2 trillion of annual revenue by 2030 and would still fall about $800 billion short even after reinvesting savings. Each sets out how large the revenue has to be.
Consumers cannot supply it. ChatGPT had about 50 million paying subscribers among 900 million weekly users in February, and a card-panel study by a16z and YipitData found that 4.5 percent of U.S. consumers pay for ChatGPT, Gemini or Claude. Fifty million subscribers at $20 a month is about $12 billion a year, a fraction of a single hyperscaler’s quarterly capital budget. The revenue that matters is enterprise, API and coding, and it depends on companies reaching returns that surveys do not yet show: PwC’s January survey of 4,454 chief executives found that 56 percent reported no significant financial benefit from AI. This publication’s coverage of enterprise AI spending and the missing returns and of pilots that stall before production reached the same conclusion from the other direction.
The losses and commitments are large relative to the revenue. Anthropic’s draft filing, as reported by the Wall Street Journal, shows a first-half net loss of $80.7 billion, about $72 billion of it non-cash charges on financing instruments, and compute commitments of up to $750 billion. A July OpenAI planning deck, reported by the Financial Times, projects cumulative free cash flow of negative $278 billion from 2026 through 2030 against roughly $856 billion of compute spending. Azure passed $100 billion in annualized revenue, but Microsoft still has not disclosed Copilot’s dollar revenue, a gap this publication covered in its analysis of Microsoft’s July earnings.
The funding is also changing character. The International Monetary Fund’s October stability report counted 14 circular financing arrangements among AI firms, with about $300 billion of equity and $650 billion of debt set against $1.4 trillion of promised future revenue, and managing director Kristalina Georgieva described the economy as “standing on one leg, the AI leg.” Meta reported a cash outflow, its first in more than a decade, and announced a split into voting and non-voting shares to raise capital.
The tape on October 22 showed some fraying. The Nasdaq Composite fell 2.6 percent, its worst day since early June, the Philadelphia semiconductor index fell 6.7 percent to more than 26 percent below its June 22 record close, and Nvidia closed at $212.21, 11.3 percent below its October 6 closing high. Chips have been in a bear market since the June peak, and the indexes show a modest pullback, short of a broad AI correction.
Where the SaaS Comparison Breaks Down
The software comparison has limits. Software never grew like this at this scale. Anthropic’s annualized revenue went from roughly $1 billion at the end of 2024 to about $9 billion a year later and more than $65 billion by the end of July, and its recognized quarterly revenue rose from $4.7 billion in the first quarter to $11.6 billion in the second and $16.8 billion in the third, the last figure told to investors and unaudited.
Demand signals are concrete as well. Google Cloud reports a backlog of $514 billion and Oracle $664 billion of remaining performance obligations. Nvidia reported second-quarter revenue of $96.2 billion, up 106 percent, well above the 40 percent growth that this publication’s May analysis of Nvidia’s capital-spending thesis said the stock was pricing in. Nvidia says it is supply-constrained. At roughly 13 times annualized revenue it is valued far below SaaS in 2021. Bears reply that much of the backlog comes from the labs, whose ability to pay depends on raising more money. That reply is accurate and concerns financing.
The pool the revenue is drawn from is far larger than software’s. U.S. compensation of employees ran at an annualized $16.35 trillion in the second quarter, so JPMorgan’s $650 billion hurdle is about 4 percent of it. If AI agents capture a small slice of labor spending, the hurdle clears. That premise underlies the labs’ own pricing and explains why a seat-count argument understates the opportunity. It also answers the objection that the world lacks the people or the money to pay for AI. In aggregate it has both.
The SaaS episode cuts against anyone using it to argue that AI has no value. Software was marked down this year because investors believed AI could do the work software does. A sector cannot be repriced for being replaced by a technology that is worthless.
The bearish case narrows to whether the revenue arrives on the schedule the prices need, at margins that repay more than $700 billion a year of capital spending, from customers beyond a few labs and hyperscalers, and with financing that does not depend on the next round closing. The prices hold only if AI succeeds on a timetable that the labs’ reported losses suggest has not yet been met. Software investors learned in 2021 that you cannot pay 20 times revenue for a growth rate and then keep the multiple when the growth rate halves. AI investors are paying 15 to 34 times revenue for growth that has not yet halved, and growth that has quadrupled in a year cannot repeat on a base this large.
SpaceX Gave Back Most of the Pop
SpaceX’s IPO is a clean test of the thesis because the facts are public and the stock has swung widely. SpaceX priced at $135 on June 11, began trading on June 12 and closed its first day at $160.95, a gain of 19.2 percent. It reached a closing peak of $201.80 on June 16, 49.5 percent above the offer. It then fell below the offer price, to a closing low of $108.27 on August 5, 19.8 percent under it, through a series of lockup expirations that this publication followed in its coverage of the first lockup short squeeze and the second tranche. It closed on October 22 at $144.75, down 9.1 percent on the day.
Measured against the offer price, the shares are up 7.2 percent. Measured against the first-day close they are down 10.1 percent, and a buyer at the June 16 closing peak is down 28.3 percent. The index comparison is modest: the S&P 500 is up about 3 percent since the listing and the Nasdaq Composite is roughly flat. A holder who received shares at $135 has modestly beaten the market, while buyers from the first day onward trail it. “Disappointing” is a fair word for the second group and the wrong word for the first. This publication’s June analysis, SpaceX Listed at Twice Its Value. Crypto Did This First., noted that Morningstar valued the company at $780 billion and Aswath Damodaran at $1.3 trillion against the $1.77 trillion at which it listed.
The business delivered. Third-quarter revenue was $14.58 billion, up 161 percent, and the AI segment, now more than half of the total, contributed $7.56 billion. Net income was $4.4 billion, capital spending was $19.6 billion, and the company guided fourth-quarter capital spending to $35 billion to $40 billion. The 9.1 percent fall on October 22 coincided with that guidance, funding worries, the disclosure that special committees were examining a merger with Tesla, and a lockup release of up to 1.3 billion shares, the largest yet. The capital-spending guidance, the funding worries and the supply of shares are forms of the same pressure: a price that assumed a great deal meeting the question of who pays for it, and who sells. Another 328 million shares come free on October 26 under the published schedule.
The SaaS Bubble Popped and the Test Moves to AI
Reality comes back as a price, seldom as an obituary. The software products behind Adobe, HubSpot and monday.com are in use and growing, and what has been repriced is the future that 2021 paid for in advance. Groupon, Wish and Peloton show what happens when demand was borrowed from a moment. Amazon, Cisco and Facebook show what recovery looks like when demand was real, and that it takes years.
The SaaS record puts one question to AI: at what point does the price need more from customers than they can pay? For software the answer arrived in 2021 and was plain by 2023. For AI, the capital-to-revenue ratio, the circular commitments and the shift toward debt put it somewhere in the next several quarters or years. The prices can break while the technology proves useful. They broke for software, and software is still being sold.
Microsoft reports on October 28 and will be asked again to disclose Copilot’s revenue, Atlassian reports on November 5, and Groupon is expected to report in early November. Anthropic has been reported to be preparing a public filing, with marketing from the week of November 9 and a listing as early as mid-November, which would put detailed financial statements from one of the two largest AI labs in front of public investors. Each is a small test of the same question: whether the growth the price assumed is arriving.
Investors who paid for 2035 in 2021 are learning the answer one earnings report at a time.
