But the SaaSpocalypse was in many ways a confusing signal. SaaS equity prices have since returned to roughly pre-SaaSpocalypse levels (see below). While market valuations are by definition a prospective measure, they should in part be informed by recent actual performance. Plus, there’s the concern of whether the SaaSpocalypse shock itself affected SaaS revenues—for example, by making hiring or new funding rounds more difficult for these companies.

To help answer this question, we developed the Stripe SaaS Index, a weekly same-business measure of the health of nominal non-AI SaaS revenues, using Stripe pay-in volume.1 The flagship aggregate index covers an average of 72,000 businesses a week so far in 2026. Because it is a same-business index, the Stripe SaaS Index captures trends in the businesses themselves without being affected week-to-week by Stripe’s own market share or changes in the composition of businesses coming onto or off of Stripe. Because it is weekly, the Stripe SaaS Index can pick up on industry shifts before lower-frequency public data or quarterly financial reports. Further details are at the end of the post; we’ll be hosting and updating the indices on our website.

To be explicit, pay-in volume is not a replacement for an income statement. Our metric cannot see payments processed off Stripe or account for a business’s operating expenses and margins. It should instead be interpreted as a barometer for trends in gross SaaS demand that filters out Stripe-specific factors.

Overall SaaS revenue growth actually accelerated through the SaaSpocalypse

The Stripe SaaS Index tells several narratives about the SaaSpocalypse and its aftermath. First, recent SaaS revenue growth has remained strong and even accelerated from its end-2025 pace. This is clearest in the year-on-year change, which is now above 30%, at or above the prior highs that prevailed in the summer of 2023. Higher-frequency growth is more volatile but so far appears to be trending higher throughout 2026, especially at 13- and 26-week cadences.

Another way of illustrating this is with the level of the Stripe SaaS Index—which shows cumulative growth rates over time. The index’s level has shifted above its pre-SaaSpocalypse trend and is now 4% higher. Between the beginning of May 2025 and the end of December, the index grew 14%, indicating solid same-business nominal revenue growth on average. Since January, the index has grown another 21% over the same length of time. This does not necessarily mean the market reaction in early 2026 was irrational or won’t show up eventually in medium-term SaaS revenues, but it does suggest the market reaction was almost entirely prospective and so far of little relevance to the short-run prosperity of SaaS firms.

This growth resilience was partially driven by larger firms…

Second, some but not all of the resilience story is about large SaaS businesses weathering the market shock. To test how typical the growth of our flagship indices are, we constructed an alternative index chained using equal-weighted median pay-in volume growth among SaaS businesses. Like our aggregate indices, the median index is same-business and resilient to changes in the composition of businesses on Stripe, but because it is unweighted, it is less indicative of the SaaS industry in aggregate and more reflective of the typical SaaS business, regardless of size. The story with our median index is different from the aggregate index: the typical SaaS business did in fact see a slowdown in weekly revenue growth during the market-sell-off, a slowdown that appears to have begun in late 2025. Once the early 2026 sell-off ended, however, typical growth reaccelerated sharply above its 2025 pace—such that cumulatively, our median index is now right back in line with its pre-SaaSpocalypse trend.

…but young SaaS businesses saw extraordinary acceleration

Third, young SaaS businesses saw particularly fast growth on average. While both young SaaS businesses (less than one year since their first Stripe charge) and mature SaaS businesses (one year or greater) accelerated though the SaaSpocalypse, the two entered the sell-off under different contexts. Young SaaS firms saw revenue growth rise sharply beginning in mid-2025; it kept accelerating throughout the SaaSpocalypse. Only recently has young SaaS revenue growth started cooling modestly, but growth remains well-above its 2025 pace. Mature SaaS in contrast saw its revenues accelerate in late 2024 and early 2025, and then begin the decelerate in late 2025 before the sell-off. The SaaSpocalypse if anything arrested this deceleration and hosted a resurgence in mature SaaS growth, now back to roughly its mid-2025 pace. Both patterns are the exact opposites of what we would expect if there were a SaaSpocalypse-driven retrenchment, and the strong performance of young SaaS businesses is, if anything, a positive harbinger of SaaS over the short and medium term.

Part of what might explain the recent relative acceleration for young businesses is a jump in AI usage. While Stripe can’t see all AI usage by a business, we do see certain signals such as Model Context Protocol (MCP) requests, agentic CLI requests, and sandbox claims. This proxy adoption has been rising gradually for both young and mature SaaS businesses. As recently as the beginning of 2026, proxy adoption—measured as the presence of at least one of our AI signals as a share of businesses—was five percentage points higher for mature SaaS. But then the gap closed rapidly, and it flipped in April. Now, young SaaS firms are four percentage points more likely to use Stripe AI tools than mature ones.

US SaaS has sustained its SaaSpocalypse acceleration

Fourth, there are notable geographic differences. After a 2025 that saw somewhat slower growth than 2024, the US saw clear acceleration going into the SaaSpocalypse, bringing growth rates back to 2024 levels and, more recently, even above. Cumulative US growth is slightly above its pre-October 2025 trend. Europe meanwhile accelerated into the market sell-off but has not sustained those higher growth rates, instead returning back to around growth rates that prevailed in late 2024 and early 2025. Nevertheless, Europe too is cumulatively above its pre-sell-off trend. The UK and the rest of the world saw growth peak in the summer of 2025 and gradually cool from there, with no clear discontinuity around the SaaSpocalypse. Cumulative UK growth is slightly above trend while it’s slightly below for the rest of the world.

Healthcare, retail, and professional services saw particular strength

Fifth, there are some key industry-level differences within SaaS. Healthcare accelerated sharply through the SaaSpocalypse, while retail and professional services have also been beating their pre-sell-off trends, but by smaller margins. Financial services and other sectors are in line with trend. Broadly speaking, SaaS revenues appear to have been resilient post-SaaSpocalypse across sectors.

One area where we do not see wide differences by industry since the SaaSpocalypse has been proxy AI adoption, suggesting this measure alone cannot explain the growing healthcare wedge in 2026.

Our goal is that the Stripe SaaS indices prove a helpful tool for evaluating the real-time health of the industry. Markets are forward-looking and fast-changing, so only time will tell how ongoing demand and valuations interact with one another. In the meantime, we will be following our metrics and others’ closely, and we aim to publish other Stripe-based metrics on related topics in the near future.

Methodology

The Stripe SaaS Indices are weekly indices of pay-in volumes (customer payments) received by non-AI SaaS companies and processed through Stripe. The aggregate indices—our flagship index and the subindices by geography and industry—are Fisher indices: that is, geometric means of week-to-week pay-in volume growth of the sample of companies that appear in weeks T-1 and T, respectively, and chained over time, making them compositionally-adjusted same-business indices. Thus, week-to-week fluctuations do not reflect entry or exit of businesses on Stripe. We further filter out effects of rapid ramp-up and ramp-down on Stripe by winsorizing (capping) extreme outlier contributions to growth. The median index is not a Fisher index, but rather the unweighted median growth rate of SaaS businesses with nonzero pay-in volume in both weeks T-1 and T; otherwise, it is chained, seasonally adjusted, and averaged similar to the aggregate indices. Underlying pay-in volume is in US dollars using a fixed-period exchange rate; this means weekly index fluctuations also do not reflect currency movements. We seasonally adjust the weekly indices and apply a Kalman filter to remove volatility.

This debate has played out before. During the COVID-19 pandemic, work from home provoked predictions about the end of cities. So too did the rise of email and instant messaging, and before that, the telephone, and earlier still, the telegraph.

There are many reasons to believe that technology could weaken agglomeration effects, but history shows that the substitution of technology for in-person interaction is not straightforward. Cities benefit from both centripetal forces that draw firms and individuals in, such as shared infrastructure, amenities, and deep labor markets; and from centrifugal forces that push people out, such as congestion, high housing costs, and crime (Krugman 1996). New technological waves often act in both directions, and it is difficult to predict ex ante what the net effect will be.

Overall urbanization has been rising for over a century, but specific cities and neighborhoods within them rise and fall relative to each other. Jobs move to new places for two reasons: existing companies grow, shrink or relocate, and new companies get founded elsewhere. Stripe’s data allows us to observe the latter particularly well.

We find that AI-era businesses (which we define as companies that were started from 2024 and on, and which includes both AI companies and non-AI companies) are far more distributed than before. Close to 40% of new businesses on Stripe this year are in metros with fewer than one million people, up 10% relative to four years ago. This year, twice as many new businesses were formed per person in Cheyenne, Wyoming as in New York City; in 2022, the rates of business formation in both cities were equal. And even for businesses still forming within major metros, they are increasingly forming in the outer suburbs rather than in city centers and high-density areas: the share of businesses forming in outer suburbs is up by as much as three to four percentage points in metros such as Houston, Austin, Tampa, Orlando, and Dallas in the last few years.

There are two caveats to the AI-era story. First, there have been other factors influencing the spread of business formation, particularly the post-COVID-19 rise of remote work, and it might be challenging to entirely disentangle these. Second, not all AI-era businesses are more distributed geographically. In fact, the companies that are most responsible for AI development itself—AI labs and AI product companies—continue to cluster in major metro areas.

New businesses in the AI era are less concentrated in major cities than in the past

At first glance, AI-era business formation appears to be broadly distributed across the United States. While larger metros do on average see higher sign-up density than smaller metros and micropolitan areas, the distribution of sign-ups is not especially concentrated. We find smaller metro areas like Hinesville, Georgia (population 90,000), Cheyenne, Wyoming (100,000), and Fayetteville, North Carolina (390,000) among the densest hubs for new business formation. Meanwhile, major metro areas that one would expect to be at the core of this surge in new businesses—like New York, Los Angeles, or San Francisco—are seeing below-average business formation relative to their sizes.

To see how this compares to historical trends, we map Stripe sign-ups relative to population across US metro areas larger than 200,000 people from 2019 (which we use as our pre-COVID-19 baseline) to 2026. Micropolitan areas are excluded from this map for legibility but included in our analysis. Here we see that the current AI-era distributed formation is actually a continuation of a broader shift in entrepreneurial activity away from major metropolitan areas (larger bubbles getting lighter) and toward secondary metros (smaller bubbles getting darker).

Business formation is up across the country, so to understand relative winners and losers, we need to look at how the share of businesses forming in different locations is changing. This confirms what the map suggests, which is a gradual move away from larger metro areas (top 10 and metros with more than one million people) and toward smaller metros and micropolitan areas.

The biggest jump happened during COVID-19 as a consequence of work from home, but the general trajectory has continued in recent years and appears to be reaccelerating since 2024, closely tracking the AI era. The shift we’re seeing now is also reflected in Census data and is roughly one third as large as the COVID-era move away from major cities.

Within major cities, AI-era businesses are also less clustered in city centers

Within major cities, we also see that Stripe sign-ups are widely distributed outside of city centers. If agglomeration effects were key to AI-era business formation, you might expect to see the share of sign-ups in city centers and high-density areas vastly outweighing population share. Instead, sign-ups track population distribution within cities closely, and close to 80% of sign-ups in major metropolitan areas are outside of the city center and high-density inner ring.

What’s more, this distribution marks a change from previous waves of business formation. Much like the trend away from major metros, the shift in business formation away from city centers and toward outer suburbs has reaccelerated recently, relative to the post-COVID-19 plateau. Houston saw 37% of new businesses forming in the outer suburbs this year, up four percentage points relative to 2023. While this is a particularly stark example, the general trend holds across most major metros. The COVID-19-era trend is well documented in the literature (Ramani and Bloom 2021, Decker and Haltiwanger 2023), though to our knowledge, the recent reacceleration in this trend has not been covered elsewhere.

Disentangling AI from work from home is complicated

The data on AI-era business formation would thus appear to be consistent with AI dampening the value of agglomeration in major metros. But the extent to which AI is independently causing this change is hard to untangle. One confounding factor is the interaction between AI and work from home. For example, AI might be unlocking latent entrepreneurial potential among individuals that relocated with work from home policies—a continuation of the entrepreneurial activity spawning from remote work documented in the literature (Kwan et al. 2024).

If this were true, we might expect to see businesses disproportionately forming in areas with high levels of work from home. Our data does not currently support this. Business formation appears uncorrelated with work from home intensity, and that holds whether we look across metro areas or at the distribution of businesses within them. Still, this is not enough to conclude that AI-era business formation would be as widely distributed absent work from home.

AI work itself is immune to this effect

That AI-era business formation is broadly distributed does not mean that agglomeration does not matter with Al. In general, knowledge at the technological frontier is highly tacit, and frontier firms cluster in major cities to benefit from knowledge spillovers in their domain. The field of AI development itself is no different. If anything, the speed at which innovation in AI is moving would suggest that the proximity of cities should be even more valuable.

Stripe data allows us to test this empirically. We can categorize AI-era businesses in three groups, according to their distance from the technological frontier. First are the AI labs themselves. Then there are businesses in which AI is part of the product. Finally, there are businesses that are AI-enabled, but that are building and selling non-AI products.

Aligned with our hypothesis, these different types of businesses show different spatial distribution, with those closest to the technological frontier clustering most tightly in big cities. Almost 80% of AI businesses closest to the technological frontier (the labs) are in San Francisco alone. Among AI product companies, which are also innovating close to the frontier, 50% are in San Francisco or top 10 metro areas. Finally, it’s the AI-enabled businesses that are most widely distributed across the country.

Conclusions

• US business formation in the AI era appears to be more decentralized than previous waves: a higher share of firms are forming outside of major cities; and even among firms forming in major cities, a growing share are forming outside the city centers.

• We saw this trend during COVID-19 with the rise of work from home, but it has reaccelerated in the last few years, coinciding with the advent of AI.

• While it’s hard to fully disentangle this shift from changes driven by work from home, the distribution of AI-era businesses appears uncorrelated with work from home intensity at the city level.

• The subset of new firms closest to the technological frontier—labs and AI product companies—continue to cluster more tightly in San Francisco and other major metro areas, presumably because of the continued benefits of agglomeration for innovation.

Alongside this mounting evidence of micro productivity gains, aggregate or macro productivity in the US has accelerated. US labor productivity has grown at roughly 2.5% over the past year—a meaningful outperformance over the 1.6% annual average of the last two decades.

It now looks highly likely that the US is in a period of high productivity growth, at least for the time being. Our Markov-switching model—a tool used by economists to judge the likelihood that an economic measure is behaving meaningfully different than before—suggests a 93% probability the US has moved into a “high” productivity period, though as the chart below shows, this alone tells us very little about how long a high-productivity regime will last.1

The natural assumption might be that the boost to aggregate macroproductivity is a simple reflection of underlying microproductivity gains from AI at the worker and firm level. But in fact, while AI is likely part of the story, it’s not in the way most might think.

To start, there are reasons to be skeptical that even well-measured microproductivity gains are moving the needle much at the aggregate level at this point. We would expect, for example, that strong microproductivity gains would show up in estimates of total factor productivity (TFP). TFP is formally the portion of productivity growth not explained by increasing the amount of capital or labor being deployed in the economy; more informally, TFP is what economists think of as “pure” technological and efficiency growth, with the important caveat that TFP cannot be directly observed, and so must be estimated using economic models. If AI was making workers more productive at the frontier, TFP estimates should be accelerating just like overall labor productivity. But they aren’t: despite the strong headline labor productivity numbers, TFP growth has been near zero over the past year under the estimates developed by John Fernald and maintained by the San Francisco Fed.2

Our same Markov-switching model shows a less-than-20% probability the US is in a high-TFP-growth period.

Moreover, industry-level productivity gains bear little relation to AI adoption. Combining recent AI adoption rates from the Census Bureau’s Business Trends and Outlook Survey (BTOS) alongside the Chicago Fed’s quarterly industry-level labor productivity (which we extend two quarters using its methodology), there is a positive, if modest, relationship between current AI adoption as self-reported by the business and recent productivity growth (top panel in the following image). But the same sectors that have adopted AI most heavily also showed stronger productivity growth over 2016–2019, when generative AI did not exist (middle panel). Strip out those pre-existing trends, and the residual correlation between AI adoption and recent productivity growth falls to essentially zero (bottom panel).

Put another way: the sectors with high AI adoption were already the ones from which we’d expect stronger productivity growth given the pre-pandemic experience. That doesn’t necessarily mean AI adoption isn’t weighing on aggregate productivity at all–—high pre-pandemic productivity might have fueled desire or ability to adopt AI now—but it does mean that such a relationship isn’t clear in the sectoral data.

This raises an obvious question: if there’s evidence that AI leads to microproductivity gains on specific tasks, why wouldn’t that be associated with stronger productivity growth in aggregate? A different strand of recent research gives one possible answer: AI may speed up bounded tasks without loosening the downstream constraints that determine final output. Economists at the Bank of Korea find that generative AI adoption reduces work time by 3.8% (about 1.5 hours per week), similar magnitudes to the Dillon paper cited earlier, but they also find that these savings have essentially no relationship with realized output growth unless workers have autonomy or strong incentives to reallocate the saved time productively (Suh et al. 2026). Other research shows the same attenuation in software: AI coding tools sharply increase commits and code volume, but the gains shrink at each higher production stage—from pull requests, to projects, to shipped releases—because review, integration, testing, release, discovery, and user adoption remain human bottlenecks (Demirer et al. 2026). That is, it speeds the inner loop, but not (yet) the outer loop, which becomes the binding constraint. In short, AI can create task-level efficiency before firms have redesigned workflows, incentives, and complementary processes enough for those gains to appear in aggregate productivity.

But if the human bottleneck hypothesis were correct, then why is aggregate labor productivity accelerating in the US? What appears to be lifting US labor productivity, rather than microproductivity gains, are macroproductivity gains from AI—specifically companies running their existing capital harder. Think longer runs of factories already built, more utilization of server racks and GPU clusters already paid for, and more occupancy of existing hotel rooms. Economists call this “capital intensity” or “utilization.” Higher capital utilization represents real economic gains, but it’s not the same as microproductivity. We can see this in the rising contribution of utilization in the Fernald TFP estimates, which—other than during the pandemic bounceback—are higher than they have been in decades. This is not the same as “capital deepening” (growth in measured productivity from expanding capital supply—for example, through capital investment). But utilization and deepening are often correlated: a firm that is motivated to boost capex will also likely be pushing its existing capital harder. Fernald himself and coauthors come to a similar conclusion in a recent VoxEU piece (Boyle, Fernald, and Li 2026).

None of this is a pessimistic read of AI. For one, much of the research is based on GPT-2 through GPT-4 era models. If task-level AI gains in the 10%–40% range are multiplied in more recent models (highly plausible anecdotally), if firms are getting better at integrating models into their workflows, and if they are sustained and become more widely diffused, then eventually those microproductivity gains will move the dial more in the aggregate data. There’s increasing academic evidence that such gains are achievable. Moreover, if human or other bottlenecks are the binding constraint on AI productivity gains, then firms will gradually reorganize to unlock them, just as they have done in response to prior technological shocks.

What the current numbers do not yet support, however, is the claim that this transmission is already well underway at the aggregate level: adoption is too thin, the cross-sectional signal is too weak, downstream bottlenecks remain, and the productivity pickup is too attributable to utilization to be confident in that conclusion. Instead, the acceleration we’re seeing so far is largely a result of companies trying to meet the demand for AI capacity by pushing the limits of their existing infrastructure.

UK statistics show business creation flat or even declining over the past 2 years. Is the UK really falling behind or is something else afoot? In this appendix to our last post, we triangulate across Stripe data and public statistics, and conclude that the UK is likely seeing its own business formation boom that government statistics are simply failing to capture.

Subscribe now

The UK lacks a clear solopreneur registration pathway

The US and France have explicit statistics related to solo or microbusinesses. The Census Bureau’s business application statistics are based on EIN filings, which, while not technically necessary to start a sole proprietorship in the US, are highly common in practice. INSEE’s data on création d’entreprises leverages the microentrepreneur regime (focused specifically on solopreneurs). The UK, however, has no such solo pathway. Companies House statistics don’t include sole proprietors or the self-employed. The Office for National Statistics (ONS) similarly only counts businesses registered for VAT or PAYE (tax withholding)—a threshold that many solopreneurs fail to meet.

The Department for Business and Trade (DBT) publishes its own business population estimate, which tries to “back out” the number of unregistered businesses (primarily self-employed and sole-traders) using the Labour Force Survey (LFS). The LFS counts people who identify as self-employed, but this likely struggles to account for individuals whose solo enterprises are a nascent or supplemental source of income—for example, a person who has a Substack or a Shopify store on the side that they monetize while still maintaining a salary job. The LFS has also been plagued by multiple issues, losing roughly half its effective sample since COVID-19 (from approximately 70,000 individuals surveyed in 2019 to fewer than 35,000 in 2023), and with worse attrition among young people who are more likely to be self-employed. The LFS publication was suspended entirely in 2023 and was relaunched in 2024, downgraded to “statistics in development” due to these challenges; the DBT statistics were then also downgraded.

Tax receipts are broadly consistent with an increase in solopreneur activity

An imperfect proxy for solopreneurship in the UK is self-assessment income tax receipts (i.e., taxes that are paid by individuals rather than directly by employers). Self-assessment receipts increased by over 70% relative to the pre-COVID period and showed a reacceleration of 12% in the last available year of data. This is consistent with growth in solopreneur activity, if not dispositive.

There are several caveats: solopreneurs do not need to register with HMRC if they make less than £1,000, which means the data excludes a large swath of the solo population who are either early in their journey or experimenting. Individuals also file self-assessment taxes for many reasons besides self-employment (e.g., declaring property income or investment income), with changes in dividend allowances in 2023–2024 artificially increasing receipts from this category.

Finally, UK tax law until recently required earners above £100,000 to file self-assessment even if their employer paid their taxes. However, this last point—while meaningful for self-assessment filings—should not materially impact receipts numbers since the vast majority of tax for this group is captured at the source (via PAYE) rather than via the self-assessment tax return. Thankfully, HMRC tax information and impact notes (TIINs) provide data on costings for each of these recent policy changes, so we can disentangle these confounding variables relatively cleanly.

Stripe data is consistent with an acceleration in UK business creation

Stripe’s UK business sign-ups have been growing rapidly in the last few years, tracking the surge seen in the US and France almost exactly. In the last two years, UK business sign-ups have doubled.

How do these trends compare to Companies House incorporations? During the early post-pandemic years Stripe signups tracked incorporations closely; since 2024 they have diverged sharply.

While this Stripe data is a strong signal of growing underlying business formation in the UK, it is possible that Stripe could simply be increasing its market share of existing UK businesses, or capturing a greater share of new UK businesses that are forming. Our analyses suggest neither of these is first order. While migrations onto Stripe in the UK are indeed growing, migrations as a share of Stripe sign-ups are declining. Put otherwise, the surge in sign-ups predominately represents net-new business activity. In addition, the surge in signups is so large that, absent an underlying surge in business formation, the probability of the implied share-gain over such a short time frame is low.

Conclusions

1. The UK is likely experiencing an increase in solopreneurship broadly in line with what we see in the US, France, and many other markets.

2. Solopreneurship in the UK is poorly captured by official statistics. If, as early analysis suggests, a surge in solo-run businesses turns out to be a major feature of the economic transition driven by AI, the UK’s inability to track it creates a major policy and economic blindspot.

By the early 2020s, these assumptions started creaking. The Census Bureau noticed that solo operators in some sectors were generating substantial revenue, and they didn’t seem to be hiring. And so in 2022, the Census Bureau systematically raised its income thresholds. The resulting counts of nonemployer businesses at higher income levels skyrocketed.

Today, Census Bureau Business Formation Statistics, cross-country business registration records, and Stripe platform data all suggest solopreneurs have continued to prosper.

This post advances three related arguments. First, solopreneurship is growing faster than employer-business formation, and the acceleration is validated by multiple independent data sources, making it unlikely to be driven by a fraud wave. Second, both the number and share of solopreneurs reaching meaningful income thresholds is rising. And third, early signals indicate that AI is filling the capability gaps that once made hiring necessary—and doing so fast enough to show up in the income distribution.

Solopreneurs are multiplying much faster than employer businesses, and it’s not fraud

New business applications in the Census Bureau’s Business Formation Statistics have been flashing an unusual signal for the past 18 months. New business applications are filings with state agencies and the Internal Revenue Service that precede actual commercial activity. These new applications reaccelerated beginning in late 2024, after spiking in 2020 at the beginning of the pandemic and remaining elevated thereafter.

This acceleration, however, is not being driven by applications from businesses that have a “high propensity” or likelihood to become employers. The Census Bureau classifies certain new business applications as “high propensity” if it deems the business as likely to hire workers within the next eight quarters. The Census Bureau uses an internal statistical model that considers industry classification, whether an Employer Identification Number (EIN) was requested, and whether the applicant indicated planned wages. High-propensity filings have stayed relatively stable over the past 18 months even as overall applications have surged.

A similar spike in new business applications occurred early in the pandemic. The Paycheck Protection Program (PPP) was a small business loan program that the US federal government established in 2020. It required an EIN and little else for eligibility, creating strong incentives for applications with no genuine entrepreneurial intent. Economic research has found a contemporaneous increase in the composition of 2020 applications toward businesses likely to be nonemployers (Dinlersoz et al. 2021). The nonemployer surge peaked and partially reversed as the PPP program wound down—though interestingly, the level of business formations in the US stayed persistently higher than prepandemic.

For the current acceleration in new business applications, there is no comparable federal subsidy creating a financial incentive to file. The PPP stopped accepting new applications in May 2021, and the last forgiveness payout was in 2024.

Three further pieces of evidence argue against fraud specifically. The first is that Stripe pay-in data is also consistent with the read that recent business sign-ups are legitimate. Stripe cannot distinguish between solopreneurs and employers among business sign-ups, but we do validate the overall acceleration and the meaningful activity of these businesses. Businesses that signed up on Stripe after 2023 reached material transaction volumes earlier than the sign-up cohorts that preceded them. The share of businesses (not just solopreneurs) reaching $1 million in cumulative revenue within a year after going live on Stripe was roughly 30% higher for the 2025 cohort as it was for the 2023 cohort, and it was roughly 3x higher for the 2025 cohort than the 2019 cohort. If more recent sign-up cohorts were increasingly driven by inactive businesses, we would expect more time to material volume thresholds, not less.

Second, as we have shown previously, the story of spiking business applications is not confined to the United States. New business registrations have risen roughly 40% in Australia, 70% in Finland, and 80% in France since 2017, with meaningful acceleration in 2025 alone. Multicountry acceleration across different regulatory environments argues for a more fundamental driver than fraud activity. And in France, where more granular data is available, the surge in business formation is driven primarily by solo founders or microentrepreneurs rather than by traditional employer businesses—similar to the US.

A final reason to be skeptical of fraud in the current acceleration is that we have also seen growth in incorporations in the US state of Delaware. Delaware incorporations have grown approximately 40% year over year since early 2025, and they have remained at or above the pandemic-era peak every month since January 2025. While Delaware represents only about 75,000 of the roughly 5.7 million business applications filed in the United States in 2025, it is the incorporation jurisdiction of choice for founders who intend to raise institutional capital or establish formal governance structures, making it a strong signal of genuine business intent. It is not the natural destination for a passive LLC filing by bad actors.

Business applications are also growing fastest in Wyoming, another jurisdiction associated with deliberate legal structuring rather than passive registration. The geographic composition of the acceleration tilts toward intent.

Solopreneurs at higher income thresholds are seeing rapid growth

In 2023, roughly four million Americans earned their primary income as solopreneurs, generating over $100,000 in annual revenue. That figure has risen substantially from the mid two-million range of the early 2010s, a period when the infrastructure enabling solo business at scale—Stripe, Substack, Kajabi, and other platforms—was still small in scale and more limited in capabilities. On its own, this doubling is already impressive. But what about solopreneurs at even higher income thresholds? Here, the Census Bureau’s data is unfortunately more limited. Methodological changes over the years make longitudinal studies of Nonemployer Statistics (NES) data challenging.

To understand what is happening with higher-income solopreneurs, we constructed a proxy index of solopreneurs on Stripe, covering approximately 115 solopreneur-focused platforms and all solo Stripe Atlas businesses. This index almost certainly understates the actual Stripe solopreneur population, since most solo operators use general-purpose infrastructure rather than solopreneur-specific platforms. But it tracks directional trends with reasonable fidelity, and it shows a consistent upward trajectory in both the count and share of solopreneurs earning above various income thresholds.

We find that there has been a substantial increase in the number of solopreneurs earning over $100,000 in our index, but an even larger increase in the number earning at higher income thresholds, with a clear acceleration since 2023. More than twice as many solopreneurs earned over $1 million in 2025 than in 2023, and close to three times as many crossed $5 million and $10 million.

Perhaps even more interestingly, the share of solopreneurs earning above these income thresholds has also doubled in the last two years, suggesting that—rather than the surge in business applications reflecting low-quality experimentation with a few lucky standouts— the cohorts of new solopreneur businesses might actually be of higher quality than in the past.

This is a proxy. There might be nonsolo businesses using these solopreneur-focused platforms. Businesses that started as solopreneurs might have since added employees. And even for those that are in fact solo, the earnings distribution within the solopreneur category is almost certainly wide. Still, these trends are quite striking and suggest a true shift in the scale solopreneurs can reach on their own and the frequency with which they do so.

AI is filling the capability gaps that once made hiring necessary

It seems likely that AI is one of the primary drivers of both the acceleration in solo business formation and the outsized performance of solopreneurs in recent years. However, untangling this effect is not always easy.

Part of the story is ease of discovery of, access to, and integration of new tools. An agent can now help you find the best tools for your business and handle your integration with minimal support. We see this effect in both Census Bureau data and our sign-up data via what we call AI-assisted sign-ups.

Census Bureau data allows us to dig further into nonemployer business applications, segmenting the data by broad sector. The following chart compares the percent growth in this nonemployer proxy since the first quarter of 2025—when it started accelerating—to AI adoption in the Census Bureau’s Business Trends and Outlook Survey (BTOS), which asks whether the business used AI in any of its business functions over the prior two weeks. The recent growth in nonemployer businesses shows a positive relationship with industry-level AI adoption, suggesting that higher AI adoption has been broadly consistent with more growth in nonemployer applications, though with notable outliers (manufacturing on the downside, transportation and warehousing on the upside).

AI-influenced user journeys now constitute nearly 4x the share of Stripe sign-ups as last January. We measure two types of AI assistance among the businesses using Stripe today: direct and passive. Direct behavioral signals indicate use of AI tools to build or manage Stripe integrations. These are things like using our Model Context Protocol (MCP), command-line interface (CLI), or Claimable Sandboxes—a Stripe feature launched in October 2025 that lets prospective businesses interact with a prebuilt demo environment before formally signing up. Passive signals indicate association with AI tools but not necessarily AI-assisted building. These are things like AI referrals, or traffic arriving from large language models like ChatGPT that recommend Stripe to businesses researching how to accept payments.

These are substantial numbers; however, they likely understate the importance of AI in the solopreneur surge. Part of the reason businesses historically tended to be built by groups was that a single individual rarely possesses all the skills needed in the entrepreneurial journey. Whether it’s how to evaluate or size a market, code an app, price a product, write and execute a marketing campaign, or close a deal, AI (and AI-augmented software) can fill many of the gaps that founders previously turned to another human for. Or, as Sam Altman so succinctly put it: “revenge of the idea guys.”

We think this phenomenon is the true engine of the AI surge in business formation we’re seeing today. The availability of this breadth of on-tap assistance allows anyone with sufficient motivation to go it alone. Given this, we think the 20% figure is a floor rather than a ceiling in terms of AI impact.

Conclusion

While the recent surge in US business formation is not reflected in high-propensity applications, the evidence points toward a structural increase in genuine small business activity over 2025 and 2026, driven by solopreneurs. Preliminary evidence of growth in AI tool usage and solopreneur growth in high-AI-adoption sectors suggests that advances in AI are responsible for a meaningful portion of this growth. We believe AI is lowering barriers to business formation and growth by expanding the capabilities of solopreneurs, further improving tools and platforms that cater to new businesses, and creating a new set of opportunities for entrepreneurs to pursue. Whether the magnitude of this effect is as large as the most optimistic readings of the data suggest remains to be seen, but we believe we might be in the early innings of a fundamental acceleration in business formation—which could have ripple effects throughout the economy.

Subscribe now

The rise of solopreneurs

New business applications are rising worldwide. In the United States, where the Census Bureau tracks filings with enough granularity to distinguish likely employer businesses from other types, a notable divergence has emerged: total applications have accelerated significantly, but “high-propensity” applications—those statistically likely to result in payroll employment—have not kept pace.

Similar wedges have opened between overall and high-propensity filings in the past, such as in the depths of the pandemic and during the implementation of the Paycheck Protection Program (PPP) in the US. The current episode appears to reflect something different. The rapid development of AI tools might be lowering both the barriers to starting a business and the returns to doing so for individual operators—contributing to a rise in AI solopreneurs: individuals using AI to establish sole proprietorships or side businesses without the need or intent to hire. We see some evidence of this in our own Stripe Atlas data, where startups have accelerated since 2023, and especially in the first quarter of 2026. The overwhelming driver of this acceleration comes from solo founders, and we see growth among both AI and non-AI startups (but AI might be enabling easier formations even for firms whose business models are not AI-based).

What the SaaSpocalypse looked like on Stripe

Over 30 days in early 2026, the software sector shed roughly $1 trillion in market capitalization. The sell-off was driven primarily by investor concerns about the long-run revenue prospects of traditional software companies, as agentic AI systems—capable of performing many of the functions that enterprise software has historically been purchased to support—advanced rapidly through 2025.

Payment volumes offer a different vantage point. Unlike equity prices, which reflect market pricing of future scenarios, pay-in volumes capture current economic activity. Looking at weekly transactions for the 100 largest non-AI software-as-a-service (SaaS) companies on Stripe, the episode appears as a brief dip followed by a swift recovery and continued growth—a pattern quite different from what equity markets were pricing. The divergence between current transaction activity and market valuations punctuates how much the SaaSpocalypse was driven by expectations and forecasts, not current ongoing activity.

The Solow paradox and AI

In 1987, economist Robert Solow famously commented that the gains from personal computers were “everywhere except the productivity statistics.” The quip proved premature rather than wrong: most economists date the visible acceleration in aggregate productivity to the mid-1990s—less than a decade after Solow’s observation. (For more reading on how economists thought about the productivity effects of personal computers in real time, before the effects were obvious in the dot-com boom, Paul David—a Stanford economic historian—has a seminal 1990 paper, The Dynamo and the Computer, that’s a classic in this genre.)

The Solow paradox—the lag between a technology’s arrival and its measurable effect on productivity—is, it turns out, historically typical. The electrification of the US economy offers a tangible parallel. For three decades after Thomas Edison began delivering commercial electricity to New York customers in 1882, inflation-adjusted output per worker grew at just 0.5% annually—a sluggish pace relative to what electricity would eventually enable. The gains became evident only as the economy’s capital stock and organizational practices adapted to the new technology: new factories designed around electric motors from the ground up, rather than retrofitted from steam. Some of the impetus for this replatforming came from the weapons production spurred by World War I. In the decade after 1917, productivity growth more than doubled its prior three-decade average.

The implication for AI is that a near-term absence of productivity acceleration in the aggregate data should not be read as evidence that no acceleration is coming. (Stripe Economics has a recent post on how the decline of travel agents might offer some lessons on how technological shocks work their way through the labor market.)

John’s talk included many more nuggets. Watch it in full below.

Some occupations, like radiologists, have proven more resistant to technological displacement than expected. Others, like bank tellers, did eventually succumb—just to later and different shocks than most predicted.

Sometimes though, displacement happens in straightforward ways, as with travel agents. Airline deregulation in the late 1970s both complicated booking prices and loosened rules on travel agent commissions. Together, these changes made the profession far more lucrative and boosted employment: between 1980 and 2000, travel agent jobs tripled, peaking at about 340,000. But then a prolonged structural decline set in, driven by a one-two punch: airlines began cutting agent commissions on ticket sales in the early 1990s; and internet platforms like Expedia, Travelocity, and Priceline emerged, advertising that everyday consumers could now access the same reservation systems agents once monopolized. The loss of clients to online providers and reduced commissions led many brick-and-mortar agencies to downsize or shut down entirely.

So the decline of travel agents was driven largely by technological displacement. AI is a very different kind of shock from the internet of the 1990s, and its effects on labor markets may turn out to be quite different too. But the story of the travel agent is still instructive, because it demonstrates what happens to workers when a technology renders their profession obsolete. As it turns out, the story is not entirely bleak.

1. Travel agent layoffs came in recessionary bursts, not a gradual decline.

Widespread displacement of travel agents didn’t happen immediately during the dot-com boom. Rather, it was the bust that drove displacement. In the wake of the 2001 recession, employment in the travel agency industry1 fell precipitously and did not claw back prior losses. By the eve of the Great Recession in 2007, there were already nearly 40% fewer workers in travel agencies than at their dot-com peak. Four years later, employment was at less than half of its peak, and today travel agency jobs are 60% lower.

2. Travel agent unemployment rates are still low in good times, but are now higher in bad.

While travel agent employment has declined steadily, so has the travel agent labor force, indicating that many agents have changed careers since 2000. As a result, the unemployment rate of travel agents—measuring the utilization of their labor force—was just 2.8% in 2025, lower than typical even before the dot-com boom.

But this low rate comes with a major asterisk. Travel agent unemployment has become far more sensitive to economic downturns. Before 2000, travel agent unemployment rates typically rose by less than the overall rate during downturns. Between 1979 and 1982, for example, the travel agent unemployment rate rose only 0.8 percentage points despite the overall rate jumping nearly 4 percentage points. The pattern repeated over 1989–1992, with travel agent unemployment rising only 0.9 percentage points versus 2.2 overall. But after 2000 the pattern flipped: the travel agent jobless rate spiked 3.9 percentage points over 2000–2003 versus 2 overall, 5.8 versus 5 over 2007–2010, and 0.7 versus 0 over 2019–2022.2

3. Pay for travel agents has outpaced overall wage growth as they have shifted to higher-value bookings for wealthier clientele.

As the market migrated online in the late 1990s and early 2000s, the remaining human travel agents pivoted sharply upmarket. They adapted by charging planning and membership fees and aligning with luxury consortia like Virtuoso to offer high-end clients exclusive upgrades, complimentary amenities, and personalized services that no algorithm could replicate. This shift is reflected in wages. Average weekly earnings at travel agencies were 87% of overall average weekly earnings back in the heyday of 2000. By 2025, the ratio had reached 99%, meaning travel agency wages had outpaced the rest of the private sector over that span.

4. Economy-wide employment has fully recovered

Travel agent employment is down by more than half from its dot-com peak. This caused hardship for agents forced to switch careers and facing greater risk of joblessness. Decades later, however, the dot-com boom has not lowered aggregate employment. While the employment-population ratio has fallen five percentage points since 2000, this is entirely because of population aging. Adjusted for today’s demographics, the employment rates of 2000 look virtually identical to the present. As travel agent jobs disappeared, new jobs emerged elsewhere, many former agents switched careers, and young workers who might once have become travel agents went into other fields instead.

History doesn’t repeat itself, but it does rhyme. AI’s effects on the labor market may not be identical to past technological shocks, including those on travel agents. But the experience of travel agents highlights risks for us to consider. On the one hand, AI disruptions to the labor market may be delayed and not salient until (perhaps even amplified by) the next recession. On the other hand, the most negative effects may be limited to specific occupations rather than broad-based, and as prices and wages adjust, those occupations may even look more favorable over time. But that’s of little comfort to anyone displaced in the near-term. These risks suggest the policy focus should be more weighted towards easing transitions for whatever displacements do take place rather than assuming mass unemployment.

Subscribe now

The K’s we see: Profits and (possibly) wealth

US corporate profits in 2025 were elevated, with profit margins reaching 17.9% over the year ending September 2025, a level not seen pre-pandemic since 1967.1 But more striking than the average level of profitability is its distribution.

Market concentration is often dismissed as a valuation bubble, but in this case, the divergence between the best and worst performing stocks is underpinned by a massive consolidation in actual earnings. The most profitable third of publicly listed US stocks now account for roughly two-thirds of total market capitalization—the highest share on record, based on data from Eugene Fama and Kenneth French. And the most valuable 10% of the S&P 500 account for approximately 59% of the index’s total profits, well above historical norms.

There are two effects here: the most profitable companies increasing their share of equities, and large businesses getting more profitable and entering the top tercile. The steep jump we saw in 2025 was driven entirely by the latter. That included some Magnificent 7 businesses (e.g., Meta, Amazon, Alphabet), but also many large, non-Magnificent 7 companies such as Netflix, Home Depot, and JPMorgan Chase. That broadening of the profit elite within large-cap equities is a meaningful signal: it suggests the dynamic is structural, not a function of a handful of hyperscale platforms.

The mechanism is consistent with superstar firm economics. Returns to scale in software, data infrastructure, and digital distribution create winner-take-most outcomes across a widening range of industries. The equity market is, in this sense, functioning as the sorting machine it is designed to be—concentrating capital and profits toward the highest-return deployments.

The equity divergence has a plausible transmission mechanism to household wealth. The S&P 500 rose 16.5% over 2025, and the top 1% of US households own approximately 40% of all equities.2 And on first glance, that mechanism seems to be firing: the share of total wealth of the top 1% has risen roughly 2 percentage points since 2022. That signal warrants some caution, however, as the rise in the top share began growing at exactly the point this data became less anchored to more-reliable wealth surveys.3 The direction of this recent trend is plausible and consistent with equity market performance, but the magnitude should be treated as provisional until more data lands.

The K’s we don’t see (yet): Consumer spending and wages

The consumer flavor of the K-shaped hypothesis might “feel” right to families because of lingering dissatisfaction with prices and affordability, which we can see in metrics like consumer sentiment. That said, while consumer spending by income group is not tracked well in real time by any single government source, the available data does not support a bifurcation narrative.

Multiple spending surveys show low-income household spending performing at least on par with high-income households over 2025. Households making under $50,000 reported spending growth of 5% over 2025, versus 4.6% for households making over $100,000. While inflation-adjusted retail spending for lower-income households grew modestly slower than it did for top-income households in 2023 and 2024, the gap converged through 2025—low- and upper-income consumption grew at similar rates. This is consistent with lagged but larger and more-reliable government data, which shows the top income decile’s share of aggregate spending at 22.8% in 2024, down from 23.4% in 2023 and below the recent peak of 23.9% in 2016 (economist Antoine Levy has a helpful deep dive on the data here).

Using Stripe data, we can also crudely proxy for consumer spending by income quartile, based on consumer MCC transactions and average income in the ZIP code of the spender. This data similarly shows no evidence of a high-income divergence: if high-income ZIP codes were seeing faster spending growth than low-income ZIPs, the ratio of the two would be rising. But in fact, the ratio has been falling since 2023. So at least on Stripe, lower-income ZIP codes have seen more robust spending growth than the top.

The picture on wages and income—which should be strongly correlated with consumption and which are more reliably measured on a household basis—is similarly undramatic. While real earnings at the 10th wage percentile have grown roughly 0.5 percentage points more slowly than at the 90th percentile since 2022, both cohorts posted positive real wage growth over the period: the bottom did not fall while the top rose. On income more broadly, lower-income households actually outperformed the middle and top from 2022 to 2024, before briefly softening in late 2024 and early 2025 and then reaccelerating. The middle income tier has grown faster than the top in 2025 and on a cumulative basis since 2022.

In short, on current data, the consumer economy looks more evenly distributed than equity market performance alone would suggest.

Why the corporate K but not the consumer K?

Several forces might explain why divergence in corporate profits has not yet translated into divergence in consumer-level consumption patterns. First, although the richest 20% hold a disproportionate share of equities, they account for less than a quarter of all consumer spending. So any equity wealth effect, even if large in percentage terms for the wealthy, is modest in aggregate demand terms. On average, only 3¢ of every $1 earned in stock market wealth is actually spent. The $6.3 trillion valuation-driven growth in equity wealth over the first 9 months of 2025 would thus be expected to support about $200 billion in consumer spending. And higher concentration of equity wealth at the less-spending-prone top can dampen this effect. The very wealthiest spend more like 1¢ for every dollar they make in the stock markets. In other words, the wealthy are getting richer, but they don’t spend enough of that extra wealth to tip the scales of the entire consumer economy. This explanation implies that the run-up in equity wealth would have to be even steeper than what we’ve seen to drive divergence in consumer spending.

Second, real wages at the bottom have been supported by continued labor market tightness and sustained gains from earlier in the pandemic, partially offsetting capital income gains accruing to the top. As the chart below shows, the strongest payroll gains since the end of 2022 have been in the lowest fifth of industries by wage level. This possibility suggests that labor market weakening would be the harbinger of K-shaped patterns in the consumer data.

We are seeing economic divergence, but primarily in business America—in profit margins, market concentration, and equity returns—where the story appears to be largely one of tech versus nontech, rather than the consumer economy, where evidence is much weaker so far.

For investors, the profit concentration story appears durable so long as returns to scale in technology and digital infrastructure persist. The broadening of the profit elite beyond the Magnificent 7 into a wider cohort of scalable businesses supports equity earnings quality over the medium term.

The consumer K meanwhile remains an open question rather than a settled finding. Wealth concentration has risen on the best available real-time data, and the transmission mechanism—equity wealth effect into upper-income spending—is theoretically coherent, but it has yet to decisively show up in what distributional spending data we have. A consumer K could become more evident as wealth spillovers become more salient, or as other shocks such as geopolitical and trade tensions or slowing labor market churn has disparate effects on households. Until then, consumer bifurcation is a hypothesis worth monitoring, not a conclusion to act on.

The K, in other words, is more Wall Street than Main Street—for now.