America’s New Firm Formation Puzzle: AI, Entry Costs, and the Return of Business Dynamism

America’s New Firm Formation Puzzle: AI, Entry Costs, and the Return of Business Dynamism

The most important economic chart in the current AI debate may not be a benchmark score, a semiconductor revenue curve, or a hyperscaler capital expenditure table. It may be the quiet but dramatic rise in United States business applications. For roughly a decade and a half before the pandemic, monthly applications moved in a familiar range, often around 45,000 to 55,000 on a seasonally adjusted four-week moving average basis. Today the series is above 120,000, more than twice the old normal. The initial break in the chart was clearly pandemic related, but the persistence of elevated formation is harder to dismiss as a one-off reopening artifact. It now overlaps with the mass adoption of generative AI and large language models.

That overlap matters because entrepreneurship is ultimately a calculation about expected return on effort. A person starts a firm when the possible upside exceeds the opportunity cost, the capital requirement, the coordination burden, and the psychological cost of uncertainty. AI changes that calculation by lowering the fixed costs of launching and operating a business. It does not simply make existing firms more efficient. It makes the first mile of entrepreneurship cheaper. It compresses the distance between idea and prototype, prototype and customer, customer and repeatable process. If this persists, the United States may be entering a structural entrepreneurship boom rather than merely enjoying a post-pandemic statistical aftershock.

The core thesis is straightforward: the rise in business applications is a signal that the supply curve of entrepreneurship has shifted outward. The pandemic may have triggered the first move, but AI can help explain why the new level has not fully mean-reverted. Large language models reduce the need for large starting teams, automate pieces of coding, marketing, customer service, research, legal drafting, documentation, analytics, and operations, and allow individuals or very small teams to perform tasks that previously required specialized labor. That raises the expected return to entrepreneurial effort and lowers barriers to entry. The result can be more firms, more experimentation, faster productivity diffusion, and greater resilience. The effect resembles the internet revolution, but with a different microeconomic channel: the internet lowered search, distribution, and communication costs; AI lowers cognitive, coordination, and production costs.

The Chart Looks Like A Regime Shift, Not A Normal Cycle

A cyclical interpretation of the data is tempting. Recessions, lockdowns, stimulus payments, remote work, and labor-market dislocation all pushed people to reconsider careers and start side businesses. Many of the first pandemic-era applications were likely necessity entrepreneurship, e-commerce experiments, local services, independent consulting, and tax or registration behavior rather than fully fledged high-growth startups. The 2020 spike was unusually sharp, so a skeptical analyst should ask whether the series simply moved from an old equilibrium to a temporary pandemic plateau.

Yet the persistence is the point. A temporary surge should normally decay as labor markets normalize, offices reopen, savings buffers decline, and household risk appetite cools. Instead, the business-application series remains far above its pre-2020 baseline. The composition also matters. The United States Census Bureau separates high-propensity business applications from broader applications; while not every filing becomes an employer firm, a sustained elevation in filings means the pipeline of possible firms is materially wider. Even if the conversion rate from application to operating business falls, the absolute number of potential entrants can still rise meaningfully.

This is where the AI interpretation becomes plausible. Structural breaks in firm formation are rare because starting a company is not only a financial decision. It is an institutional decision. It requires access to tools, customers, knowledge, accounting, compliance, distribution, and labor. The internet changed several of these constraints. Cloud computing changed them again by turning large up-front technology expenditure into variable cost. AI now attacks another set of constraints: the need to buy specialized cognitive labor before the business has proven demand. A founder who can draft contracts, build a landing page, generate code, analyze competitors, produce marketing copy, answer customer inquiries, and summarize regulation with machine assistance faces a lower entry hurdle than the same founder did five years ago.

There is an important distinction between a boom in ideas and a boom in firms. Ideas are abundant; operational execution is scarce. AI does not eliminate that scarcity. It does, however, reduce the penalty for trying. If the cost of an experiment falls from tens of thousands of dollars to a few hundred dollars and a month of focused work, the number of experiments should rise. Economic history suggests that productivity waves often begin as waves of experimentation before they become waves of measured output. The early internet did not immediately create stable profits for everyone who registered a domain. It created an environment in which the cost of testing business models collapsed. Many failed. A few became platforms. A larger number became normal firms using the new infrastructure.

AI Changes The Startup Cost Function

The cleanest way to understand the entrepreneurship boom is to think in terms of the startup cost function. A simple new firm has fixed costs, variable costs, and coordination costs. Fixed costs include incorporation, research, product design, software development, branding, basic legal work, accounting setup, and initial sales material. Variable costs include customer acquisition, hosting, inventory, transaction processing, and support. Coordination costs include the time needed to hire, manage, communicate, and align people with different skills.

Generative AI directly lowers fixed and coordination costs. A founder can use AI to write a first version of a business plan, produce a financial model, draft customer emails, design a product specification, build code scaffolding, create customer-support scripts, translate content, summarize market research, and compare legal templates. None of these outputs should be treated as perfect. But perfection is not required at the entry stage. What matters is that the founder can reach a minimum viable version faster and cheaper.

A useful formula is:

Expected entrepreneurial surplus = p(success) x payoff - fixed cost - opportunity cost - coordination cost - financing friction.

AI can affect every term. It may raise the probability of success by giving founders better research and faster iteration. It can raise the payoff by letting a small team serve a larger market. It lowers fixed cost by automating draft work and technical scaffolding. It lowers opportunity cost by allowing people to test ideas part-time before leaving employment. It lowers coordination cost by substituting software agents for some early hires. It lowers financing friction because a founder who needs less starting capital can self-fund longer and avoid early dilution.

This framework also explains why the effect can be broad rather than limited to venture-backed software. A restaurant supplier can use AI for quoting, inventory analytics, customer outreach, and bookkeeping. A real-estate services firm can use AI to generate market reports and automate lead qualification. A solo consultant can turn expertise into repeatable products. A medical billing startup can automate documentation. A small manufacturer can use AI for procurement research and sales collateral. The common denominator is not that every business becomes an AI company. It is that every business receives a cheaper operating layer.

Ronald Coase argued that firms exist partly because market transactions are costly. When it is expensive to find, contract with, and coordinate independent specialists, activities move inside the firm. AI complicates this boundary. If software can perform some specialist tasks on demand, the minimum efficient size of a firm can fall. The same entrepreneur can coordinate a smaller human team with a larger digital toolkit. That reduces the scale required to enter a market. In Coasean terms, AI lowers some transaction costs and some internal organization costs at the same time, changing the make-or-buy boundary for new firms.

The same logic applies to managerial span of control. In classic models of entrepreneurship and firm size, including work associated with Lucas and later heterogeneous-firm models, managerial talent is scarce and firms differ because founders vary in productivity and ability to coordinate resources. AI can augment managerial capacity by reducing the amount of time spent on routine communication, documentation, analysis, and monitoring. It may not turn a weak founder into a strong one, but it can expand the operating range of a capable founder. That matters for the long tail of small businesses as much as for high-growth startups.

The Decline In Barriers To Entry Is A Productivity Story

Business formation is not automatically productive. A new firm can be a lifestyle business, a duplicate service provider, or a short-lived experiment. Many entrants fail. But an economy with too little entry becomes stagnant. Research by John Haltiwanger and coauthors has shown the importance of young firms in job creation and the concern that U.S. business dynamism had weakened over prior decades. Decker, Haltiwanger, Jarmin, and Miranda documented a decline in dynamism and a falling startup rate before the pandemic. That decline mattered because productivity growth depends not only on incumbent efficiency but also on reallocation: resources must move from less productive firms to more productive firms.

AI-enabled entry can help repair that channel. The first-order effect is not that every new business becomes a productivity champion. The first-order effect is that more experiments enter the selection process. In models such as Jovanovic’s theory of firm learning and Hopenhayn’s industry dynamics, firms discover their productivity after entry, and market selection reallocates resources toward better firms. Lowering entry costs increases the number of draws from the productivity distribution. If selection remains healthy, the economy benefits from more chances to discover high-productivity firms.

There is a Schumpeterian interpretation as well. Innovation comes through creative destruction: new combinations challenge old routines. AI gives entrants a new production technology and a new organizational technology at the same time. A small firm can challenge an incumbent not because it owns more capital, but because it can move faster, personalize more cheaply, and operate with fewer layers. A large incumbent still has data, distribution, capital, brand, compliance infrastructure, and customer relationships. But the entrant’s disadvantage narrows when the cost of building a credible product, reaching customers, and servicing demand falls.

The productivity channel can be described in three stages. First, AI increases the number of entrepreneurs who can attempt a business. Second, it increases the speed of experimentation by reducing the time between hypothesis and market feedback. Third, it increases the quality of surviving firms because low-cost experimentation allows founders to pivot before capital is exhausted. If the economy can provide financing, infrastructure, and competitive markets, the surviving cohort should be more productive than a world in which only heavily funded teams can experiment.

This matters because the United States has been searching for a productivity impulse. After the late-1990s acceleration, productivity growth slowed. The internet generated enormous consumer surplus and platform value, but aggregate productivity statistics often lagged the visible technological excitement. Robert Solow’s famous observation that computers appeared everywhere except in the productivity statistics remains relevant for AI. General-purpose technologies require complementary investments: organizational redesign, training, new business models, and legal adaptation. Business formation is one mechanism through which those complements emerge. New firms are not locked into old workflows, so they may adopt AI-native processes faster than incumbents.

Brynjolfsson and Hitt’s research on information technology emphasized that the gains from IT are largest when paired with organizational change. AI may follow the same pattern. A bank, insurer, law firm, or hospital can buy AI tools and still struggle to redesign incentives and processes. A new firm can build the workflow around AI from day one. That difference is why elevated formation could matter for measured productivity over time. The startup is not simply using AI as a plug-in; it can be organized around AI as an operating principle.

Small Teams Can Now Attack Problems That Once Required Large Payrolls

The most striking implication is the changing relationship between headcount and ambition. Historically, many businesses required a minimum set of specialists: engineer, designer, marketer, customer-support representative, analyst, operations manager, and legal or finance support. Even if each role was part-time, the coordination burden was real. A founder without funding could not easily assemble that stack. AI does not fully replace these people, but it can approximate the first draft of their work and allow a smaller team to decide where human expertise is truly necessary.

This changes the economics of bootstrapping. A bootstrapped founder can now stretch capital by using AI for the rough work and hiring humans for judgment, domain expertise, trust-sensitive tasks, and final quality control. The bottleneck moves from “Can I afford the first team?” to “Can I define the problem well enough, reach customers, and maintain quality?” That is still difficult, but it is a higher-quality difficulty. It rewards clarity, taste, and customer knowledge rather than only access to capital.

There is also a geographic implication. If AI reduces the need to hire a full local team, entrepreneurship can diffuse beyond traditional startup hubs. Remote work already loosened the geographic constraint. AI can loosen the talent-density constraint. A founder in a smaller city can access software, design, research, and marketing leverage that once required proximity to a deep labor market. This does not mean geography disappears. Networks, capital, universities, and customers still cluster. But the marginal founder outside a hub is less disadvantaged than before.

For the labor market, the implications are ambiguous but powerful. AI-enabled startups may hire fewer people per dollar of revenue, which can reduce job creation per firm. At the same time, more firms can be created, and the surviving firms may scale faster. The aggregate employment effect depends on the balance between lower labor intensity and higher firm count. This is one reason the business-application data are important but incomplete. We need to track conversion to employer firms, payroll creation, revenue growth, survival, and productivity. A million AI-enabled filings that remain side projects would mean something different from a broad cohort of employer firms that survive and scale.

The same ambiguity applies to wages. If AI makes skilled individuals more productive, it can raise returns to entrepreneurial and technical talent. If it automates routine white-collar tasks, it can pressure some roles. If it creates many small firms, it can increase demand for specialized human judgment, sales, compliance, design, and relationship management. The likely outcome is dispersion. Workers who combine domain expertise with AI leverage may see rising productivity and bargaining power. Workers whose tasks are easily standardized may face more competition from both software and AI-enabled entrants.

The Boom Could Strengthen Economic Resilience

A higher rate of business formation can make an economy more resilient if the new firms diversify sources of income, employment, products, and local services. A concentrated economy dominated by a few incumbents can be efficient in stable periods but fragile when shocks hit. More entrants create redundancy. They test alternative supply chains, serve niche customers, and create local options. In a world of geopolitical risk, supply-chain disruption, cyber risk, and climate shocks, distributed entrepreneurial capacity has macro value.

This resilience point is often underappreciated because financial markets prefer scale and profitability. Public-market investors naturally focus on dominant platforms, margins, and returns on invested capital. But from a macro perspective, an economy also benefits from option value. Each new business is a small call option on a new product, process, or local service model. Most expire worthless. Some become modestly useful. A few become extremely valuable. AI lowers the premium paid for these options. When option premiums fall, rational actors buy more options.

A venture capitalist would recognize this logic immediately, but the same option-value principle applies at national scale. More experiments mean more chances to discover new productivity pockets. The expected value of a broad entrepreneurship boom depends on the distribution of outcomes, not the median outcome. If AI produces thousands of mediocre businesses and a handful of transformational firms, the aggregate payoff can still be large. That is why dismissing elevated applications because many will fail misses the point. Failure is part of the discovery process.

There is a link to endogenous growth theory. Romer-style models emphasize ideas, non-rival knowledge, and the role of innovation in long-run growth. AI can increase the effective supply of idea implementation, not merely idea generation. Many people already have ideas, but they lack the means to test them. If AI converts more latent ideas into experiments, it increases the economy’s innovation throughput. Aghion-Howitt creative destruction models similarly emphasize the role of new innovations displacing old technologies. More entrants using AI-native workflows can accelerate that displacement.

Why The Skeptical Case Still Matters

The optimistic interpretation should not become a blind narrative. There are several reasons to be cautious. First, business applications are not the same as operating firms. The data are an early signal, not a final outcome. A filing can represent a side project, a tax entity, a single-person consulting vehicle, or an inactive company. The key question is conversion. Are applications turning into employer businesses? Are they producing revenue? Are they surviving beyond two or three years? Are they raising productivity or merely increasing churn?

Second, AI may create too much low-quality entry. If the cost of launching falls dramatically, markets may fill with thinly differentiated products, automated content, weak software, and service providers using similar tools. This can create noise for customers and compress margins for entrants. Low entry barriers are good for experimentation but difficult for profitability. The internet produced both great companies and a flood of fragile business models. AI will likely do the same.

Third, incumbents also use AI. Lower barriers to entry do not guarantee lower concentration. Large firms have proprietary data, distribution, cloud infrastructure, compliance teams, and capital budgets. They can integrate AI into existing products and acquire promising entrants. The result could be a barbell: many more small firms at the bottom, continued dominance by large platforms at the top, and pressure on mid-sized firms that lack both speed and scale. That would still be a structural change, but not necessarily a simple decentralization story.

Fourth, regulation and trust can slow the translation from applications to real businesses. AI-generated legal drafts, medical advice, financial recommendations, and customer communications can create liability. Industries with high compliance burdens may not allow tiny AI-enabled teams to move as freely as software demos suggest. Trust remains a human and institutional asset. In many markets, customers do not only buy output; they buy accountability.

Fifth, the macro environment matters. High interest rates, tighter credit, weak consumer demand, or a downturn in small-business lending could slow the boom. AI lowers some costs but not all costs. Businesses still need customers, working capital, payment systems, insurance, and time. The surge in applications should therefore be treated as a leading indicator of possible dynamism, not proof that productivity acceleration is guaranteed.

Investment Implications: More Entry, More Dispersion

For investors, the entrepreneurship boom points toward dispersion. If AI lowers entry costs, more firms can attack niches, more industries can be disrupted, and the distance between winners and losers can widen. The obvious beneficiaries are AI infrastructure providers, cloud platforms, semiconductor suppliers, and software companies selling productivity tools. But that is only the first layer. The second layer is the set of businesses that use AI to change cost structures in old industries. The third layer is the pressure placed on incumbents whose margins depended on high customer friction, slow service, or scarce expertise.

This suggests a different way to think about AI exposure. Owning the infrastructure winners captures part of the theme, but the broader economic effect may appear through changes in competitive intensity. Industries with high information-processing costs and fragmented customer needs are especially exposed. Professional services, marketing, education, healthcare administration, insurance distribution, compliance, recruiting, software development, local services, and business process outsourcing all have tasks that can be compressed by AI. Some incumbents will become more profitable by automating internal work. Others will face new entrants that use AI to underprice or out-personalize them.

Public markets may initially reward incumbents because they have the resources to deploy AI quickly and the margins to show near-term efficiency gains. Over time, however, the entry effect can matter more. If small firms can reach customers cheaply and operate leanly, incumbents may lose pricing power in specific verticals. This is not a universal short thesis against large companies. It is a warning that AI is both an efficiency tool and an entry tool. The first supports margins; the second attacks them.

Private markets may see a larger structural change. The cost of reaching product-market fit could fall, which changes the financing stack. Founders may need less seed capital to build a first product, but more capital to scale distribution once traction is proven. This can shift bargaining power toward founders in the earliest stages and toward investors with distribution, data, and go-to-market expertise in later stages. It may also increase the number of small profitable companies that never need traditional venture funding. That would be an important change because the venture model is built around a small number of very large exits, while AI may also enable a broad layer of durable micro-multinationals.

The labor-capital split is another investment question. If AI allows revenue to scale with fewer employees, profit margins can rise for successful firms. But if entry explodes, competition can pass those productivity gains to customers through lower prices. The distribution of gains depends on market structure. In markets with network effects, data advantages, or regulatory barriers, AI productivity may accrue to firms. In markets with low switching costs and many entrants, it may accrue to consumers. Investors should therefore focus less on generic AI adoption and more on whether adoption creates defensible advantage.

The Internet Analogy Is Useful, But Incomplete

The natural comparison is the internet boom of the late 1990s and early 2000s. That comparison is useful because the internet also created a surge in experimentation by lowering communication and distribution costs. A small team could reach national or global customers without owning physical storefronts. Search engines, email, websites, online payments, and later social media reduced the cost of discovery and customer acquisition. Cloud computing then lowered infrastructure costs and made it possible to rent computing capacity rather than build it.

AI extends this sequence, but the mechanism is not identical. The internet made markets easier to access. Cloud made infrastructure easier to rent. AI makes capability easier to invoke. It turns pieces of expertise into an on-demand service. A founder can ask for a market map, a draft contract, a prototype, a customer segmentation, a translation, a spreadsheet model, a code review, or a support workflow. The output still requires verification, but the first pass is no longer blocked by waiting for a specialist. That shifts bottlenecks from access to knowledge toward judgment about how to use knowledge.

This distinction matters for productivity. The internet allowed many firms to sell more broadly, but it did not automatically redesign the internal process of producing a product or service. AI is closer to the production function itself. It can touch the work of analysts, engineers, lawyers, marketers, teachers, designers, operators, recruiters, and support teams. That is why the business-formation effect may be more diffuse. The marginal entrepreneur does not need to be selling an AI product. He or she can be using AI to make a boring business cheaper to start and easier to run.

The historical analogy also warns against excessive short-term certainty. The internet bubble produced real overinvestment and real long-term infrastructure. Both were true. Many firms disappeared, but the fiber, software habits, consumer expectations, and platform models survived. AI may follow a similar pattern. Some current business applications will fail because the founders overestimate what AI can do, underestimate customer acquisition, or ignore compliance. Yet the experimentation itself can leave behind skills, workflows, reusable code, datasets, and customer insights. Even failed firms can contribute to diffusion if workers carry AI-native habits into the next venture or employer.

The better analogy may therefore be not just the dot-com boom, but the combination of electrification, the personal computer, the internet, and cloud computing. Each general-purpose technology required complementary assets before it transformed measured productivity. Factories had to be redesigned around electricity rather than merely replacing steam engines with electric motors. Offices had to change workflows around computers rather than treating them as faster typewriters. Firms had to reorganize around the internet rather than adding a website to an unchanged business. AI will likely require the same kind of complementary redesign. New firms are often where that redesign is easiest because they do not have to ask permission from legacy processes.

Policy, Institutions, And The Quality Of Selection

Whether the entrepreneurship surge becomes a productivity boom depends heavily on the quality of the selection environment. Low entry costs are valuable only if good firms can grow and bad firms can exit without excessive friction. That requires competitive markets, flexible labor mobility, reasonable bankruptcy rules, access to payment systems, functioning small-business finance, and a legal environment that protects customers without freezing experimentation. Entrepreneurship is not merely an individual act. It is an institutional ecosystem.

The United States has several advantages in this respect. It has deep capital markets, a culture that tolerates failure, relatively flexible labor markets, strong universities, large domestic demand, and a long history of commercializing general-purpose technologies. These advantages help explain why a technological shock can translate into firm creation. But there are also constraints. Healthcare tied to employment can make leaving a job riskier. Housing costs in high-productivity regions can limit mobility. Occupational licensing can slow entry in local services. Data privacy rules, intellectual property uncertainty, and AI liability questions can make founders cautious.

Policy should therefore focus less on subsidizing every AI startup and more on improving the environment in which experimentation is sorted. That means supporting portable benefits, reducing unnecessary local entry barriers, expanding access to technical education, improving small-business credit channels, clarifying AI liability standards, and maintaining competitive pressure against dominant platforms when they use data or distribution to block entrants. The goal is not to guarantee success. The goal is to make useful experimentation easier and selection cleaner.

There is also a measurement challenge. Traditional statistics may understate the significance of AI-enabled microfirms because many can generate meaningful revenue with few employees. If policymakers look only at payroll growth, they may miss a rise in high-revenue solo or tiny-team businesses. If they look only at applications, they may overstate the boom. The right measurement approach should combine applications, employer conversion, revenue, payroll, survival, productivity, and sector-level entry. Over time, tax data and business microdata will be essential for separating real operating dynamism from administrative filings.

The quality of selection also depends on customer trust. If AI lowers entry costs too far without quality control, markets can become polluted by low-quality providers. That would raise search costs for customers and reduce the value of experimentation. Reputation systems, professional standards, certification, insurance, and platform governance may become more important. The paradox is that AI lowers the cost of production but can increase the need for trust infrastructure. The most successful new firms may be those that combine AI leverage with credible human accountability.

Capital Allocation May Become More Barbell-Shaped

The financing implications deserve separate attention. A lower startup cost does not simply mean less demand for capital. It changes when capital is needed and what kind of capital is useful. Before AI, a founder might need funding early to hire engineers, designers, marketers, and operations staff before testing demand. With AI, some founders can delay that financing. They can build, test, and iterate with a smaller budget. That pushes the riskiest early experimentation away from institutional capital and toward founder time, customer prepayments, revenue, and small checks.

Once a business shows traction, however, capital needs may return quickly. Distribution, sales, compliance, security, enterprise integration, and working capital still require money. AI makes the prototype cheaper, but it does not make trust, customer acquisition, or scale free. This suggests a barbell financing structure. At one end, many small firms bootstrap for longer and may never raise traditional venture capital. At the other end, the firms that prove large markets may raise substantial capital to scale aggressively. The middle could become more difficult: companies that are too complex to bootstrap but not differentiated enough for major funding may struggle.

For venture capital, this can change the meaning of early-stage investing. If the cost of building a prototype falls, investors may place less value on the existence of a demo and more value on proprietary insight, distribution advantage, data access, customer urgency, regulatory positioning, and founder judgment. The scarce asset is no longer just the ability to build. It is the ability to decide what is worth building, sell it to the right customer, and defend the resulting position. AI may commoditize parts of execution while increasing the value of taste and strategy.

For public-market investors, the same logic implies that margin expansion from AI should be analyzed alongside entry risk. A company that uses AI to cut costs in a market with strong moats can produce durable shareholder value. A company that uses AI to cut costs in a market with weak moats may simply trigger a price war as new entrants adopt the same tools. The question is not “Does AI improve productivity?” The question is “Who captures the productivity?” In some markets it will be shareholders; in others it will be customers; in still others it will be specialized workers who become dramatically more productive.

This is why business-formation data belong in the AI investment conversation. They are not just a macro curiosity. They are a clue about competitive supply. If the number of potential entrants remains structurally higher, then the long-run AI trade is not only about buying the vendors of AI tools. It is also about identifying which incumbents can withstand a world where more people can build credible competitors. ## The Incumbent Response Will Decide How Much Dynamism Survives

The entrepreneurship boom will not unfold in a vacuum. Incumbents will respond, and the nature of that response will determine whether AI produces broad dynamism or simply strengthens existing leaders. Large firms can use AI to automate support, write code faster, personalize marketing, analyze customer behavior, and reduce administrative expense. If those gains are reinvested into better products and lower prices, incumbents can improve consumer welfare while maintaining scale. If they are used mainly to defend distribution and lock in customers, the entry boom may struggle to become a competitive boom.

This is where antitrust and platform governance intersect with entrepreneurship. Many AI-enabled startups will depend on cloud providers, app stores, search platforms, payment networks, data vendors, and social distribution. The same infrastructure that lowers entry costs can become a gatekeeper. A founder may be able to build a product cheaply, but still face high costs in discovery, trust, payment acceptance, and platform ranking. The bottleneck can move from production to distribution.

Investors should therefore separate two questions. First, does AI let more people build viable products? The answer increasingly appears to be yes. Second, does the market structure let those products reach customers and keep enough economics to survive? That answer varies by industry. In open, fragmented markets, the entrepreneurship boom can translate into real competition. In platform-controlled markets, entry may rise but surplus may be captured by the platform layer.

The most constructive outcome is one in which incumbents become more efficient while entrants keep them honest. That is the classic productivity bargain: technology lowers cost, competition passes some gains to customers, and the best firms grow. The less constructive outcome is one in which AI increases activity but not contestability. The business-application chart tells us that more people are trying. It does not yet tell us whether markets will let the best of them scale. ## What Would Confirm Or Refute The Boom

The next stage of analysis should move beyond applications. Several indicators matter. One is the conversion rate from business applications to employer identification numbers and active payrolls. Another is new-firm survival beyond two years. A third is revenue per employee among young firms, which would show whether AI is enabling leaner scaling. A fourth is sector composition: if the boom is concentrated in low-productivity administrative entities, the macro implication is weaker; if it appears in technology-enabled services, healthcare administration, software, manufacturing support, logistics, and professional services, the implication is stronger.

We should also watch productivity statistics with patience. General-purpose technologies diffuse slowly. The internet’s largest productivity effects took time and required complementary investment. AI may be faster because cloud distribution and software adoption channels already exist, but organizational redesign still takes time. The most important near-term evidence may come from microdata: small teams shipping products faster, professional-service firms increasing output per employee, customer-support operations handling more volume, and new firms reaching revenue milestones with fewer employees.

Another confirmation would be a widening gap between AI-native firms and legacy firms within the same industry. If AI is merely a tool everyone buys, the productivity effect may be broad but not especially disruptive. If AI-native entrants build different workflows, pricing models, and service expectations, then the competitive structure changes. The business-application surge would then be an early sign of a deeper reorganization.

A refutation would look different. If applications remain high but employer formation does not improve, if survival rates fall sharply, if revenue creation is weak, and if productivity growth fails to respond over several years, then the boom may be mostly administrative churn. That would still matter for labor-market behavior and household risk-taking, but it would not justify a strong productivity narrative. The burden of proof belongs to the optimistic case. Elevated applications are a signal, not a conclusion.

Conclusion: AI Turns Entrepreneurship Into A Cheaper Experiment

The United States appears to be experiencing a meaningful change in business formation. The rise from a pre-pandemic norm near 45,000 to 55,000 monthly applications to levels above 120,000 is too large to ignore. The pandemic explains the initial shock, but persistence at more than twice historical norms suggests something deeper may be happening. AI offers a coherent mechanism: it lowers startup costs, reduces the need for large initial employee bases, automates early cognitive and operational tasks, and raises the expected return on entrepreneurial effort.

The most important economic effect of AI may not be immediate replacement of workers or instant productivity miracles. It may be the expansion of who can try to build a business. When the cost of experimentation falls, more people experiment. When more people experiment, more firms enter the selection process. When selection works, the economy discovers more productive firms and reallocates resources toward them. That is how a business-formation boom can become a productivity boom.

This does not mean every application is meaningful, every AI startup is valuable, or every incumbent is vulnerable. The skeptical case remains important. We need conversion, survival, revenue, payroll, and productivity evidence. But the chart should be taken seriously. It may be showing the early macro footprint of AI as an entrepreneurial technology, not merely a corporate efficiency tool.

The internet lowered the cost of reaching customers and distributing information. Cloud computing lowered the cost of accessing infrastructure. AI lowers the cost of cognition, coordination, and first-draft execution. Put together, those forces can turn entrepreneurship from a capital-intensive leap into a cheaper, faster, more repeatable experiment. If sustained, that shift could boost productivity growth, revive business dynamism, and increase the resilience of the U.S. economy. The market will debate which companies win from AI. The deeper question is whether AI increases the number of people who can become companies in the first place.