The American company with no employees is not a marginal phenomenon. There are approximately 29.8 to 30.4 million of them (data from 2022-2023), which together produce about 6.8% of American GDP. Each month, a growing number of new ones are added. This phenomenon was already circulating before AI. What has changed is what a single operator can now produce.

Jensen Huang said it plainly during a Q&A session for media at the Nvidia GTC conference in San Jose in March 2026: within ten years, Nvidia aims to have its 75,000 projected employees working alongside 7.5 million software agents, or 100 agents per person. This ratio is not an anecdote about an exceptional technology company. It is a signal about what the minimum viable size of an organization is becoming.

The Essentials

  • Approximately 29.8 to 30.4 million companies with no employees exist in the United States (data from 2022-2023), generating about 6.8% of GDP, with a growing number of new creations each month
  • Jensen Huang shared his vision that Nvidia would ultimately deploy 7.5 million AI agents for 75,000 humans, or 100 agents per employee — a projection over a ten-year horizon, not a current operational reality
  • A solo operator equipped with a stack of agents can produce what a team of ten did two years ago, according to the Microsoft Work Trend Index 2026
  • No American social protection system was designed for this profile: no unemployment insurance, no publicly funded continuous training, no automatic retirement
  • Solutions exist — portable rights, adapted taxation — but no country has seriously implemented them

A Solo Operator with a Stack of Agents Produces What a Team of Ten Did Yesterday

The Microsoft Work Trend Index 2026 documents what many intuited without being able to measure it. A sole founder, equipped with AI agents for prospecting, writing, customer service, accounting, and product development, reaches production levels that required a small team twenty-four months ago. This is not an abstract claim about productivity: it is a transformation in the relationship between organization size and production capacity.

The effect is cumulative. Each layer of automation adds leverage. An agent that responds to customer emails frees up hours. An agent that generates first drafts of content reduces creation time. An agent that monitors metrics and generates reports eliminates an entire function. Layered together, these gains do not simply add up: they multiply.

Fortune has documented several concrete cases: solo founders managing six-figure revenues in sectors as varied as recruitment consulting, e-commerce, and content creation, operating with stacks of agents that would have been called science fiction in 2022. The limit is no longer available human time. It becomes the ability to orchestrate agents coherently.

This displacement of the constraint is fundamental. For two centuries, the size of an organization was constrained by the cost of coordinating humans. Ronald Coase theorized this in 1937: the firm exists because it is sometimes cheaper to coordinate internally than to resort to the market. AI does not destroy this reasoning. It radically changes its parameters by bringing the cost of coordinating cognitively intensive tasks down to near zero.

Nearly 30 Million Companies with No Employees: a Critical Mass Preceding AI

It is important to place this figure in its historical context. The approximately 29.8 to 30.4 million American nonemployer firms are not a creation of AI. They have existed for a long time, driven by decades of platformization, the rise of independent work, and an entrepreneurial culture that values autonomy. What AI does is change the nature and potential of this ecosystem.

Before AI agents, a company with no employees was structurally limited to what a single human could produce alone. An independent consultant could bill their hours, nothing more. A content creator was bounded by their own pace. A freelance developer could only take on a certain number of projects in parallel. These physical and cognitive constraints defined the ceiling on achievable solo revenues.

These constraints do not disappear, but their threshold recedes. A consultant who automates the production of standard deliverables can take on twice as many clients. A creator who delegates research, editing, and distribution to agents can multiply their presence on platforms. A developer who orchestrates coding agents can deliver faster and penetrate markets that were closed to them before.

The direct consequence: some of these millions of companies that were stagnating at subsistence-level revenues can cross thresholds of genuine economic viability. And new creations arrive in an environment where the entry ticket to productive activity has never been lower. AI skills pay five times more than a master’s degree: the return on mastering these tools is multiplied for those working alone.

What Nvidia Reveals About Large Companies Also Applies to Small Ones

Jensen Huang’s vision deserves careful examination. Nvidia’s ambition of 100 agents per human does not imply that Nvidia employees would each run 100 agents themselves. It means that the organization as a whole would integrate agents into its workflows at a scale far exceeding the number of its employees — Nvidia currently counting approximately 36,000 to 42,000 according to 2025-2026 data, with a projection of 75,000 within ten years.

What strikes is that the logic is identical at the scale of a solo operator. A single founder orchestrating a stack of ten agents is 10 agents per human. The difference with Nvidia’s vision is a matter of degree, not of nature. The organization of work in both cases rests on the same principle: delegate repetitive, procedural, or parallelizable tasks to agents, concentrate human effort on judgment, relationships, and direction.

This principle is not new in its intent. What is new is the scope of tasks an agent can handle. In 2020, a chatbot managed FAQs. In 2026, an agent can draft a competitive analysis, extract data from a contract, generate functional code, or design a complete email campaign. The expansion of the delegable scope is the real rupture.

For large enterprises, this raises questions about workforce structure that Daron Acemoglu and Simon Johnson have begun to document in their work on the capture of productivity gains by capital. For solo operators, the dynamic is different: they capture the gains themselves. The business owner and the AI user are the same person.

The Black Hole of Social Protection

This is where the picture changes color. The American solo operator running five agents and generating $200,000 in annual revenue has no access to unemployment insurance if they lose their clients. They do not contribute to any automatic retirement system. They benefit from no collectively funded continuous training provision. If they fall ill and can no longer work, their activity stops.

The American social protection system was designed for the permanent employee of an identifiable employer. The independent worker has always been poorly treated within it: this is a reality predating AI. But AI accelerates the rise of this profile without institutions following suit. If newly created companies each month integrate a growing proportion of solopreneurs amplified by agents, the mass of active workers not covered by protection grows at rapid speed.

The paradox is visible: a solo operator with a stack of agents can be economically more productive than an employee in a large enterprise, while being far more exposed to shocks to income, health, and retirement. Productivity increases. Security does not follow. The labor share of GDP is declining less than once believed: but this overall maintenance masks profound recompositions in how work is organized and protected.

This problem is not unique to the United States. In France, the self-employed worker system has created a category of independent workers whose retirement and training rights remain far below those of employees. In the United Kingdom, debates about the status of gig workers have occupied courts for years. No developed country has produced a satisfactory answer to the question: how do you protect the productive independent worker without discouraging them from being productive?

Solutions Exist, No One Has Seized Them

Economic literature has identified the problem for a long time. The most discussed solution has a name: portable rights. The idea is simple in principle: rights to training, retirement, and certain forms of social protection follow the individual, not their employment contract. Whether they are an employee, independent, or in transition between the two, their rights continue to accumulate.

In the United States, proposals in this direction have circulated since at least the late 2000s, notably around Aspen Institute work on “independent worker benefits.” The idea of an individual training account funded by each hour worked, regardless of employment status, resurfaces regularly in political circles. It has never passed the legislative stage at the federal level.

In France, the Personal Training Account (CPF) is the most advanced approximation of this principle among developed economies: it accumulates independently of status and follows the individual. But its amounts remain modest compared to the real cost of qualifying training, and its financing does not adapt to the logic of a solopreneur whose revenues fluctuate significantly.

On taxation, the path is equally well-marked. Productive micro-enterprises amplified by AI do not resemble traditional service self-employed workers. They can generate high margins with few expenses. A tax system that encourages reinvestment in productive tools, including subscriptions to agent platforms, would treat these companies as what they are: miniature capitalist entities, not workers disguised as businesses.

This analytical framework is not out of reach. What is missing is the political will to define a new economic category, neither the salaried worker of the twentieth century nor the precarious independent. A solo operator with millions of potential virtual agents at their disposal is neither one nor the other.

The Question That Remains Open

AI reduces the entry cost to significant economic production. This is good news for millions of individuals who had neither the capital nor the team to launch a viable activity. It broadens access to a form of genuine economic autonomy, not merely symbolic.

But the productivity of a solo operator is not the same thing as their economic security. The two can advance together, provided that institutions adapt to the reality they have been seeing form for years. Nearly 30 million companies with no employees represent a clear enough statistical signal to justify a serious political response.

The question is not whether this transformation of work will continue. It will. The question is whether protection and support mechanisms will be designed for the world arriving, or will continue to apply to the one departing.

Sources

  1. Fortune / Microsoft Work Trend Index 2026 — Solo founders using AI to replace entire teams: fortune.com
  2. U.S. Census Bureau — Nonemployer Statistics (data on companies with no employees)
  3. Microsoft Work Trend Index 2026 — microsoft.com/en-us/worklab/work-trend-index
  4. Jensen Huang Statement on Nvidia Agents — reported at Nvidia GTC conference, San Jose, March 2026 (media Q&A session); Jensen Huang also spoke at Stanford’s SIEPR Economic Summit in March 2024
  5. Aspen Institute — Future of Work Initiative, work on “independent worker benefits” (aspeninstitute.org)
  6. Acemoglu, D. & Johnson, S. — Power and Progress, PublicAffairs, 2023
  7. U.S. Census Bureau – Nonemployer Statistics 2022 (NES): https://www.census.gov/library/stories/2025/05/smallest-businesses.html
  8. U.S. Census Bureau – Nonemployer Statistics 2023 (NES-D): https://www.census.gov/newsroom/press-releases/2025/business-owner-characteristics.html
  9. U.S. Census Bureau – NES Growth 2012-2023: https://www.census.gov/library/stories/2025/07/nonemployer-business-growth.html
  10. Fortune – Jensen Huang GTC 2026, 7.5M agents projection: https://fortune.com/2026/03/19/jensen-huang-nvidia-ai-agents-future-of-work-autonomous/
  11. SIEPR Stanford – Jensen Huang at SIEPR 2024 (not 2025): https://siepr.stanford.edu/siepr-events/summits/2024-siepr-economic-summit
  12. MacroTrends – Nvidia Employee Count: https://www.macrotrends.net/stocks/charts/NVDA/nvidia/number-of-employees
  13. Microsoft Work Trend Index 2026: https://www.microsoft.com/en-us/worklab/work-trend-index/agents-human-agency-and-the-opportunity-for-every-organization
  14. Coase 1937 – The Nature of the Firm: https://en.wikipedia.org/wiki/The_Nature_of_the_Firm
  15. Aspen Institute – Portable Benefits Resource Guide: https://www.aspeninstitute.org/publications/portable-benefits-resource-guide/
  16. Acemoglu & Johnson – Power and Progress (MIT): https://shapingwork.mit.edu/power-and-progress/