If $1B Exits Are Dead, the $1B Company Is Changing Too

Andreessen Horowitz recently published a conversation with David George and VenCap CIO David Clark titled Why $1B Exits Are Dead. The title is deliberately provocative, but the underlying argument is more nuanced. AI companies are reaching meaningful scale unusually quickly, and if the largest companies of this generation ultimately become much larger than their predecessors, a $1 billion exit may no longer have the significance within a venture portfolio that it once did. For investors trying to generate top-tier returns, the implication is that finding and maintaining ownership in the relatively small number of enormous winners becomes even more important.

We think there is another implication worth examining, one that begins much earlier in the lifecycle of the company.

If AI changes the potential size and velocity of venture outcomes, it is unlikely to leave the economics of the underlying company untouched. The more interesting question, then, is not only whether the next generation produces more $10 billion, $50 billion, or $100 billion companies. It is what those companies will look like while they are being built. How much capital will they consume? How many people will they employ? Where will that capital go? And, perhaps most importantly for venture investors, what happens to the relationship between capital invested and productive capacity created?

That relationship has been remarkably durable throughout the history of modern venture capital.

Venture capital has always been partly a financing model for labor

We wrote recently that venture capital was built around what we call headcount economics (we wrote about that in a previous blog post; see here). The basic idea is straightforward: for much of venture's history, one of the principal ways capital increased a company's productive capacity was by allowing it to employ more people.

A seed round funded the first engineers and product hires. A Series A allowed the company to deepen product development and begin building go-to-market. Subsequent rounds financed sales organizations, customer success, new geographies, additional products, management layers and the operational infrastructure necessary to coordinate an increasingly complicated organization.

There were always other uses of capital, and software had already made companies dramatically less labor-intensive than their industrial predecessors. Still, for most venture-backed technology companies, growth in revenue, headcount, organizational complexity and capital requirements tended to move in roughly the same direction. As the company became larger, it generally required more people; as it required more people, it generally required more capital.

AI introduces the possibility of weakening that relationship.

That does not mean, as some of the more breathless predictions about AI employment suggest, that successful companies will simply stop hiring. There are good reasons to be skeptical of that conclusion. Productivity improvements can expand the amount of economically viable work rather than merely eliminate existing work, and companies facing larger opportunities have a tendency to reinvest productivity gains into growth. Indeed, some early evidence suggests that companies adopting AI aggressively are not necessarily the companies reducing headcount.

What is changing is subtler and, from an investment perspective, potentially more important: productive capacity can increase without headcount increasing at the same rate.

The denominator is starting to matter

We can already see unusually high levels of output per employee in some AI-native companies, although the sample is young enough that we should be cautious about turning examples into universal rules.

BCG estimated that Cursor was producing approximately $3.3 million in annual revenue per employee in 2025, compared with roughly $350,000 for the median public SaaS company in its benchmark. That is about 9.4 times as much revenue per employee. BCG appropriately characterizes some of its company-level estimates as directional, and comparing a rapidly growing private AI company with a broader group of mature public software companies is not apples-to-apples. Still, the difference is large enough to be noteworthy.

Lovable offers another useful example. The company said earlier this year that it had crossed $400 million in annual recurring revenue with 146 full-time employees. On a simple basis, that is approximately $2.74 million in ARR per employee: $400 million ÷ 146 employees = approximately $2.74 million per employee.

Again, we should resist extrapolating too much from companies experiencing extraordinary early growth. We do not yet know what these organizations will look like at maturity, how their cost structures will evolve, or whether today's remarkable revenue-per-employee figures persist as they build enterprise sales, compliance, support and international operations. But we do not need to know the eventual equilibrium to recognize that the frontier has moved.

The relevant shift is not “AI companies don't need people.” It is that a given number of people can increasingly command substantially more productive capacity.

An engineer working with capable agents can explore, build and maintain more software. A product team can test more hypotheses without proportionately expanding design and engineering resources. Support organizations can absorb greater customer volume before adding staff. Small teams can operate systems that would previously have required specialized functions around them. None of this makes human expertise irrelevant. In many cases, it makes experienced judgment more valuable because execution is no longer the binding constraint it once was.

For investors, however, it creates an important question: if productive capacity no longer scales linearly with labor, why should capital requirements scale the way they used to?

A $1 billion company may no longer require a $1 billion-company organization

For a long time, company value and organizational scale were imperfect but reasonably correlated. Companies became more valuable as they accumulated customers, products, markets, employees, managers and infrastructure. A billion-dollar software company generally looked and behaved like a substantially larger organization than a hundred-million-dollar software company.

AI gives us reason to question how durable that relationship will be.

A company reaching $100 million in revenue with 75 employees is not necessarily an immature version of a company that will eventually require 750. It may be approaching the same economic scale through a different organizational architecture. Some AI-native businesses will certainly become enormous employers, particularly if the technology opens markets large enough to justify continued reinvestment in people. But the important distinction is that headcount becomes more of a choice about where human judgment creates incremental value and less of an automatic consequence of growth.

Capital allocation should change accordingly. Payroll has historically represented one of the largest uses of venture financing. If software begins supplying a greater share of incremental productive capacity, companies have more discretion over where the next dollar goes. For some, it may go toward compute or proprietary data. For others, distribution, acquisitions, regulatory infrastructure or geographic expansion may become the scarce resource. And some companies may discover that they simply do not need as much capital as an equivalent business would have required a decade earlier.

This is where the implications extend beyond operating models and into venture portfolio construction.

AI may stretch the distribution in both directions

The a16z argument focuses on the upper tail, and appropriately so. Venture is a power-law business. If AI creates a new class of companies capable of reaching tens or hundreds of billions of dollars in value unusually quickly, then missing those companies becomes increasingly expensive. A portfolio filled with respectable billion-dollar exits may perform poorly relative to one containing meaningful ownership in a single generational company.

But there is another possibility that deserves more attention. AI may improve the economics of outcomes that venture has increasingly learned to dismiss as too small.

Consider two hypothetical companies that each sell for $500 million. The first has raised $150 million on its way there. The second has raised $15 million. Ignoring dilution, preferences, ownership and time for the moment, the ratio of exit value to capital raised is dramatically different:

$500M ÷ $150M = 3.3x

versus

$500M ÷ $15M = 33.3x.

Those figures are emphatically not investor returns. Actual returns depend on entry price, ownership, dilution, liquidation preferences, secondary transactions, follow-on investment and holding period. But the example exposes something that headline exit values obscure: a $500 million company built with $15 million of external capital is economically different from a $500 million company that consumed ten times as much capital to reach the same destination.

If AI allows more companies to build meaningful revenue with less labor and therefore less external financing, venture may experience two changes simultaneously. The upper tail may become considerably larger, making the biggest winners more important. At the same time, the capital efficiency of outcomes further down the distribution may improve enough to make some of those outcomes more attractive than their valuations alone suggest.

Those ideas are not contradictory. In fact, they arise from the same underlying change.

We may need better measures of company quality

For years, venture has relied on a set of proxies that made sense in a world where organizational capacity was expensive to build. Headcount growth signaled investment. A large engineering organization suggested technical capacity. Rapid hiring after a financing round was often interpreted as evidence that a company was scaling. Later-stage companies were expected to accumulate functions and management layers because that was what operating at greater scale required.

Some of those signals are becoming less informative.

Revenue per employee is an obvious replacement metric, but it is an incomplete one. A company can maximize revenue per employee by underinvesting in areas that matter for long-term durability. The more interesting measures will probably examine how efficiently companies convert scarce inputs, including capital, human judgment and time, into durable enterprise value.

This also changes the questions boards should ask management teams. The useful question is no longer simply how many engineers the company needs to hire next year. It is which constraints are actually preventing the company from producing more value, and whether another employee is the highest-leverage way to remove each one. Likewise, fundraising should become less about financing a predetermined stage of organizational expansion and more about identifying what scarce resource additional capital actually purchases.

That sounds obvious, but much of the venture ecosystem is institutionally organized around the older model. Financing stages, hiring plans, management structures, benchmarks and even our language around “scaling” emerged during a period in which growing the company and growing the organization were closely linked.

They may no longer be.

The financing model eventually has to follow the operating model

One of the stranger possibilities created by AI is that companies may begin reaching what we traditionally consider growth-stage economics while retaining what looks, organizationally, like an early-stage company.

A 30-person business could have the technical capacity that once required 100 people. A 50-person company could serve a global customer base. A company could reach substantial recurring revenue before building many of the functional departments investors historically expected to see at that stage.

If that happens with any frequency, the familiar Seed-to-Series-A-to-Series-B progression becomes less descriptive of what is actually happening inside companies. Those rounds have never been purely about headcount, but they have been closely associated with predictable stages of organizational development. When those stages compress, overlap or disappear, round labels tell us progressively less about the maturity and capability of the underlying business.

There is an important caveat here. We do not yet have enough longitudinal evidence to know whether today's extraordinarily lean AI companies remain lean. It is entirely possible that some are merely postponing organizational complexity rather than eliminating it. Enterprise requirements, regulation, customer support, international expansion and the mundane work of running large institutions have historically created headcount regardless of how efficiently the underlying product is built. I cannot confirm that AI permanently changes that dynamic.

But investors do not need certainty about the final organizational form to recognize that the assumptions embedded in the current one deserve reexamination.

The most consequential change AI brings to venture may therefore be broader than bigger exits. It may change the relationship between the size of a company, the size of its organization and the amount of capital required to build it.

The enormous outcomes will attract most of the attention, as they should. If this generation produces companies worth hundreds of billions of dollars at unprecedented speed, those companies will determine fund returns and reshape entire markets.

But underneath those outliers, something potentially more pervasive is happening. The cost of creating organizational capacity is falling. That makes it possible to build meaningful companies with different combinations of people, capital and time than venture has historically assumed.

So when we ask whether the $1 billion exit is becoming obsolete, we may be starting at the wrong end of the equation. Before AI changes what a company can be worth, it changes what is required to build one.

And if the next $1 billion company can be built with materially fewer people, less capital, or in substantially less time than the last one, then it is not simply a more efficient version of the company venture capital was designed to finance; it is a different economic object. Eventually, venture will have to price it that way.

---

Sources