US Venture Capital Report 2024 — Two Markets

2024 should, on the standard framework, have been a recovery year for US venture capital. Rates fell. Public markets performed well. The 2021 vintage's problems were largely acknowledged and marked. The conditions that caused the contraction had reversed.

The recovery was partial and uneven, and understanding why is the substance of the year.

The constraint had moved. In 2022 the problem was the cost of capital — higher rates made long-duration assets less valuable, and the repricing followed mechanically. By 2024 the problem was the availability of capital at the LP level, which is a different variable and one that rate cuts do not directly address.

The chain described in the 2023 report had not been repaired. Distributions were still well below the level needed to sustain commitment pacing, because exits had not reopened at scale. Rate cuts lower the cost of capital, but they do not make an institution's private portfolio return cash. A constraint that operates through liquidity rather than price is not relieved by making price cheaper.

The second defining feature was the deepening separation between two populations of companies. Businesses with a credible AI position experienced a market resembling 2021 — abundant capital, rapid processes, competitive rounds. Businesses without one experienced a market resembling 2023. Both were true simultaneously and within the same asset class.

Underneath both, a slower change was occurring that received almost no attention: the pipeline of new venture firms was thinning. That is a change to the industry's composition rather than its conditions, and it compounds over a much longer horizon.

Why rate cuts did not reopen exits

This is the most important analytical point of 2024 and it generalises well beyond venture capital.

The standard model says lower rates support asset prices and reopen capital markets. That model is correct when the constraint is the cost of capital.

By 2024 the venture constraint was elsewhere. The loop runs: exits → distributions to LPs → new commitments → fund deployment. It had broken at the first link, and each subsequent link had constricted in turn.

Rate cuts affect this loop only indirectly, and each transmission step is slow and lossy:

• Lower rates raise public market valuations, which should improve the environment for listings.
• But listing decisions depend on more than valuation — on whether a company's financial profile can withstand quarterly disclosure, on comparable company trading, and on whether existing shareholders will accept the achievable price relative to their carrying value.
• And the backlog was very large. Four years of accumulated unrealised positions cannot clear through a modestly improved window.

There is a further friction that is rarely stated. A company whose last private round priced it above what public markets will pay has an incentive to wait, because listing crystallises a loss for existing investors that remains theoretical while private. That incentive persists until either the company grows into the valuation or its investors accept the adjustment. Neither happens quickly, and both are decisions made by people whose compensation is affected by the outcome.

Rate cuts change what capital costs. The 2024 problem was that capital had not come back — a distinction that determines which policy levers work and which do not.

The general principle is worth stating on its own: when a system has multiple potential constraints, relieving one that is not binding produces no effect. Diagnosing which constraint is actually binding is the analytical work, and it is frequently skipped in favour of applying the familiar framework.

Two populations, one asset class

The separation that became sharp in 2025 was clearly visible in 2024, and its mechanics are worth setting out at the transaction level.

Population one comprised companies with a credible artificial intelligence position — at the model layer, in infrastructure, or in applications with demonstrable adoption. For these companies:

• Rounds were competitive and frequently pre-emptive.
• Processes ran in weeks rather than months.
• Valuations were set by competition among investors rather than by comparables.
• Capital was available in size, at multiple stages, from both traditional and non-traditional sources.

Population two comprised everything else. For these companies:

• Processes were long and outcomes uncertain.
• Diligence was extensive, with particular focus on capital efficiency and path to profitability.
• Valuations were frequently below prior marks, or flat with structure.
• Existing investors were the most likely source of new capital, which limited the price discovery available.

The same investors participated in both markets, applying different standards depending on which population a company fell into. That is not inconsistency — it reflects genuinely different competitive conditions in the two segments.

The practical difficulty this created was classification. Whether a company counted as population one was frequently ambiguous, and the answer materially affected its valuation and its access to capital. This produced predictable behaviour: companies repositioned their narratives toward AI, sometimes reflecting genuine product changes and sometimes reflecting the incentive to be classified favourably.

For investors, this made diligence harder in a specific way. The question shifted from "is this a good business" to "is this AI position real" — and the second question requires technical assessment that not every investment team was equipped to perform.

When a category changes its capital requirement

A structural shift occurred in 2024 that changed what venture capital could fund, and it received less attention than the valuations did.

Artificial intelligence had previously been a software category. Software has favourable capital characteristics: development costs are largely people, marginal costs approach zero, and a company can reach substantial revenue on modest capital.

Frontier model development is not that. It requires:

• Large-scale computation, purchased or rented at costs that dwarf typical software development budgets.
• Data acquisition and preparation at scale.
• Specialised talent competed for by very well-capitalised buyers.
• Ongoing rather than one-time capital, since each model generation requires the cycle again.

This changes the economics of funding such a company by roughly an order of magnitude. A venture fund that could seed twenty software companies could seed a fraction of one frontier model developer.

The consequences reshaped the market's structure:

• Non-traditional capital became essential at the frontier — strategic investors, sovereign funds, and corporate balance sheets, none of which operate on venture return requirements.
• Compute-for-equity arrangements emerged, in which infrastructure providers supplied capacity in exchange for stakes, blurring the line between customer, supplier and investor.
• Traditional venture repositioned toward the application layer, where the capital requirements still resembled software.
• The definition of a venture-scale opportunity narrowed at the infrastructure layer, because the capital required exceeded what most funds could provide even in aggregate.

The 2024 slot-A report describes this reclassification at the macro level. At the venture level, its significance is that a category which venture capital had considered its own moved partly outside the range that venture capital can fund.

The emerging manager problem

The least visible development of 2024 concerns the composition of the industry rather than its conditions.

New venture firms — first and second-time funds — are the mechanism by which the industry renews itself. They introduce new strategies, new geographic focus, new networks, and they are frequently where the highest-returning funds of a vintage originate, because a small fund concentrated in an under-covered area can produce returns a large diversified one cannot.

Raising a first fund in 2024 was extremely difficult, and the reasons were structural rather than performance-related:

• LPs prioritised re-ups. With limited new commitment capacity, institutions supported existing relationships. A new manager competes for the residual.
• Track record requirements tightened. In a constrained environment, the burden of proof rises, and a first-time manager has no fund-level record to offer.
• Institutional minimum cheque sizes exclude small funds mechanically. An LP that must write $20m cheques cannot invest in a $50m fund without taking an uncomfortably large share of it.
• The fee mathematics are hard. A small fund generates modest management fees, which limits the team it can support during the years before carry.

The effect compounds slowly and is hard to reverse. A manager who cannot raise a first fund in 2024 does not raise a second in 2027 or a third in 2030. The cost appears a decade later, as a gap in the population of established managers — precisely analogous to the company-level cohort gap described in the 2025 and 2026 reports, operating one level up the capital chain.

Reserves over new deployment

A practical change in fund management became dominant in 2024: the reserve allocation decision.

Venture funds hold back a portion of capital for follow-on investments in existing portfolio companies. In favourable conditions this is straightforward — support the winners, let the others resolve.

By 2024 the calculation had changed:

• More companies needed capital, because the market was not funding them from outside.
• Fewer had clearly earned it, since the milestones that would justify support had become harder to reach.
• New commitments were constrained, so a fund could not simply raise the next vehicle and move on.
• The alternative to supporting a company was frequently its failure, which is a different decision from declining to increase a position.

The result was that a much higher share of deployment went to existing positions rather than new ones. That is rational at the portfolio level but has a market-level consequence: capital allocated to reserves is capital not available to new companies. Some of the reduction in new company formation in 2024 reflects this reallocation rather than any judgement about the opportunities.

The classification problem

The two-population structure created a diligence difficulty that was new in kind rather than degree, and it is worth setting out because it changed what venture diligence has to be able to do.

The shift in the question. In a normal market the diligence question is whether a business is good — the market, the team, the product, the economics. In 2024 an additional question came first and frequently dominated: is this company's AI position real?

Why that question is harder:

• It requires technical assessment. Distinguishing a genuine capability from a wrapper over a third-party model requires understanding what the model does, what the company adds, and whether the addition is defensible. Not every investment team was equipped for this.
• The incentive to reposition was strong. A company classified favourably raised more easily at a better price. Repositioning a narrative toward AI was rational whether or not the underlying product changed, and both happened.
• The boundary was genuinely ambiguous. A software company that uses a model to improve a feature and a company whose entire product is the model occupy the same category in most datasets and are entirely different investments.
• The answer determined the price. Which meant getting it wrong was expensive in both directions — overpaying for a repositioned business, or missing a genuine one because the assessment could not be made.

The practical questions that separated the two:

• What breaks if the underlying model is replaced with a competitor's? If nothing, the company is a distribution and workflow business that uses a commodity input — which may be a fine business, priced differently.
• What proprietary data or workflow integration exists? These are the defensibility mechanisms available at the application layer, and they are assessable.
• What is the gross margin, and where does it go as usage scales? A business whose input cost scales with revenue has a different margin structure from software, and it shows up here first.
• Is retention improving or is growth from new customers? Net revenue retention distinguishes an embedded product from a trialled one.

When a category determines price, classification becomes the diligence. That is an uncomfortable position, because it means the most consequential judgement in the process is the one least amenable to the tools most investors have.

What an allocator could act on

Diagnose which constraint is binding before applying a remedy. Rate cuts address the cost of capital and do nothing for its availability. In 2024 the constraint was availability — distributions had not recovered, so commitments could not be made. A familiar remedy producing no effect is diagnostic information, and the correct conclusion is that the constraint has moved rather than that more of the remedy is needed.

Measure distributions as a percentage of NAV. It measures capital returned. Listing counts measure activity. Only the first relieves the constraint that determines commitment pacing, and Bain publishes it free.

Separate the capital intensity question from the thematic one. A category whose capital requirement rises by an order of magnitude may move outside the range a venture fund can fund, regardless of how attractive the theme is. A fund sized for software economics cannot participate at infrastructure scale, and repositioning toward the application layer was the rational response rather than a retreat.

Watch first-time fund formation. NVCA/PitchBook reports it separately and free. A manager who cannot raise a first fund in 2024 does not raise a third in 2030, and the cost appears a decade later as a gap in the population of established managers — the same cohort-gap mechanism operating one level up the capital chain.

Ask about reserve policy. A much higher share of deployment went to existing positions in 2024 than to new ones. That is rational at the portfolio level and it means capital allocated to reserves is capital not available to new companies. Some of the reduction in new company formation reflects this reallocation rather than any judgement about opportunities.

What 2024 established for US venture

• Constraint migration was demonstrated: relieving the cost of capital does nothing when the binding constraint is its availability.
• The two-population structure became clear, along with the classification problem it creates for diligence.
• A category outgrew venture's capital range at the infrastructure layer, requiring non-traditional capital.
• The emerging manager pipeline thinned, a compositional change whose cost appears a decade out.
• Reserve allocation displaced new deployment as the dominant use of fund capital.