US Venture Capital Outlook 2026 — What Has to Happen

This is an outlook, not a retrospective. Every statement about 2026 below should be read as a scenario or a monitoring question, not a forecast.

US venture capital enters 2026 with three unresolved problems operating on different timelines.

The exit backlog is five years deep. Companies that raised in 2020 and 2021 and have not yet found liquidity now represent a substantial accumulation of unrealised value held well past its underwritten horizon. This is the fastest-resolving of the three — a genuinely open listing window would clear a meaningful share of it within a year — and it is the one that unlocks the most, because distributions drive commitments and commitments drive everything downstream.

The early-stage gap is slower and cannot be fixed by any 2026 decision. Companies not funded at seed in 2022 through 2025 do not exist to raise Series A in 2026. Capital availability does not create them. This problem resolves only through several years of restored seed formation, and the earliest a restart in 2026 would show up at Series A is 2028.

The AI capital cycle is on its own clock, set by the depreciation schedules of assets already built. As the 2026 slot-A report argues, the capital expenditure decision was made on a demand forecast, but the depreciation charge arrives on a calendar. That timing is not contingent on anything, which makes it the most predictable of the three tests.

A fourth issue is administrative but consequential: the 2021 vintage reaches its terminal year for most funds. Positions deferred through bridges, extensions and continuation discussions since 2022 face a structural deadline.

Why distributions unlock the most

Of the variables that could change in 2026, distributions to limited partners have the widest downstream effect. The chain is mechanical and each link is real:

1. Exits produce distributions.
2. Distributions fund new LP commitments, since institutions operate to a target allocation rather than continuously adding money.
3. New commitments fund the next generation of venture funds.
4. Those funds deploy into companies, including at seed.

Five years of constrained exits have constricted every link. The important observation is that the constraint is now self-reinforcing: fewer commitments mean fewer funds, which means less deployment, which means fewer companies reaching the stage where they could exit, which means fewer future distributions.

A genuine improvement in exits relieves several problems simultaneously, which is what makes it the highest-leverage variable:

• Distributions restore commitment capacity.
• The denominator effect eases as private marks are realised and allocations return toward target.
• Reported IRRs improve even where multiples are unchanged, because return is time-sensitive and holding periods have extended.
• Carrying values are validated, closing the gap between marks and realisable value that has persisted since 2022.
• Seed funds can raise again, which is the only path to closing the cohort gap.

What to watch, and it is not IPO counts. The meaningful measure is distributions as a percentage of net asset value, because it measures capital actually returned rather than transactions announced. An IPO in which existing shareholders are locked up for six months produces no distribution in the year it occurs. Bain publishes this figure free in its annual private equity report.

The Series A signal problem

The cohort gap becomes observable in 2026, and it will produce a data pattern that is easy to misread.

The mechanism: seed formation outside favoured categories was constrained from 2022. Those companies would have reached Series A between 2024 and 2026. Fewer exist.

Meanwhile, capital available for Series A has not fallen proportionally — funds raised in 2021 and 2022 with growth mandates still need to deploy, and the reserve pressure described in the 2024 report has partially resolved as portfolios have been triaged.

Too much capital and too few qualifying companies produces rising prices. That is the signal that will be misread.

A market can produce rising Series A valuations for two entirely different reasons:

• Demand strength — more capital is competing because companies are performing better and the opportunity set has improved. This is a healthy market.
• Supply shortage — the same capital is competing for a smaller number of companies. This is a thin market that looks healthy in price data.

They are indistinguishable in valuation data and clearly distinguishable in count data:

Rising valuations with rising deal count is demand strength. Rising valuations with falling deal count is a supply shortage. In 2026 these will look identical in every headline and opposite in the underlying series.

The practical guidance is to read Series A deal count in 2026 against the 2019–2021 baseline rather than against 2023–2025, which were themselves constrained. A recovery relative to a depressed year is not a recovery.

The AI return test

The 2026 slot-A report treats this at the macro level. At the venture level, three questions matter and each has a different evidence trail.

Does application-layer revenue growth justify infrastructure-layer investment? The capital deployed at the infrastructure layer was underwritten on expected demand from applications. If application revenue accelerates, the investment is validated. If it plateaus, capacity was built ahead of demand. Evidence: revenue disclosure at listed application companies, and reported growth rates at private ones where available.

Does the model layer remain differentiated? Much venture capital at the model layer was underwritten on the assumption that leading models retain a capability advantage. If capability converges — if the gap between the best and the merely good narrows — models become a commodity input, pricing power erodes, and the value accrues to whoever owns the distribution or the data instead. Evidence: pricing behaviour, and whether customers switch providers on price.

Do applications build defensibility? The application layer's central question is whether businesses built on top of a commoditising input can construct durable advantage — through workflow integration, proprietary data, or distribution. Evidence: net revenue retention and gross margin, both of which are disclosed by listed comparables.

The most useful thing to note is that the evidence appears in a specific order. Infrastructure evidence — depreciation, utilisation, capex guidance — is publicly disclosed quarterly and appears first. Model layer evidence appears in pricing. Application layer evidence appears last, in retention data that takes multiple periods to become meaningful.

An investor who wants to update early should watch the filings, not the funding announcements.

The 2021 vintage's terminal year

An administrative deadline arrives in 2026 that will force resolution of positions deferred for four years.

Most venture funds have a ten-year term with extension provisions, typically two one-year extensions at the manager's discretion. A 2021-vintage fund reaches the end of its extended term around 2033 — but the investment period ended years ago, and more immediately, funds raised in 2016 and 2017 that made their final investments in 2021 are now at or past their terminal dates.

More practically, the positions themselves have reached a limit. A company that raised in 2021, bridged in 2022, extended in 2023, restructured in 2024 and held on through 2025 has exhausted the available deferrals. The options remaining are narrow:

• Exit at an achievable price, accepting whatever the market provides.
• Move to a continuation vehicle, which requires the asset to be attractive enough for new buyers.
• Restructure again, which requires existing investors to fund it.
• Wind down.

The consequence is that 2026 should see a higher rate of resolution — in both directions — than the preceding years. That will make the data look worse in some respects and better in others: more shutdowns recorded, more sales at prices below prior marks, but also a clearing of positions that have been distorting portfolio statistics since 2022.

Resolution is healthy even when the individual outcomes are poor. A portfolio in which every uncertain position has been resolved is a portfolio that can be assessed. One in which they have all been deferred cannot be.

What would change the picture

Rather than forecast, it is more useful to name what would constitute evidence.

Evidence the exit constraint is easing: distributions as a percentage of NAV moving toward long-run averages — not IPO counts, and not announced transaction values.

Evidence the early-stage gap is closing: seed deal count rising against a 2019–2021 baseline. Seed deal value alone can rise on concentration and tells you nothing.

Evidence the AI capital cycle is working: utilisation staying high on deployed capacity; useful-life assumptions in filings staying constant rather than being extended; application-layer net revenue retention holding.

Evidence the industry's composition is recovering: first-time fund formation rising. NVCA/PitchBook reports this as a separate line, and it is the cleanest indicator of whether the manager pipeline is being repaired.

Evidence that concentration is easing: the share of total deal value accounted for by the largest rounds falling, alongside rising deal count.

Each of these is a specific, publicly-available series. None requires a view about market direction, which is the point — an outlook is more useful when it specifies what would change its mind than when it states what it expects.

What the archive says about the three problems

Each of 2026's three problems has a precedent in the archive, and the precedents are informative about how they resolve.

On the exit backlog: corrections deferred get larger, not smaller. The 2019 report describes a private-market correction interrupted by the 2020 policy response and arriving in 2022 against a wider gap. The 2023 report describes bridge rounds deferring adjustments that cost more when they expired. In both cases the deferral consumed the resources that would have made the adjustment survivable.

Applied to 2026: five years of accumulated unrealised positions is a larger backlog than four, held by funds with less remaining life and by LPs with less patience. The backlog does not improve with age.

On the early-stage gap: the cost of a funding shortfall appears years after it is incurred. The 2024 report describes the emerging manager pipeline thinning, with the consequence appearing a decade later as a gap in the population of established managers. The company-level version is the same mechanism one level down.

Applied to 2026: no decision made this year closes the gap. Companies not funded at seed in 2022–2025 do not exist to raise Series A now. A restart in 2026 shows up at Series A in 2028 at the earliest — and the intervening years produce the rising-price, falling-count pattern that will be misread as strength.

On the AI capital cycle: evidence appears in filings before it appears in prices. The 2023 report describes one variable — duration — producing a bank failure, an exit drought and an index concentration, reported as three unrelated stories. The 2024 report describes constraint migration going undiagnosed because the familiar framework kept pointing at the wrong variable.

Applied to 2026: the infrastructure evidence is quarterly, public and free. It will be available well before the funding market responds to it, and the failure mode is not lack of data but a framework that is not looking at it.

The archive's most consistent finding is that the informative evidence is usually published and usually not read, while the widely-discussed evidence is usually a lagging summary of it.

The practical version of that finding for 2026 is a short list: hyperscaler capex and depreciation from quarterly filings, seed and Series A deal count from Carta, first-time fund formation from NVCA/PitchBook, and distributions as a percentage of net asset value from Bain. All four are free, all four lead the commentary, and together they cover every variable this outlook identifies.