Andreessen Horowitz’s latest essay, “Catching the (Venture) Bus”, makes a claim that would have sounded like special pleading a decade ago and now reads as a description of the market. Technology, to borrow Marc Andreessen’s line, is “the dog that caught the bus”. Venture is no longer a cottage allocation to be sized at 5–10% of a portfolio; it is where a growing share of the economy’s value is created. SpaceX listed in June at a $1.77 trillion valuation, the largest venture-backed IPO by a factor of ten. Anthropic was valued at $965 billion in May, OpenAI at $852 billion in March.
We agree. The evidence has moved decisively in a16z’s direction, and the case for a larger, more deliberate venture allocation is stronger today than at any point in the last twenty years.
Where we differ is not on the destination but on the route. a16z’s prescription, to concentrate capital into the few managers and companies that capture the power law, is correct for most of the dollars in the asset class. It is incomplete for the rest. The fare a fund pays to board, set by stage, ownership and entry price, decides whether the ride compounds. And the way cash returns to limited partners is about to change, as the largest private companies leave the private market.
THE BUS IS REAL
Venture is no longer a cottage industry. The data now says so as loudly as the venture industry does.
The largest companies in the S&P 500 are overwhelmingly technology businesses built with venture capital, and the Magnificent Seven alone account for roughly a third of the index. a16z calls the power law “the shape of our economy”; the index agrees. Behind it, the number of US listed companies has roughly halved since the late 1990s, while the number of unicorns has risen from three in 2006 to 945 today. As we set out in “Venture Is No Longer Setting Its Own Price”, what once would have been public equity is now private equity.
The exits have followed. US venture-backed exit value reached $2.19 trillion in the first half of 2026, ten times the $217 billion recorded in the whole of 2025. SpaceX accounted for $1.77 trillion, but even excluding it, the first six months of the year delivered almost twice the exit value of the previous twelve.
So have the returns. Cambridge Associates’ US venture index returned 21.1% in calendar 2025, against 7.6% for buyouts. The institutions that held early venture exposure have been rewarded: the University of North Carolina’s endowment returned 37.8% in its 2026 fiscal year, roughly half of it attributed to SpaceX through a 2009 venture fund commitment.
The concentration a16z warns about is equally real, and our own research reached the same conclusion before their essay was published. In “Born Winners: The Age of Day-Zero Unicorns”, we found that 40% of US venture dollars in 2025 went into the ten largest deals, and nearly half of LP capital to the ten largest funds. In the first half of 2026, a16z, Thrive and Founders Fund raised 48.1% of all US venture capital between them. And of 2,143 global venture funds from the 2000–2018 vintages, a16z’s PitchBook screen finds that only 51 returned five times their capital in cash.
On the core argument there is little to dispute: for most institutional portfolios, being underexposed to venture is now the larger risk.
FOR GROWTH CAPITAL, ONLY THE KINGS COMPOUND
Most venture dollars now enter at prices that leave room for only one kind of outcome.
a16z’s argument is strongest where most of the money sits. In 2025, 68% of US and Canadian venture dollars went into late-stage rounds. In the first quarter of 2026, late-stage rounds took 82% of global venture funding, and four companies, OpenAI, Anthropic, xAI and Waymo, took 65% on their own.
For that capital, the arithmetic is unforgiving. In “Born Winners” we worked through a typical example: an investor that takes half of a $200 million round at a $1.8 billion post-money valuation owns 5.5% of the company. A $20 billion exit, a result almost any founder would celebrate, returns $1.1 billion on that stake, a sum that barely moves a multi-billion-dollar fund. At growth-stage prices the problem compounds. A tenfold return from a $10 billion entry requires a $100 billion exit before dilution, and only around 30 venture-backed companies have ever reached that value. Even Stripe, whose valuation rose from $91.5 billion to $159 billion between its February 2025 and February 2026 tender offers, falls well short of the multiple a growth fund needs from a single position.
The fund math, by stage
When a growth investor is in a genuine category winner, the model works. Menlo Ventures invested more than $500 million in Anthropic at a valuation of around $18 billion; the stake is now reportedly worth around $14 billion. When it is not, the outcome looks like Tiger Global’s $12.7 billion 2021 fund, still showing losses three years later and ranked in the bottom decile of its vintage.
For growth capital, the question is not whether to own the winners. It is whether you can afford not to.
THE KINGS ARE CROWNED LATE, BUT OWNED EARLY
The largest venture outcome of the cycle was seeded from a $220 million fund.
This is where a16z’s argument stops short. “Only the winners matter” describes growth capital. It does not describe the whole asset class, because dollar-weighted data is not the same as outcome-weighted data. The largest venture-backed IPO in history was not built on a growth cheque. Founders Fund first invested in SpaceX in 2008, $20 million from a $220 million second fund. Across its vehicles the firm invested around $600 million, reportedly worth more than $50 billion at the IPO.
Anthropic tells the same story within a single cap table. An investor in its 2021 Series A paid roughly 1,600 times less for the company than the Series H investors of May 2026, before dilution. An investor in the September 2025 Series F paid around five times less. Both backed the same winner. Only one bought an early fare.
The mechanism has intensified. The crowning round, the pre-emptive raise that certifies a company as a category leader, is arriving earlier. Crunchbase counted 59 companies that reached unicorn status at seed or early stage in 2025, and 47 in the first quarter of 2026 alone. The median step-up for AI rounds reached 2.2x in the first quarter, against 1.5x for everything else, and a16z’s own data shows the gap between unicorn funding rounds has fallen to a year. Every one of those rounds marks up the investors already on the cap table.
We see the effect directly in our own programme. One underlying fund in our fund-of-funds portfolio is already at 8x DPI. Three more have the potential to be marked above 5x by the end of the year, on positions that were king-made at Series B by precisely the growth and sovereign capital a16z describes. For the capital that wrote it, the crowning round was the prize. For the funds that arrived first, it was the payoff on a bet already placed.
The broader data points the same way. In an analysis of more than 1,300 funds in PitchBook’s data, 25% of funds under $350 million delivered a TVPI above 2.5x, against 17% of funds above $750 million. The access still exists; it has simply moved. New capital entering an oversubscribed mega-round buys minimal ownership and no influence. A smaller pool of capital, sub-$1 billion funds and family offices structured to take early risk, continues to win allocation in the same companies before they are crowned. That capital barely registers in dollar-weighted statistics. It does not barely register in the returns it produces.
A $1–10 billion outcome held with real ownership still produces 8x and better at the fund level. The power law has not closed to smaller funds. It has closed to late fares.
AS THE KINGS LIST, SECONDARY CAPITAL MOVES DOWN THE LIST
The capital raised to buy the largest private companies is about to lose its targets. It will not sit idle.
In “The Rise of Venture Capital Secondaries” we described a market that clears for perceived winners and little else. That remains true. The US venture secondary market reached $121.7 billion over the twelve months to June 2026, but the top 20 names accounted for 86% of trading value on Hiive in the second quarter, and the top five for more than half.
What is changing is the top of the list. SpaceX is now public. Anthropic filed confidentially for an IPO on 1 June and is reported to be preparing a listing before November; OpenAI filed a week later, and its chief financial officer has told staff it will be public by 2027. As these names list, the most sought-after assets in private markets move into public ones.
The capital built to buy them does not disappear. Secondaries fundraising reached a record of around $166 billion in 2025, 48% above the previous peak. Evergreen funds now hold around $120 billion at fair value, 46% of it in secondaries, and more than half of secondary buyers run evergreen vehicles that must deploy continuously. New channels have opened for private investors, from Robinhood’s listed venture funds to dedicated pre-IPO vehicles. At the top of the market, demand has run ahead of supply: OpenAI’s latest employee tender sold only $6.6 billion of a $10.3 billion cap, because employees chose to hold.
That capital has to move down the list, and the early evidence suggests it is starting to. Jefferies expects growing demand for venture and growth secondaries, including in mid-stage AI companies. PitchBook’s list of the next large secondary trades now reads Stripe, Databricks, Anduril, Shield AI, Epic, Discord, Perplexity, Crusoe, Kraken and Polymarket. Company-run tender offers on Carta rose 200% by value year on year in the first half of 2026, at a median discount of zero.
This is where a new DPI window opens for earlier funds. The natural sellers of the next tier are the seed and Series A funds that hold those companies at a fraction of today’s price, many of them from vintages whose distributions have lagged; on PitchBook’s data, 2021-vintage venture funds have so far returned an average of 0.05x DPI. A buyer paying a modest discount to the last round is still paying many multiples of an early fund’s cost. A discount to the last round is not a discount to cost.
The window is forming, not yet open. Venture-dedicated secondary dry powder is only around $10–12 billion, with most of the record fundraising aimed at buyouts. Pricing softened over the summer, with the median Forge trade moving from par in June to a 7% discount in July, and only a fifth of eligible SpaceX stock has so far been released from lock-up. As our March piece warned, the post-IPO secondary market will offer less proven companies and less competitive bidding. That is a risk for buyers. For early funds holding the next winners at low cost, it is an opportunity to crystallise returns earlier than the IPO calendar would allow.
The next wave of DPI for early-stage funds will not come from the IPOs that dominate the headlines. It will come from the secondary capital those IPOs set free.
NOT EVERY CROWN IS EARNED
King-making works when the capital follows the business. It fails when the business follows the capital.
The strongest counterargument to our own position deserves a fair hearing. If early funds benefit from later rounds, they also inherit the pricing of those rounds, and not all of it rests on fundamentals. As we documented in “Venture Is No Longer Setting Its Own Price”, a meaningful share of capital at the top of the market is circular: Nvidia and Microsoft invested in Anthropic, and Anthropic committed to buy Azure capacity and Nvidia’s next generation of chips. The investment is, in part, the purchase order.
Public markets have already tested some of those assumptions. SpaceX traded more than 30% below its debut level by late July before recovering to its IPO price in August, and Cerebras was down around 35% from its debut within weeks of listing. The last time capital crowned companies this quickly, in 2021, fewer than 30% of that year’s unicorns raised again within three years, and half of those that did accepted down rounds.
The distinction that matters is between a crown backed by the business and a crown backed by the balance sheet of the buyer. In “‘Uberization’ of AI Models”, the durable winners were those embedding themselves in customer workflows: more than 1,000 companies now spend over $1 million a year on Claude. That is a crown earned from customers rather than conferred by capital, and it is the test we apply before treating any mark-up as a signal.
WHAT AI IS DOING TO PRIVATE EQUITY
a16z measures venture against private equity’s high-water mark. The more important comparison is what AI is doing to private equity’s largest sector.
The same technology producing venture’s power law is repricing the largest sector in buyout portfolios. Software accounted for around 18% of US private equity deal value in 2025, against a ten-year average of 14%; the industry leaned in just before the ground moved. In February, the launch of Anthropic’s Claude Cowork plugins for legal, finance and data work took the IGV software index 36.6% below its peak by April.
Public markets have since recovered most of that loss, and the index now sits around 10% below its high. But the recovery has been selective. By July, Snowflake was up 21% for the year while Salesforce and ServiceNow were down 35% and 27%. The market has stopped pricing software as a single asset class and started pricing winners and losers, the same sorting we see in venture. The losers look a great deal like the typical buyout portfolio company: seat-based, mid-market and leveraged.
Private markets have not recovered, because they never fully repriced. Private equity software valuations fell by around 8% in the first quarter, against roughly 30% in public markets. The credit market tells the truer story. At the end of August, after the equity rebound, software loans still traded at 87.7 cents on the dollar against 97.1 for every other sector, the only sector in the leveraged loan index with a negative return this year. US software buyout deal value fell 65.7% year on year in the second quarter, and more than a quarter of private equity-owned software companies have now been held for five years or more. The economics explain why: as we showed in “‘Uberization’ of AI Models”, AI-first applications operate at gross margins of around 50–60%, against the 80–90% that underwrote SaaS multiples.
The exposure is not limited to software. IT services and business process outsourcing look like services businesses, but their value proposition is labour arbitrage, precisely what AI agents automate first. Accenture fell around 18% in a single day in June after cutting its growth guidance, and Concentrix cut its revenue forecast, with one analyst noting that some clients had “simply withdrawn customer support altogether”. The picture is not uniform, but where labour is the product, the savings AI generates accrue to the client, not the provider.
The more insulated businesses share a common trait: their value sits in physical assets, regulated delivery or on-site work that AI augments rather than replaces. Bain notes that buyers now favour businesses with “physical or labor-intensive components”. Private equity accounted for 72% of the $63 billion invested in US data centres in 2025; energy deal value reached $90 billion in the first half of 2026, the strongest pace in more than a decade; and aerospace and defence deal value rose to $11.3 billion in the first quarter from $7.7 billion a year earlier.
In venture, AI is the source of the power law. In private equity, it is the source of the markdown still to come.
WHAT THIS MEANS FOR ALLOCATORS
a16z is right about the destination. The fare, and the exit, are where returns are won.
For allocators, the implications are fourfold.
First, size the allocation for the economy we have. a16z is right that a 5–10% venture allocation was built for a cottage industry. For most institutional portfolios, underexposure is now the larger risk.
Second, build a barbell, not a line. Spread smaller, diversified commitments across the earliest stages, through funds with the ownership and access to be on the cap table before the crowning round, and concentrate larger growth-stage exposure only alongside the leading firms, in companies that have already been crowned. The segment to avoid is the middle: growth-stage prices without either the early fare or the certainty of a winner.
Third, underwrite the exit, not just the entry. As the largest names list and secondary capital moves down the list, the managers that deliver DPI will be those willing and able to sell part of a winner into that bid, through tenders, secondaries and continuation vehicles, rather than waiting for an IPO. Liquidity is becoming a manager skill.
Fourth, underwrite private equity by AI exposure, sector by sector. Labour-arbitrage services deserve the same scrutiny as seat-based software, and physical, regulated and on-site businesses carry a genuinely different risk profile.
CONCLUSION
a16z’s essay is right about the most important thing. Venture is no longer a niche allocation, and the power law has become the shape of the economy.
But agreeing on the destination is not the same as agreeing on the fare. For the majority of venture dollars, entering at growth-stage prices, only the category winners compound. For the smaller pool of capital that boards early, a $1–10 billion outcome held with real ownership still produces returns most growth funds cannot match, and the crowning rounds that follow are the mark-up, not the threat. As the largest private companies list, the secondary capital built to own them will move down the list, and the funds holding the next generation of winners at an early-stage basis will be its natural sellers.
The prudent seat on the bus is either at the front, alongside the firms crowning today’s winners, or at the first stop, before the fare rises. It is not in the middle, paying growth-stage prices for a seat that is neither early nor certain.
SOURCES & NOTES
- Andreessen Horowitz (Jen Kha), “Catching the (Venture) Bus,” 11 September 2026: SpaceX, Anthropic and OpenAI valuations; 5–10% allocation norm; PitchBook screen of 2,143 funds (2000–2018 vintages); time between unicorn rounds. a16z.com/catching-the-venture-bus/
- PitchBook-NVCA Venture Monitor, Q2 2026: H1 2026 US exit value $2.187 trillion (SpaceX $1.766 trillion); a16z, Thrive and Founders Fund 48.1% of US venture capital raised. NVCA 2026 Yearbook: 2025 exit value $217.1 billion. PitchBook-NVCA Q1 2026: AI step-ups. CNBC (12 June 2026): SpaceX IPO pricing. PitchBook (July 2026): post-IPO performance of SpaceX and Cerebras. pitchbook.com/news/reports/q2-2026-pitchbook-nvca-venture-monitor
- Morgan Stanley Investment Management, “Public to Private Equity in the United States: A Long-Term Look,” September 2026: listed company counts; 945 unicorns. Forbes (September 2026): Magnificent Seven weight in the S&P 500. morganstanley.com/content/dam/im/assets/publication/thought-leaders…
- Cambridge Associates, US PE/VC Benchmark Commentary, calendar year 2025. WRAL (23 September 2026): UNC endowment FY2026 return. cambridgeassociates.com/insight/us-pe-vc-benchmark-commentary-calen…
- Reference Capital, Venture Pulse, “Born Winners: The Age of Day-Zero Unicorns,” February 2026: capital concentration in top deals and funds; worked growth-stage fund-math example. referencecap.com/reports/born-winners-the-age-of-day-zero-unicorns
- Crunchbase News: 2025 North American funding by stage; Q1 2026 global funding by stage and company; early-stage unicorn counts. Dealroom, “Power law outcomes”: companies at $100 billion+. news.crunchbase.com/venture/north-american-startup-funding-2025-dat…
- CNBC (24 February 2026) and Axios (27 February 2025): Stripe tender offer valuations. TechCrunch (23 June 2026): Menlo Ventures’ Anthropic stake; (10 December 2024): Tiger Global PIP 15. cnbc.com/2026/02/24/stripe-value-stock-sale-tender-offer.html
- The Next Web (June 2026): Founders Fund’s SpaceX investment and value at IPO. TechCrunch (17 December 2007): Founders Fund II, $220 million. thenextweb.com/news/founders-fund-spacex-50-billion-windfall
- Anthropic funding rounds, company announcements and press coverage: Series A (2021, ~$0.6 billion), Series C (2023, $4.1 billion), Series D (2024, $18.4 billion), Series E (2025, $61.5 billion), Series F (2025, $183 billion), Series G (2026, ~$380 billion), Series H (May 2026, $965 billion). Multiples are pre-dilution and illustrative.
- Santé Ventures analysis of PitchBook data, via Institutional Investor (December 2023): TVPI by fund size. Forbes Finance Council (June 2026): economics of new capital in oversubscribed mega-rounds. institutionalinvestor.com/article/2cjotgawx7nn70st06k8w/portfolio/b…
- Reference Capital, Venture Pulse, “The Rise of Venture Capital Secondaries,” March 2026. referencecap.com/reports/the-rise-of-venture-capital-s…
- PitchBook, Q2 2026 US VC Secondary Market Watch; Hiive via PitchBook (August 2026): Q2 2026 concentration and next large secondary trades; Augment: Q4 2025 concentration; Forge Global (August 2026): trade pricing; PitchBook (August 2026): 2021-vintage DPI and tender offers. pitchbook.com/news/reports/q2-2026-us-vc-secondary-market-watch
- Anthropic, confidential draft S-1 announcement (1 June 2026); Reuters (September 2026): listing timing; CNBC (June and August 2026): OpenAI filing and timing. anthropic.com/news/confidential-draft-s1-sec
- Secondaries Investor: 2025 fundraising. Jefferies, Global Secondary Market Review, July 2026: evergreen funds and demand outlook. Evercore, H1 2026 Secondary Market Review. Carta, Tender Offer Update, H1 2026. jefferies.com/wp-content/uploads/sites/4/2026/07/Jefferies-Global-S…
- Reference Capital, Venture Pulse, “Venture Is No Longer Setting Its Own Price,” June 2026: circular financing; 2021 unicorn outcomes. referencecap.com/reports/venture-is-no-longer-setting-its-own-price
- Reference Capital, Venture Pulse, “‘Uberization’ of AI Models,” April 2026: AI application gross margins; Claude enterprise customers. referencecap.com/reports/uberization-of-ai-models
- PitchBook (February and July 2026): software share of US PE deal value and Q2 software deal value. Bespoke (28 August 2026) and StockAnalysis (September 2026): IGV drawdown and recovery. 24/7 Wall St (6 July 2026): dispersion across software names. Bain & Company, Private Equity Midyear Report 2026: private software marks. PitchBook (2 September 2026): software loan prices and returns. S&P Global Market Intelligence (May 2026): hold periods for PE-owned software. pitchbook.com/news/articles/august-wrap-loans-return-0-93-buoyed-by…
- The Motley Fool (18 June 2026): Accenture; Investing.com (30 June 2026): Concentrix and RBC comment; S&P Global Market Intelligence (May 2026): US data centre investment; Foley & Lardner citing KPMG (September 2026): energy; PitchBook via Penn Mutual (July 2026): aerospace and defence. spglobal.com/market-intelligence/en/news-insights/articles/2026/5/p…
- Goldman Sachs Global Investment Research, “The Potentially Large Effects of Artificial Intelligence on Economic Growth,” March 2023.
- Reference Capital fund-of-funds portfolio data, September 2026.
