Late-Stage Venture Returns Are Marks, Not Distributions

Key claims

• Per Cambridge Associates, late-stage venture funds posted the strongest returns of any venture stage in 2025, while paying the highest entry prices of the cycle.
• Per PitchBook, larger VC funds finished 2025 up 18%; funds under $250M stalled near 3%. Most of the gap came from access to a handful of AI megadeals driving markups: “marks, not distributions” (see DPI).
• By mid-2026 median pre-money valuations had passed their 2021 highs at every series. For AI companies at the venture-growth stage the median rose 204% in H1 2026 alone. Unlike 2021 (low rates lifting everything), one theme is carrying the late stage.
• LPs are leaning in anyway: through Q3 2026, funds of $500M+ took 78.1% of the $108.5B raised in US VC while being only 6% of new funds.
• The midterms as a stress test: broad indexes rallied after Democrats won both chambers (Dow up ~1,500 points the next day), but the AI supply chain fell. Neoclouds like CoreWeave and Nebius dropped more than 7%, as did several chipmakers, on expectations of slower data center construction.
• Several of the largest private AI companies were waiting for the election before listing; their IPOs will now be priced against a less predictable policy backdrop. Firmus pulling its IPO the month before showed how wide the private/public pricing gap can get.
• His conclusion: late stage has earned its recent returns, but the next leg depends less on company performance than on what public markets and Congress decide AI infrastructure is worth.

Notable quotes

Those are marks, not distributions.

In 2021, low rates lifted everything. This time, one theme is carrying the late stage.

Easier to underwrite isn’t the same as de-risked. When entry prices sit at a cycle peak and exits depend on a narrow IPO window, the margin for error shrinks.

If your late-stage gains were marked at peak valuations and the exit path just got more political, how are you haircutting them in your 2027 pacing?

Discussion

One reply at capture time, from Przemek G. (Sotapoc): paper gains vs. cash returned is the right split, and the same gap appears inside AI-heavy portfolio companies, where “the valuation reflects the AI story, while the P&L still shows the old cost base.” LPs will start asking for cash back and for evidence the AI spend changed margins.

How it connects

• DPI — the post’s core distinction: markups vs. cash returned.
• LP, Venture Capital — capital concentrating in $500M+ funds.
• Power Law — returns driven by access to a handful of megadeals.
• IPO — the narrowing exit window for private AI companies.
• Innovation tech doesn’t create… (tweet) — the same week’s argument over whether more venture capital grows the pie or competes returns away.