Thirty venture figures checked against the organisation that produced them, five that circulate widely and do not hold up, and five more we could not trace past a blog.
This page exists because the same statistic kept turning up in different articles with different values, and in several cases with a source attached that had never published it. If you are writing about venture markets, quoting a figure in a deck, or deciding where to allocate, start here and follow the links to the original.
Status codes and the procedure behind them are set out in the FinanceBeef Data Standard.
Before anything else: there are three different “global totals”
Three credible providers publish a first-half 2026 global venture figure, and all three differ. This is the single most common way people end up quoting a number that contradicts the deck they are sitting in.
| Provider | H1 2026 | Scope | Status |
|---|---|---|---|
| Crunchbase | $510B | Global, Crunchbase inclusion rules | A |
| KPMG Venture Pulse | $560.3B | Global. Built on PitchBook data, includes some financings with partial debt | A |
| PitchBook-NVCA | over $400B | United States only | A |
Both: pick one provider for any series you are going to chart and name it every time you quote from it. If you need a second opinion, use Crunchbase or CB Insights rather than another PitchBook wrapper, and publish both numbers with their scope.
Verified: global funding
| Figure | Value | Status |
|---|---|---|
| Global venture funding, H1 2026 | $510B, a record half-year | A |
| Global venture funding, full year 2025 | $440B | A |
| OpenAI and Anthropic combined, H1 2026 | $217B, 43% of global startup funding | A |
| Anthropic alone, Q2 2026 | $65B, close to a third of global Q2 funding | A |
| AI share of global capital, Q2 2026 | over 70%, from just under 50% a year earlier | A |
| US share of global capital, Q2 2026 | two thirds, down from 83% in Q1 | A |
| Billion-dollar rounds, Q2 2026 | 16 companies, $108.6B, 53% of the quarter | A |
If you are allocating: two companies took 43% of global startup funding. Any market-wide average you are shown this year is being set by them. Judge a deal against its stage and sector, never against the global figure.
If you are raising: if you are not an AI company, the 70% share is the number that describes your market, not the record total. Plan the round against the remaining 30%.
Verified: seed
| Figure | Value | Status |
|---|---|---|
| Global seed funding, Q2 2026 | $12B | A |
| Seed rounds of $100M and above, Q2 2026 | $2.8B | A |
| Seed rounds of $10M and under, Q2 2026 | $5B | A |
| Seed to Series A within 24 months, 2022 cohort | about 17% | A |
| Same, a normal year (2018 cohort) | 25% to 30% | A |
If you are raising at seed: plan for 24 months of runway, not 18. The graduation rate says roughly five in six companies do not reach a Series A inside two years, and that was measured before deal count hit a decade low.
If you are allocating at seed: price the follow-on risk explicitly. A seed position in this market needs either reserves for a bridge or an explicit view on the company reaching profitability without a Series A.
Verified: fund economics and liquidity
| Figure | Value | Status |
|---|---|---|
| US VC fundraising, 2025 | $66.1B across 537 funds, lowest since 2018 | A |
| 2025 fund count against the 2021 peak | 30% | A |
| Andreessen Horowitz, January 2026 close | $15B, over 18% of all new US VC commitments since January 2025 | A |
| Net cash flows to limited partners since 2022 | almost $200B negative | A |
| Aggregate value of US unicorns | $4.3T | A |
| Concentration of 2025 venture dollars | half of all dollars into 0.05% of deals | A |
| AI share of US venture dollars, H1 2026 | $355.9B of $412.7B, about 86% | C |
If you are allocating: the $4.3 trillion figure is the one to act on. PitchBook’s own comment on it is that in the past many of these companies would have gone public, and that as they stay private, investors have shifted strategy to include private-market companies. If you want exposure to that growth, decide now whether your route is a fund, a secondary or a direct position, because waiting for the IPO is no longer a strategy.
If you are raising: almost $200 billion of negative net cash flow to LPs is why a partner who likes your company still says no. It is rarely about the deck. Ask where the fund is in its cycle before you read the room as a rejection.
Verified: cost of capital
| Figure | Value | Status |
|---|---|---|
| ECB deposit facility rate | 2.25% | A |
| Main refinancing operations rate | 2.40% | A |
| Marginal lending facility | 2.65% | A |
| Size of the move | 25 basis points, all three rates | A |
Both: if a European model in front of you still assumes the previous rate, every discounted figure in it is wrong and the venture debt line is understated. Check the effective date, not just the rate.
Verified: accelerators and venture studios
| Figure | Value | Status |
|---|---|---|
| Accelerator value added | most show negative value added against a no-accelerator benchmark; a small right tail generates large gains | A |
| Sample size | roughly 750,000 US startups, 329 accelerators | A |
| Selection effect | systematic: better ventures are more likely to enter, and to sort into higher value-added programs | A |
| Y Combinator batches per year, 2026 | four | A |
| Venture studios worldwide, September 2024 | 1,107 established | A |
| Studios that had ceased operating | 154 | A |
| Q3 2024 studio openings against closures | 17 new registrations against 20 closures, which the authors note would make 2024 the first year of net decline | A |
If you are allocating: a program logo tells you a company was selected, not that it was improved. The paper separates those two things and finds the selection effect is systematic. Ask what the company looked like on the day it applied, and judge the program by the right tail it sits in rather than by the category.
If you are raising: if you are paying for a program, ask for its own graduation and follow-on numbers against a comparable cohort that did not enter. Most cannot produce them, and that answer is itself the data.
Weakly sourced: use as indicative, not as evidence
Not necessarily wrong. We simply could not get past a secondary source.
| Claim | Circulating value | Problem | Status |
|---|---|---|---|
| Series A readiness benchmark | $3M to $5M ARR, MoM growth above 15% | Appears only in blog and advisory content. No primary dataset located | N |
| Y Combinator unicorn rate | about 4.5%, against about 2.5% for other venture-backed seed companies | Not traceable to a YC publication | N |
| YC Winter 2026 batch size | 196, 199 or 214 companies | Sources disagree, likely different treatment of stealth and dropped companies. Write “around 200” | C |
| European Series A round size | $8M to $12M, against $15M to $20M in the US | Traces to a single blog post | N |
| Bank lending tightening | net 26% of firms reporting tighter conditions | Attributed to the ECB Bank Lending Survey but read through a blog, not verified at source | C |
Both: you may use these in conversation. Do not put them in a document that someone will act on, and do not attach the implied source to them. If a figure matters enough to drive a decision, spend the ten minutes to trace it or drop it.
Does not hold up
This is the section we would most like other people to copy.
Both: search your own deck for any of these five. If one is there, remove it rather than softening it, because a hedged version of an unsourced number is still an unsourced number.
Then apply the pattern that catches all five: ask who benefits from the figure being true. Four of the five are an industry reporting on itself. That question alone would have stopped them.
How these were checked
Of the figures checked for this page, roughly one in three did not survive in the form it was circulating. That ratio is the most useful thing here.
The full procedure, the status codes above and the running corrections register are on the FinanceBeef Data Standard. The short version is seven steps: find the producer rather than the article quoting it, open the original, compare the exact wording, check the vintage rather than just the date, check who benefits, check the definition behind the label, and when two credible sources disagree publish both with their denominators.
Corrections and additions
If a figure here is wrong, out of date, or if you have a primary source for something in the weakly sourced section, tell us. Corrections are credited on the page and added to the register.
How to cite this page
Bianova, M. “Venture Data 2026: Which Numbers Check Out and Which Do Not.” FinanceBeef. Updated 10 September 2026.
Related
- Venture Capital’s $1.9 Trillion Exit Record Was Mostly a Valuation, Not Cash — why a $1.9 trillion exit quarter is mostly a valuation rather than a payment
- The FinanceBeef Data Standard — the five rules, the status codes and the corrections register
- Record Venture Funding 2026, Record Trouble Raising — the analysis these figures came from