Demo Day Heat Check
An independent data study
Does Demo Day
hype predict success?
We verified 5,022 YC companies across 20 years against genuine, contemporaneous press coverage — not hindsight labels — to find out if Demo Day buzz predicts long-term outcomes.
It mostly doesn't.
1.04
Era-adjusted odds ratio
hot vs. unicorn odds — not significant
5,022
Companies analyzed
20 yrs, verified via research
2.5%
Actually become
unicorns ($1B+), full population
~19%
Ultimately
fail entirely (verified)
Not affiliated with Y Combinator. Independent research by @Trace_Cohen
Press hype doesn't pick unicorns
Odds ratio 1.05 after controlling for era + industry — not statistically significant. Most decacorns were never a Demo Day press pick.
But it does predict scale
Press-covered companies build meaningfully bigger teams, are more likely to reach growth stage, and raise ~1.3x more capital (N=930, replicated).
Power law dominates everything
A handful of extreme outliers generate most of the portfolio value — and hype can't tell you in advance which companies those will be.
The companies nobody wanted
are worth $735B+ today.
Airbnb was rejected 7+ times. DoorDash was bottom half of its batch. Coinbase couldn't fill a $1M round. Some of the companies investors fought over at Demo Day raised hundreds of millions and are now worth $0 — but on average, across 5,022 verified companies, press coverage barely moves the odds once you adjust for the era each batch launched in.
The Sleepers
$0B+
Combined valuation of 107 companies that were overlooked or struggled at Demo Day
The Darlings That Died
$0
22 companies that were hyped at Demo Day but raised $1.5B+ and failed
Top 25 most valuable
Orange = sleepers nobody wanted. Green = hot from day 1. Blue = moderate. Red = hot but failed.
242 notable companies
An illustrative subset of the full 5,022-company verified dataset, labeled using real hot/unicorn signal. Search by name, industry, or batch. Filter by category.
Outcome status — all 5,154 companies
Full verified population, not just the illustrative subset below
| Stripe | S09 | $159B |
| Airbnb | W09 | $84B |
| DoorDash | S13 | $72B |
| Coinbase | S12 | $43B |
| S05 | $33B | |
| Scale AI | S16 | $29B |
| Rippling | W17 | $20.8B |
| Deel | W19 | $17.3B |
| Helion Energy | W14 | $15.5B |
| Whatnot | W20 | $11.5B |
| Supabase | S20 | $10.5B |
| Instacart | S12 | $10B |
| Gusto | W12 | $9.5B |
| Razorpay | W15 | $9.2B |
| Meesho | S16 | $9B |
| Groww | W18 | $9B |
| Replit | W18 | $9B |
| Flock Safety | S17 | $8.4B |
| Zepto | W21 | $7B |
| Athelas | S16 | $7B |
| Dropbox | S07 | $6.4B |
| Oklo | S14 | $6.16B |
| Rigetti Computing | S14 | $6.16B |
| Benchling | S12 | $6.1B |
| EquipmentShare | W15 | $5.7B |
| Fivetran | W13 | $5.6B |
| Legora | W24 | $5.6B |
| Faire | W17 | $5.2B |
| Rappi | W16 | $5.2B |
| Mercury | S19 | $5.2B |
| Brex | W17 | $5.15B |
| Zapier | S12 | $5B |
| GitLab | W15 | $4.7B |
| GrubMarket | W15 | $4.5B |
| Relativity Space | S16 | $4.2B |
| Vanta | W18 | $4.15B |
| Webflow | S13 | $4B |
| GOAT Group | W11 | $3.7B |
| Flutterwave | S16 | $3.25B |
| Segment | S11 | $3.2B |
| Retool | W17 | $3.2B |
| Ironclad | S15 | $3.2B |
| Podium | W16 | $3B |
| Lattice | W16 | $3B |
| Outschool | W16 | $3B |
| Corgi Insurance | S24 | $2.6B |
| Honeylove | S18 | $2.3B |
| Algolia | W14 | $2.25B |
| Zip | S20 | $2.2B |
| Docker | S10 | $2.1B |
Nobody wanted them.
Rejected at Demo Day. Now worth billions.
First customer in 2 weeks via YC network. Sequoia early. Still private at $159B.
No genuine "top picks" press coverage survives for W09's Demo Day at all — only neutral, complete rosters with no evaluative framing. Sold cereal boxes to survive early on.
Bottom half of batch at Demo Day. Nearly died in 2017–18.
Only $320K committed when pitching a $1M seed round.
19 years from YC to IPO. Longest timeline of any YC company.
19-year-old founder. Pivoted during batch. Raised Series A in 2017.
+ 101 more sleepers in the full directory above
Everyone fought over them.
Hottest at Demo Day. VCs piled in. Never hit unicorn scale — status varies below.
TechCrunch top 8. $2M run rate. 112% weekly growth. Dead by 2016.
TechCrunch top 8. Bitcoin exchange. Dead by April 2015.
TechCrunch top 8. Banking API. Acqui-hired by SVB.
TechCrunch top 8. 20K devs. Acqui-hired by Palantir.
TechCrunch top 10. 40% job placement rate. Shut down ~2022.
Mark Zuckerberg's first angel investment. K-12 data analytics.
+ 16 more in the full directory
Three batches where
the pundits missed it.
In these three illustrative batches, TechCrunch's “Top Picks” missed the biggest eventual winner — but across all 5,022 verified companies, press coverage doesn't reliably predict outcomes in either direction once era is controlled for.
S13
2013top picks failed or acqui-hired
BIGGEST WINNER NOT PICKED
DoorDash ($71B IPO)
SpoonRocket, Buttercoin, Standard Treasury
W14
2014top picks failed or acqui-hired
BIGGEST WINNER NOT PICKED
Cruise ($1B exit)
Dating Ring (#2), Kimono Labs, Boostable, BatteryOS…
W19
2019top picks failed or acqui-hired
BIGGEST WINNER NOT PICKED
Deel ($17.3B)
Flockjay, Docucharm, Parker (bankrupt)
5 companies = 72% of all value
YC's returns follow an extreme power law. Airbnb, Coinbase, DoorDash, Instacart, and Rigetti Computing account for nearly three-quarters of all verified YC portfolio value — and none of them were named as Demo Day press picks.
of YC companies are decacorns
10 of 5,154 verified companies, >$10B
become unicorns at all
136 of 5,154 verified companies
companies confirmed inactive
20% of the full verified population
of the top 5 had no Demo Day press pick
Coinbase, DoorDash, Instacart, Rigetti Computing confirmed overlooked — Airbnb's batch predates available press coverage
Category Breakdown
Every one of these 243 companies has been individually cross-checked against YC's own directory — 22 companies previously in this dataset were removed after failing to verify as actual YC alumni.
107
Sleepers that won
$735B total
10
Hot & won
$56B total
104
Moderate & won
$26B total
22
Hot & failed
$0B total
How fast did they scale?
Press coverage is one proxy for Demo Day heat, but not the only one — how fast a company closes its next round, and which investors show up, matter too. We don't have free, comprehensive data on round oversubscription or investor prestige across 5,000+ companies, but we can measure speed: time from founding to unicorn status. The fastest got there in 1–2 years. Some took over a decade. Many never made it.
6.0y
Avg. years to unicorn
1.9y
Avg. seed → Series A
120
Reached unicorn status
29
Completely dead
Fastest to Unicorn
Years from founding to $1B+ valuation
Time to Unicorn Distribution
Current Status
243 companies tracked
The analytics
Total value created by YC batch year. Some vintages are worth 100x more than others.
Even YC can't predict winners
“
The hard part was predicting how tough and ambitious they would become. I learned to maintain a completely open mind about which startups in each batch would turn out to be the stars.
Paul Graham
YC Co-founder
“
About 25% of YC companies appear to be on a trajectory for billion-dollar valuations at the end of YC, but only a handful actually achieve it.
Sam Altman
Former YC President
“
Tar pit ideas get enthusiastic Demo Day reception but trap founders. They've been tried since the '90s. The real advice is simpler: just don't die.
Dalton Caldwell
YC Managing Director
“
There are no specific events that will automatically get money from an investor. I've seen founders cross $1M ARR and still not be able to raise.
Aaron Harris
Former YC Partner
We ran the regression.
Here's the receipt.
Logistic regression on unicorn outcome (binary) as a function of press coverage at Demo Day (hot), controlling for batch era and industry (press coverage skews ~2x toward Consumer companies, so industry composition alone could otherwise explain a spurious effect). Press coverage is one measurable proxy for early heat — not a complete score. Funding velocity and investor prestige likely matter too, but aren't available as free, comprehensive data across 5,000+ companies. N = 5,022 companies with verified contemporaneous press signal.
$ python3 logit_regression.py --dep unicorn --controls era,industry
Model specification
logit(P(unicorn)) = β₀ + β₁·hot + β₂·era₁₋₃ + β₃·industry₁₋₆
Coefficient on hot (statsmodels Logit summary)
coef = 0.0468 std err = 0.319 z = 0.147 P>|z| = 0.883
OR(hot) = exp(0.0468) = 1.048 95% CI [0.560, 1.960]
→ CI spans 1.0. Cannot reject the null. No significant effect.
Forest plot
Odds ratio for "hot" on each outcome, era + industry adjusted. Whiskers = 95% CI. Dashed line = no effect (OR=1).
| Dependent var. | β (hot) | OR | p-value |
|---|---|---|---|
| unicorn (raw) | 0.503 | 1.65 | 0.104 |
| unicorn (era + industry adj.) | 0.047 | 1.05 | 0.883 |
| failed / shut down (era + industry adj.) | -0.296 | 0.74 | 0.056 |
| IPO (era + industry adj.) | 0.322 | 1.38 | 0.559 |
| acquired (era + industry adj.) | 0.073 | 1.08 | 0.650 |
| reached Growth stage (era + industry adj.) | 0.381 | 1.46 | 0.006 |
| total funding raised, log (era adj., OLS, N=793) | 0.284 | 1.33 | 0.007 |
The real, significant findings
Press coverage doesn't predict who becomes a unicorn — but it does predict who scales and who raises. Press-covered companies have 23% bigger teams (p=0.009) and are 1.46x more likely to reach YC's "Growth" stage (p=0.006), both holding after controlling for batch era and industry composition. We also individually researched total funding raised for 930 companies (310 hot, 620 matched not-hot — a dataset that doesn't exist anywhere else) and found press-covered companies raise 1.43x more capital (median $3.0M vs $2.1M, p=0.0098 Mann-Whitney, N=792 with usable data). We replicated this by independently re-sampling the not-hot group — the two samples agreed (p=0.73, no sampling artifact) — but a high-confidence-only cut of the newer sample showed an attenuated, non-significant result, so treat the direction as solid and the exact multiplier as a rough estimate. Hype is a real, measurable signal for scale and fundraising — it just doesn't reliably predict the extreme, rare outcome of hitting $1B+.
Robustness: mature batches only
Restricting to batches ≤ 2019 (9+ years to mature): OR(unicorn) = 1.29, p = 0.42. Same conclusion — no reliable hype effect on unicorn odds specifically.
Sample construction
132 companies excluded — no surviving contemporaneous press exists for their batch (7 batches, mostly pre-2012). Excluded, not mislabeled.
Why R² is so low for unicorn odds
Startup outcomes are power-law distributed — a handful of decacorns dominate total value. Binary hype labels can't explain variance dominated by a few extreme outliers, even when they do explain scale.
The signal isn't in the sizzle.
It's in the scale — just not the outliers.
We verified 5,022 YC companies across 20 years against genuine, contemporaneous press coverage — not hindsight labels. The raw numbers show hot-covered companies with a modest edge on unicorn odds (odds ratio 1.65), but once we control for both batch era and industry composition (press skews heavily toward Consumer), that edge nearly vanishes (odds ratio 1.05, not statistically significant). Demo Day press coverage does not reliably predict which companies become unicorns. But it does predict something real: press-covered companies end up with meaningfully bigger teams, are significantly more likely to reach growth stage, and — based on individually researched, independently replicated funding data for 930 companies — raise about 1.3x more capital, all holding after controlling for era. Hype is a genuine scaling and fundraising signal — it just can't pick the handful of extreme, power-law outliers that end up dominating total portfolio value.
We're not the only ones counting.
Other independent analyses of YC's outcome data land at different numbers than ours. Here's the honest comparison, and why they likely diverge.
| Metric | This study | Other estimates |
|---|---|---|
| Unicorn rate | 2.6% (136 of 5,154) | 1.8% – 4.5%, depending on cohort maturity |
| Failure / shutdown rate | 20.1%, verified per company | 13%, self-reported estimate |
| Mature-cohort unicorn rate | 5.6% (batches ≤2019) | 5% – 6.5%, using paid valuation databases |
Once you control for cohort maturity, our number converges almost exactly with everyone else's: 5.6% of our pre-2020 batches became unicorns, right in the 5%–6.5% range other analyses report using paid valuation databases. The headline 2.6% figure looks lower mainly because our dataset runs all the way through 2025, including hundreds of companies still too young to have hit $1B yet — the same maturity-bias every other analysis has to correct for. Our failure rate is likely the most accurate number in this table, since it comes directly from YC's own status field verified individually per company, not a self-reported or modeled estimate. Every serious analysis lands on the same underlying conclusion regardless of the exact percentage: YC roughly doubles a startup's unconditional odds of reaching unicorn status versus the general venture-backed baseline.
Sources & Methodology
⬇ Download full dataset (5,154 companies, CSV)Y Combinator's public company directory (all 5,154 companies, batches S05–S25)
TechCrunch, VentureBeat & The Next Web contemporaneous Demo Day coverage (30+ batches)
Wayback Machine archival recovery of dead TechCrunch gallery pages
CB Insights & Crunchbase current unicorn lists (cross-referenced)
YC's own "top company" flagship designation (91 companies)
813 individually researched Public/Acquired company outcomes
930 companies individually researched for total funding raised (310 hot, 620 matched not-hot, replicated across two independent samples)
Rebel Fund: What Predicts YC Success (Jared Heyman)
Rebel Fund: Power Law of YC Startups
Palle Broe, "Pulling Back the Curtain on the Magic of YC" (Lenny's Newsletter, Feb 2025)
5,154 total companies; 5,022 analyzed where genuine contemporaneous press exists
Not affiliated with Y Combinator. “Hot” = named in genuine press coverage published at the time of that batch's Demo Day (not retrospective lists), recovered via live search and Wayback Machine archives. 132 companies from 7 batches with no surviving contemporaneous coverage (mostly pre-2012) were excluded from the hot/not-hot comparison rather than mislabeled. “Failed” = shut down per YC's own status field, individually verified.