Why most memecoins go to zero — the data

Across 18.67 million pump.fun tokens launched between 14 January 2024 and 18 June 2026, 68.67% — 12.8 million of them — recorded their last trade on the day they were created. Around 80.37% were dead within two days. Only 4.55%, roughly 850,000 tokens, survived past 90 days. Those are CoinGecko's figures, updated 23 June 2026.

Most memecoins do not go to zero the way a failing company does, through a long visible decline. They go to zero by never having a second day. The modal outcome of a token launch is not a rug pull, a scandal or a collapse. It is silence within hours, and the data now exists to say so at a scale that makes argument difficult.

The survival curve, and the caveat CoinGecko attaches to it

Source: CoinGecko research, updated 23 June 2026, covering pump.fun tokens launched 14 January 2024 to 18 June 2026.

The caveat is CoinGecko's own and it cuts against the headline, which is why it belongs in the third paragraph rather than a footnote: they track bonding-curve trades only. A token that graduated to a decentralised exchange and carried on trading there can look silent on the curve while still being alive elsewhere. CoinGecko therefore states that the 90-day survival figure is understated. The true share of tokens still trading after three months is higher than 4.55% — nobody has published how much higher, and we are not going to guess.

That correction matters far less than it sounds like it should, because it barely touches the number that carries the argument. The 68.67% same-day figure is not sensitive to where a token trades later, since a token that graduates and moves on is by definition one that traded actively enough to get there. Two out of three tokens never reach a second day of trading on the venue that created them. Whatever the exact 90-day tail turns out to be, the front of the distribution is where almost all the mass sits.

What "goes to zero" actually means here

The phrase is used loosely, and the looseness hides three different outcomes.

The first is the one the data above measures: a token stops trading. Nobody buys, nobody sells, the pair goes quiet. Its quoted price may still be some non-zero number, because the last trade printed a price and nothing has moved it since. This is not a price of zero. It is worse in practical terms, because a price with no volume behind it cannot be realised — there is no counterparty at any size.

The second is a token that still trades but at a small fraction of its peak, on liquidity too thin to exit at the quoted price. This is the common fate of tokens that did briefly work. The quote is real, the depth is not, and the gap between them is what people mean when they say they could not get out.

The third is an outright rug: liquidity pulled, supply minted, contract drained. This is the outcome that gets written about and the rarest of the three by a wide margin. Almost all capital lost in memecoins is lost to indifference rather than theft, which is unsatisfying, harder to write headlines about, and considerably more useful to know.

Graduation is a different question, and we will not give you one number

A token "graduating" — filling its bonding curve and migrating to a decentralised exchange — is the earliest checkpoint that separates a launch from a non-event. It is also the statistic most frequently mangled, and the mangling is nearly always the same mistake: quoting a rate without the observation window that produced it.

The most rigorous published figure comes from a preprint by Kamat, arXiv 2607.02823, revised 13 August 2026, covering 832,941 Solana launches across 34 days from 8 May to 10 June 2026. It reports a pooled graduation rate of 0.198%, with a 95% confidence interval of 0.189% to 0.208%. That number is sound and it is routinely misused, because the study observed each mint for roughly six minutes after launch and states explicitly that the result should be read as a fast-regime graduation rate and a lower bound on the true 24-hour rate. The author self-corrected the paper twice and withdrew a sensitivity analysis that had attempted to recover late graduations, after API verification found zero graduations in a 100-mint sample.

Other published figures sit well above it. Cryptopolitan reported approximately 1.15% on 19 February 2026 — and reported it explicitly as the highest level in over seven months, following pump.fun's cashback-token launch, so it is a peak being quoted elsewhere as a baseline. Solana Compass put it at fewer than 2% on 10 June 2026. A 1.4% figure circulates via Odaily and Bitget with no window or sample attached, which is why we treat it as uncertain rather than citing it as an equal.

The honest position is that the defensible range for a true 24-hour graduation rate in 2026 is roughly 0.5% to 2%, and that any single point estimate quoted without its measurement window is close to meaningless. The dispersion between 0.198% and 1.4% is overwhelmingly a definitional problem, not evidence that the market changed between February and June. The full set of published figures with their windows attached is in memecoin graduation rates by launchpad, and the structural reasons behind the low rate are in why so few memecoins graduate.

A general warning applies to every figure in this area: several rates in circulation trace back to articles that no longer resolve. If a graduation number cannot be followed to a live source with a stated sample and observation window, it should not be quoted, including by anything citing this page.

The market-cap arc: concentration, not breadth

Aggregate value tells a different story from launch counts, and the two are often confused.

Memecoins as a category peaked at $150.6bn in December 2024, against $88.0bn at the 2021 cycle top in October 2021, and stood at $47.2bn in November 2025 — a drawdown of roughly 69% from the peak (CoinGecko, State of Memecoins 2025). That is a real contraction, but it describes prices, not survival, and the two move for different reasons.

The composition is the more revealing figure. As of that report, DOGE alone accounted for 47.3% of total memecoin market capitalisation, and dog-themed tokens excluding DOGE for a further 39.5%. Nearly 87% of the value in the category sits in one token plus its thematic relatives, almost all of which predate the current launchpad era by years. The 18.67 million tokens launched on a single venue since January 2024 collectively account for a small share of what the category is worth.

That is the shape of the thing. It is not a market where value is distributed thinly across many assets. It is a market where a handful of pre-existing assets hold nearly all the value and an enormous churn of new launches holds almost none.

The mechanics that produce a 68% same-day death rate

None of this requires bad actors to explain, though bad actors are present. Four structural features are sufficient.

There is no cash flow to fall back on. A memecoin has no revenue, no assets and no discount rate. Its price is entirely a function of what the next buyer will pay, which means there is no floor that arithmetic can establish. An equity that stops attracting buyers still owns things. A token that stops attracting buyers owns nothing, so attention is not one input into the price — it is the whole of it. When attention goes, there is nothing underneath.

Launching is nearly free and exiting is not. Creation on a bonding-curve venue costs effectively nothing; pump.fun's creation fee is zero and graduation costs on the order of 0.015 SOL. When the cost of producing a token rounds to zero, the equilibrium quantity is enormous, and roughly 42,000 launches a day on that one venue (10 June 2026) is what that equilibrium looks like. Attention, meanwhile, is fixed. Millions of tokens compete for the attention of a market that can only look at a few dozen at a time. Most receive none, and receiving none is fatal on day one. How the pricing curve itself works is covered in what is a bonding curve.

Liquidity is thin by construction until it isn't. A pre-graduation token's liquidity is the curve itself, seeded with virtual reserves. Depth grows only as buying grows. Early on, a small sell moves the price a long way, which means the first participants who want out impose most of the damage on everyone behind them. The moment liquidity moves to a real pool is its own hazard: what changes at migration and why price so often falls is set out in what happens at migration.

A meaningful share of launches are not attempts to build anything. Bundled launches, where the deployer takes a large share of supply through wallets that look unrelated, and sniper wallets buying in the first blocks, both transfer value from later buyers to earlier ones by design. The detection patterns are in bundled launches and sniper wallets. Some of the visible activity is not activity at all — manufactured volume is common enough that reading a volume figure without checking it against liquidity depth and holder growth is a mistake, which is the subject of how to spot fake volume and wash trading.

What actually shifted the odds in the data

The Kamat study is more interesting for its correlates than its headline rate, and both need reading carefully.

Launches that advertised a Telegram group graduated at 1.485%, against 0.166% for launches that did not — a differential of 8.94 times, with a Cox hazard ratio of 5.40. Initial market capitalisation above the venue's 30 SOL default was the strongest single predictor in the model, with a hazard ratio of 4.51.

Two things follow, and the second matters more than the first.

The first is that effort and capital at launch are visible in the outcome data. Tokens that arrive with a community channel and more than the default initial capital are several times more likely to clear the first checkpoint. That is a measurable association, not a demonstrated cause; both variables plausibly proxy for a launcher who has done more of everything, and neither was randomly assigned.

The second is the one worth carrying: an 8.94-times improvement on 0.166% is 1.485%. The best-performing observable cohort in a study of 832,941 launches still failed to graduate 98.5% of the time in the observation window. Multiplying a very small probability by a large factor produces a slightly less small probability. Presented as a lift, the Telegram finding sounds like a strategy. Presented as a level, it is a reminder that the base rate dominates everything.

The single most useful number in the whole dataset

pump.fun had processed roughly 11.9 million cumulative launches since January 2024 as of 10 June 2026. Of those, 18 tokens had ever exceeded a $10M market capitalisation, and 96 had ever exceeded $1M.

Ninety-six is not a rate that survives being expressed as a percentage in any meaningful way. It is a count small enough to list.

Note that this count and CoinGecko's 18.67 million do not reconcile — 11.9 million as of 10 June 2026 against 18.67 million as of 18 June 2026, eight days apart. The two come from different sources with different inclusion rules, and neither publishes its methodology in enough detail for us to reconcile them. We report both with their dates rather than picking the one that reads better, and the ratio implied by the second is not materially different from the first. How the venues themselves rose and fell across this period is the subject of the launchpad wars, 2024 to 2026, and the current per-venue figures sit in our analytics hub.

What this data does not tell you

It is not the whole market. The 18.67 million-token dataset is pump.fun, on Solana. That venue held 85.9% of Solana token issuance as of 8 February 2026, so it is a large and reasonable proxy, but it is one venue on one chain and other venues have different mechanics that plausibly produce different survival curves. Nobody has published an equivalent dataset for Robinhood Chain, Base or BNB.

It understates late survival. CoinGecko's bonding-curve-only tracking means the 4.55% figure is a floor, not an estimate. If you need the true 90-day survival rate, this dataset cannot give it to you.

"Dead" is defined as no trades, not as a price of zero. Tokens in the 68.67% still have contract addresses, quoted prices and, in many cases, holders. What they do not have is a market.

Graduation is not success. Clearing a bonding curve means a token reached a threshold in its first minutes or hours. It says nothing about whether it held value afterwards, and the migration event itself is one of the more reliable points at which price falls.

And none of it is predictive of any individual token. A base rate describes a population. It tells you what happens to launches in aggregate, which is genuinely useful for calibrating expectations, and it tells you nothing about which specific token does what. We are not in the business of saying which ones those are, and anyone who is should be read with that in mind.

Figures published by Meme Central from its own index describe tokens Meme Central has indexed, with a coverage window that varies by venue. Those are a sample, not whole-market data, and are labelled as such wherever they appear.

Frequently asked questions

What percentage of memecoins fail?

Across 18.67 million pump.fun tokens launched from 14 January 2024 to 18 June 2026, 68.67% recorded their last trade on their creation day and about 80.37% were dead within two days (CoinGecko, 23 June 2026). Only 4.55% were still trading after 90 days, and CoinGecko notes that figure is understated because it counts bonding-curve trades only.

Do memecoins actually go to zero, or just close to it?

Mostly neither, in the literal sense. The common outcome is a token that stops trading entirely while still showing a non-zero last price. There is a quote but no counterparty, which is functionally worse than a zero because it cannot be realised. Outright zeroes — liquidity pulled, contract drained — are real but far rarer than abandonment.

What percentage of memecoins graduate to a DEX?

There is no single defensible number. Published 2026 figures range from 0.198% (a six-minute observation window, explicitly a lower bound) to about 1.4% (undated). The reasonable range for a true 24-hour rate is roughly 0.5% to 2%, and any figure quoted without its measurement window should be treated as unusable.

Does having a Telegram group make a memecoin more likely to survive?

In the Kamat data, launches advertising a Telegram graduated at 1.485% versus 0.166% without — 8.94 times better. That is an association in observational data, not a proven cause, and the level matters more than the lift: 98.5% of the advertised cohort still did not graduate within the window.

Is the memecoin failure rate getting better or worse?

We cannot say, and neither can anyone else from published data. Market capitalisation fell about 69% from its $150.6bn December 2024 peak to $47.2bn in November 2025, but that measures price, not survival, and no comparable survival series exists for earlier periods. Anyone stating a trend in failure rates is extrapolating.

---

Lock the liquidity before anyone has to take your word for it

If you are on the launching side of these numbers, the one commitment a buyer can verify without trusting you is what happens to the pool. Team Finance — built by TrustSwap, which also builds Meme Central — locks LP tokens for a fixed term on Ethereum, Robinhood Chain, Polygon, Base and BNB, and the lock shows as a verified badge on the token's page in the Meme Central feed. It does not stop a creator selling their own allocation, and it does not move a token out of the 68% — nothing on this page suggests any mechanism does.

---

Nothing here is financial, legal or tax advice. Memecoins are extremely high-risk: most lose most of their value, and the majority of tokens launched never reach a decentralised exchange at all. Never spend money you cannot afford to lose entirely. Meme Central does not recommend any specific token. Data described as Meme Central's own reflects tokens indexed by Meme Central and is not whole-market data.