The Closing Window

Every venture investment rests on a quiet assumption about time. An investor puts money in, and then there is an interval — months, usually years — during which the company generates the evidence that the investment was sound. Customers renew or they churn. Cohorts hold their value or decay. The margin proves durable or erodes. A competitor responds, or fails to. Growth turns out to be the front edge of a market or the exhaust of a promotion. The interval is not dead time; it is the period during which a bet becomes a fact, or fails to. Underwriting, properly understood, is the purchase of evidence with time. The investor pays at the start and learns across the span, and the markup — the higher price the next round will bear — is supposed to be earned by what the span revealed.

Speed is seductive to capital for a single reason: it creates the appearance of proof before proof can exist. A steep line accumulates faster than anything else a company can show — faster than a track record could ever form — and so it arrives first, the first evidence on the table when the price is being set. Because it arrives first and arrives alone, it is the easiest thing to underwrite and the most exciting thing to price. This is the same error in the family of errors these letters keep returning to — the market fixing on the one number that is easy to see and sweeping under the rug everything that is hard to. There it was recurring revenue, read without its margin. Here it is velocity, read without the evidence it has not had time to produce.

To see why that is an error and not merely a simplification, you have to separate two things the word "evidence" hides. Some evidence scales. You can observe a million users this quarter instead of waiting a year to accumulate a thousand; you can watch adoption, engagement, and top-line growth pile up as fast as the company can generate them, and a fast company genuinely generates more of this kind of evidence, faster, than a slow one ever did. But some evidence does not scale, because it is governed by a clock rather than by volume. Whether a customer renews at the third renewal takes three renewal cycles to learn, at any size. Whether a cohort retains its value over five years takes five years, whether the cohort is a thousand users or a hundred million. Whether a margin survives a full demand cycle takes a cycle. Whether a competitor can replicate the product takes however long the competitor takes. Whether pricing power holds when the discounting stops takes the time it takes. No rate of growth compresses these. They are functions of elapsed time, not of scale, and they are precisely the evidence that tells you whether a business is durable rather than merely large.

This is the distinction the fast company collapses, and the one the market collapses with it. A company can pile up a mountain of scale-governed evidence — users, bookings, a vertical revenue line — and reach the thresholds that trigger the largest valuations while possessing almost none of the clock-governed evidence those valuations are supposed to rest on, for the simple reason that it cannot exist yet. The markup arrives before the proof, not because anyone was careless, but because the only proof available at that speed is the proof that scales, and the proof that matters most is the proof that doesn't. Capital, reading the abundant fast evidence and finding it genuinely abundant, prices as though the other kind were present too. It mistakes a large quantity of the evidence that forms quickly for the presence of the evidence that forms slowly. The two are not substitutes. A million users this quarter tells you nothing about the five-year retention curve, and the five-year retention curve is the thing the valuation assumed.

What makes this dangerous rather than merely uncertain is that the speed and the blind spot are the same phenomenon. It is tempting to treat velocity and risk as independent — a fast company is simply a good company in a hurry, and the speed is a feature to admire while the risk is managed separately. But the speed is the source of the risk, because the speed is what prevents the slow evidence from forming. A company growing slowly enough to be watched reveals its durability, or its lack of it, on a schedule that lets capital reprice as it learns. A company growing too fast to be watched conceals nothing deliberately; the durability simply has not had time to demonstrate itself, in either direction. The investor is not fooled by the company. He is outrun by it.

History supplies the same lesson in larger letters. The companies of the late-1990s internet boom were financed and revalued on a tempo that left no room for the ordinary work of verification — pre-revenue businesses reaching the public markets on timelines that, in any prior era, would not have sufficed to establish that a business existed at all, priced on a story because the story was the only thing that had had time to form. Two decades later the growth-stage rounds of the 2021 cycle compressed the interval again: at the peak, late-stage valuations were more than doubling between rounds on an annualized basis, with median round-to-round step-ups reaching multiples not seen before, each markup priced off the last markup rather than off any accumulated evidence of durability — because the rounds were arriving faster than the durability could. When the cadence broke, the same companies found themselves raising again only after the interval had finally stretched back out, into flat rounds and down rounds at valuations the fast markups had never earned. The eras differ in their technology and their casualties. The mechanism is one mechanism. Speed buys the price and skips the proof.

The most recent cycle ran the experiment again at its limit. A category of companies reached billion-dollar valuations in a matter of months — by one accounting, several times faster than the already-fast norm of their sector — and reached eight-figure revenue at an average company age measured in low single-digit years. By the speed of their ascent they were the great successes of their moment, and the fast evidence was real: the users were real, the revenue was real, the growth was the steepest ever recorded. But whether the revenue would retain, whether the margin would survive contact with scale, whether the demand would outlast the novelty — none of that could be known, because knowing it requires a duration the companies had not yet lived through. The valuations were not wrong because the companies were bad. They were unverifiable because the companies were fast.

It is fair to object that some companies genuinely are extraordinary, that their speed reflects real and durable demand, and that an investor who refuses to move quickly will lose every good deal to someone who will. True, and it is the trap rather than the refutation. Many fast companies are superb, and the speed is real. The point is that the investor cannot tell which is which in the available time, because the evidence that would distinguish the durable fast company from the hollow one is exactly the evidence that only time produces, and the time has not passed. The fast good company and the fast hollow company look identical on the only axis that has had time to form — the scale axis — and identical precisely where it matters most, on the time axis, because neither has a history there yet. Paying up for speed does not resolve that ambiguity. It funds it.

There is a stronger objection: that this proves too much, that all venture investing is underwriting under uncertainty, and that demanding a full verification interval would mean never investing early at all. Correct, and the letter does not ask for certainty — early investing is the deliberate purchase of an unverified claim, and that is its honorable function. The argument is narrower. It is that the price should reflect how much of the clock-governed interval has actually elapsed, and in the cycles that matter it does the opposite: the less time a company has had to prove the things that can only be proven over time, the more the market pays, because the unproven trajectory is steeper and steeper trajectories command higher multiples. If evidence were what capital were buying, the relationship between velocity and price would be inverse — pay less for the company you have had less time to know on the axis that counts. The observed relationship is the reverse. The market pays the most for exactly what it has verified the least.

So a valuation placed on a very fast company is not a measure of what the company has proven. It is a measure of what its trajectory implies, priced as though the implication were already confirmed by evidence that cannot, at that speed, yet exist. In the ordinary case, where growth is slow enough to be watched, the gap between implication and confirmation closes on a schedule capital can act on. In the extraordinary case — the one the market finds most exciting and pays most for — the gap cannot close before the price is set, because the closing of it is a function of time and the time has not passed. Venture capital prices the trajectory and discovers, when the interval finally elapses, whether there was ever anything underneath it.

The correction is not to refuse speed. It is to read speed for what it is — abundant evidence of one kind, and silence on another — and to price the silence honestly. It means treating a closed verification window as a reason for a lower price and a smaller position, not a higher one, and refusing the cycle logic that turns velocity into its own justification. It means asking, of any company growing faster than it can be known, not "how steep is the line" but "how much of what this price assumes is the kind of thing that could possibly have happened yet" — and discounting hard for the answer. The investors who survive the fast cycles are not the ones who moved fastest. They are the ones who remembered which forms of proof are functions of time rather than scale, and refused to pay for the ones the clock had not yet had a chance to deliver.

Speed is the one thing capital can see at a glance, and durability is the one thing it cannot — and the faster the company, the wider the gap between them.

Sources & Methods. The argument is structural; the contemporary figures are illustrative and the historical parallels are offered as pattern, not as measured comparison. The empirical record the analysis is set against:

■ The velocity figures — CB Insights, AI Agent Bible (2025): the coding-AI category minting billion-dollar valuations in as little as six months, characterized as roughly four times faster than the sector norm, and reaching $10M+ revenue at an average company age of about 3.8 years. The six-month and step-change figures are consistent with independently verified company trajectories (e.g., Cursor/Anysphere ~$100M→$500M ARR, Jan–June 2025). Note: 3.8 years is the age of the companies, not any measure of executive or CRO tenure, and is not conflated with the latter.

■ The 2021 step-up cadence — PitchBook Q3 2021 US VC Valuations Report: median "relative velocity of value creation" (annualized round-to-round valuation growth) of ~115% YTD 2021, i.e., valuations more than doubling between rounds on an annualized basis, with median rolling late-stage step-up multiples reaching ~2.0x for the first time. Contrast (PitchBook, 2023–2024): that velocity collapsed to ~14.5% late-stage and ~2% at venture-growth, and median time between rounds stretched to ~1.76 years (longest in a decade), with flat/down rounds reaching a decade high in 2024 — the interval re-opening and re-pricing the markups the fast cadence had never earned.

■ The dot-com tempo — the compression of the formation-to-public-markets interval during the late-1990s internet cycle (pre-revenue and minimal-revenue companies reaching public markets on timelines that left no room for verification) is well documented in the historical record. No specific "median months to IPO" figure is asserted, as a clean, citable dot-com-era median was not established; the claim is deliberately qualitative. For directional support: NVCA-derived data shows average funding-to-exit rising from ~3.1 years (2001) to ~6.8 years (2014), and founding-to-IPO averaging ~11 years in modern samples — i.e., the verification interval is long and was radically compressed in the bubble.

■ Evidence volume vs. evidence duration — the central distinction (some evidence scales, some is governed by a clock — retention curves, multi-cycle margin durability, competitive response, renewal behavior, pricing power) is a structural argument about the nature of underwriting evidence, not an empirical claim requiring a dataset. Cambridge Associates and Correlation Ventures data on holding period, evidence accumulation, and outcome dispersion are the baseline for the premise that the evidence which discriminates outcomes is substantially time-governed.