Cheap to Build, Hard to Fund

A data-backed read on why AI collapsed the cost of building a company and raised the bar to finance one, and why, in a market this crowded, capital increasingly moves toward whoever makes the most persuasive argument.

Overview

Building a company has rarely been cheaper or faster than it is in 2026. Raising money for one has rarely been harder. AI drove down the cost of producing a working product and a polished pitch, and in doing so it multiplied the number of products and pitches competing for the same investors. The money itself has not dried up; the first half of 2026 was the largest on record. It has concentrated. Capital is pooling around a narrow set of companies with demand that lasts, and passing over the rest.

The headline figures look like a boom. Global venture funding reached a record $510 billion in the first half of 2026, more than the $440 billion invested in all of 2025, per Crunchbase. The distribution is where it gets uneven. More than 70% of second-quarter money went to AI companies, up from just under 50% a year earlier, and two of them, OpenAI and Anthropic, took $217 billion between them, 43% of everything raised worldwide in the half. There is more capital around than there has ever been, and it is reaching fewer companies.

This report covers what changed on the build side, the capital environment founders are now raising into, the rising bar to get funded, the durability problem that separates fundable revenue from the kind that churns, and what a raise actually runs on once the demo stops being the differentiator.

The Collapse in the Cost to Build

The clearest read on how far the cost of building has fallen comes from Y Combinator. CEO Garry Tan said that for roughly a quarter of the Winter 2025 batch, about 95% of the code was written by AI, with some founders reaching as much as $10 million in revenue on teams of fewer than ten people. YC called the cohort the fastest-growing in the fund's history, aggregating roughly 10% week-over-week growth.

The market's fastest-scaling tools tell the same story. Anysphere's Cursor, an AI coding tool, went from $300 million in annualized revenue in April 2025 to more than $500 million by June, with revenue doubling roughly every two months along the way. The behavior is mainstream now. A Gusto survey found about 60% of new U.S. business owners used AI to launch in 2025, double the rate of two years earlier, and that AI-using founders were twice as likely to secure venture or angel funding.

The benchmarks investors use to judge traction have moved with it. Andreessen Horowitz reported that the median enterprise AI startup now reaches more than $2 million in annualized revenue in its first year, against an old best-in-class bar of $1 million, and raises a Series A about nine months after it starts charging. Fast building used to signal a strong team. Now that most teams can do it, it has stopped signaling much at all.

One caution sits under the optimism. A randomized controlled trial by METR in July 2025 found experienced open-source developers were about 19% slower on real tasks when using AI tools, while believing they were roughly 20% faster. The study reflects early-2025 tools, and the firm has since noted newer ones may not show the same drag; the lasting point is narrower. Cheap to build does not guarantee built well, and a team convinced it is moving faster while it is measurably moving slower is exactly the kind of risk a careful investor has learned to discount.

The Capital Environment

The money is there, and it is pooling. Investors put $305 billion to work in the first quarter of 2026 and another $205 billion in the second, per Crunchbase, and more than 70% of the second-quarter total went to AI. Concentration is the defining feature. OpenAI and Anthropic together raised $217 billion in the half, 43% of all venture funding worldwide, and Anthropic's $65 billion round alone was close to a third of everything invested in the second quarter. Sixteen companies raised billion-dollar rounds in that quarter, taking $108.6 billion between them, 53% of its funding. The megaround has stopped being an event and become the market.

Pricing for the rounds that do get done has climbed. Per Carta, the median Series A post-money valuation reached about $78.7 million in the fourth quarter of 2025, up 37% year over year, with seed post-money at an all-time high near $24 million. The lift is uneven, though. The premium investors once paid for the AI label by itself is narrowing as the category fills up, and diligence is shifting toward whether the demand under the label holds. What clears the bar is moving from "uses AI" toward "has demand that defends itself."

The same concentration that lifts the averages thins out the middle. Carta counted 966 venture-backed shutdowns in 2024, up about 26% year over year, and the correction has since moved up the stack. SimpleClosure's 2025 shutdown data shows Series A companies jumping from about 6% to 14% of all closures, a 2.5x increase in a year, with AI companies near 16% of the total and thin application-layer products facing the sharpest correction. A record total at the top and rising mortality among funded companies describe one market, not two. Money is collecting where the demand looks durable and draining away from where it does not.

The Bar to Raise – Signal Versus Noise

Cheaper building lowered the barrier to ship, and it lowered the barrier to apply. Gusto found that about 75% of AI-using founders used AI to develop the business idea itself, not only the product. The result is a flood of similar-looking companies arriving at the same investors with similar-sounding traction. Once anyone can produce a clean demo and a tidy deck, the demo and the deck carry less information than they used to.

Investors have responded by sharpening their filters. The recurring phrase across the market in 2025 and 2026 is the "wrapper problem" – a thin layer over a foundation model is easy to replicate and hard to fund on its own. Darren Mowry, who runs Google's global startup organization, put it plainly in February, saying that if the back-end model is doing all the work, the industry has run out of patience for it. What clears the bar instead is some mix of proprietary first-party data, non-public integrations, deep workflow embedding, regulated-domain compliance, and a real distribution advantage. The revenue bar has moved with it – Andreessen Horowitz now describes the old best-in-class ramp to $1 million as the lower end of the growth it sees. Early revenue is no longer proof of product-market fit; it is where diligence begins. The work that used to come after the product, articulating why this company, against which alternative, on what evidence, and in what order, has become the work that decides the round.

The Durability Problem

The hardest lesson of the AI funding cycle is that revenue and durable revenue are not the same asset. RevenueCat's 2026 subscription-app analysis found that AI apps earn about 41% more revenue per paying user but churn roughly 30% faster, with annual retention of 21.1% against 30.7% for non-AI apps, and refund rates about 20% higher. They are easy to acquire into and hard to hold onto.

The same pattern shows up beyond consumer apps. On the enterprise side, the most-cited finding of the year, MIT's report that about 95% of enterprise generative-AI pilots delivered no measurable profit-and-loss impact, is a warning about demand quality as much as about adoption. A pilot that never converts into durable spend will not anchor a fundable revenue line.

For a founder, the implication is plain. Investors are no longer underwriting whether a company can reach first revenue; AI made that cheap and common. They are underwriting whether the revenue stays and grows once the novelty wears off. Retention has become the part of the story a demo cannot fake.

The Scoreboard – What Raises Versus What Stalls

A high-altitude view of the signals separating the companies clearing the bar in 2026 from the ones being passed over. The dividing line is the same one running through the capital data – demand that is hard to copy against a demo that is easy to copy.

Signal	Standing	Why capital responds this way
Proprietary data / workflow depth	Funds	Hard to replicate; the moat investors now screen for first.
Strong retention and expansion (NRR)	Funds	Demand that lasts; the revenue survives past the pilot.
Distribution advantage	Funds	Answers the standing-out problem cheaper building created.
Regulated-domain / compliance edge	Funds	SOC 2, HIPAA and GDPR act as a barrier to entry.
First $1M revenue, no expansion	Repricing	No longer read as product-market fit; the start of diligence.
AI label without a moat	Fading	Investors now discount the label itself and re-price on substance.
Thin model wrapper	Stalls	Easily replicated; the textbook un-fundable profile.

The Bottom Line

AI did not shrink venture capital so much as redistribute it. Once a working product and a clean pitch stopped being scarce, money moved toward the thing that still is – demand that holds up under competition and keeps growing after the launch buzz fades. The record fundraising totals and a failure line that has climbed from seed into funded Series A companies are two readings of one market, taken at different altitudes.

The question that decides a raise in 2026 is no longer whether the thing can be built. Almost anyone can build it. It is whether what was built has a reason to keep getting paid for once the novelty wears off, and whether the pitch can show that with evidence rather than adjectives. When capital concentrates, the margin between a round that comes together and one that stalls narrows to the quality of the argument made for it.

What a Raise Runs On

A crowded, selective market produces a short list of recurring moments for a founder. Each is settled less by the product itself than by how it is argued to the investor, the partner, or the buyer on the other side. The materials are where that argument lands.

Raising the round. The investor who used to be impressed by a clean deck now sees a hundred of them. Standing out means an investor narrative and pitch deck built on the argument, a financial model an investor will trust, and a data room that survives diligence.

Proving the revenue is durable. The retention and expansion story that now decides the round has to be shown, not asserted – cohort economics, the model behind them, and the metrics narrative that frames them honestly.

The next conversation. Partnerships, business development, and an eventual exit each run on their own materials, from a one-pager that opens a door to a confidential information memorandum and teaser when it is time to sell.

Generative AI has made producing any of these documents nearly free, which is exactly why the bar has risen. When everyone can generate a clean deck and a tidy memo, the document itself carries less information, and the investor looks past it to the argument underneath. What AI has not made cheap is the judgment to know that an argument is aimed at the wrong audience, built on the wrong comparison, or ordered in the wrong way. In a market that reprices on the strength of that argument, the judgment is the product.