A Venture Curator analysis translated by TechFlowPost says a large share of U.S. unicorns are still carrying paper valuations set during the 2021-2022 funding boom, even though many of those prices have not been tested by the market since.
PitchBook’s Q3 2026 Quantitative Perspectives report, cited in the article, shows that among 964 active U.S. unicorns, 312 last completed a priced round in 2021 and another 225 did so in 2022. That leaves 537 companies, or about 56% of the total, still marked at values established during one of the most expensive periods in venture history.
More than half of active U.S. unicorns still sit on 2021-2022 pricing
The article says founders and early employees are often taught to treat the last round valuation as a scoreboard. It is the number on the cap table, the number in the press release, and the figure people use to estimate what their equity is worth. But newer market data suggests that, for many U.S. unicorns, those numbers have gone years without being checked by a fresh price-setting event.
PitchBook grouped active U.S. unicorns by the year of their most recent priced financing round. The result points to a backlog of companies still carrying marks from the bubble era. The same analysis notes that distributions from venture funds now equal 7.9% of net asset value, almost half the long-run average of 14.5%. PitchBook says returns have been dragged down directly by inflated valuations recorded between 2020 and 2022.
When companies do face a live market test, the article says many struggle to hold prior pricing. Public market investors have not been willing to pay private market prices. Chime is cited as one example, with its IPO valuation falling well below its private peak. In secondaries, buyers are also applying steeper discounts to companies whose last financing happened years ago than to businesses that raised more recently.
The piece says a second pattern sits beneath that. Many companies are not frozen because the business failed. They are staying private and staying quiet because any new financing or listing could force them to accept a lower number. As long as there is no new pricing event, a 2021 valuation can remain on paper. That, the article argues, is how the market can show both record paper value and record trapped value at the same time.
Its conclusion here is careful: a four-year-old valuation is not automatically false, but it is unverified. For founders and employees at companies that last raised in 2021 or 2022, the cap table valuation may simply be the most optimistic number their equity has ever had, not the amount the market would pay today. That matters for how a company approaches the next round, how employees think about exercising options, and how management explains what those options really mean. The same logic applies to funds reporting strong unrealized gains. Those marks are accounting figures, not cash returned.
The real risk, the article says, is not only that the valuation may be lower than it appears. It is that hiring, fundraising or option decisions get made as if a 2021 number were still today’s clearing price.
The AI premium is still there, but investors are slicing “AI-native” into different buckets
The analysis says calling a startup “AI-native” can still help on valuation, but investors are now asking a harder follow-up: how much of the business is actually defensible because of AI?
Carta data for the first half of 2026, as cited in the piece, shows AI startups raising roughly the same amount of money in seed rounds as non-AI companies, but at valuations about 50% higher. The premium is still real. What changed is how broadly it gets applied.
The article splits AI companies into sharply different categories. Commodity AI wrappers are framed at 3x to 8x revenue. Vertical AI products with sticky proprietary data are placed at 10x to 20x. Companies with both intellectual property and proprietary data reach 25x to 40x.
Scrutiny gets tighter as companies mature. At seed, businesses without protectable IP can face a 20% to 30% discount relative to more defensible peers, according to the analysis. By Series A, that gap widens to 30% to 40%.
That helps explain why adding AI to a product is becoming less meaningful on its own. Investors want to know what remains after the AI label is stripped away. Could another startup rebuild the same product by calling the same foundation-model API? Does each new customer generate proprietary data that improves the product over time? Is the software embedded deeply enough in the workflow that replacing it would be painful? If inference volume rises 10x, do the unit economics still work?
The piece says founders heading into the next round should pressure-test four areas: technical differentiation, proprietary data, workflow lock-in, and unit economics at scale.
Model substitution, data feedback loops, workflow depth and cost structure now define the split
Its proposed test for technical differentiation is simple: swap the underlying model. If a company can replace its main model provider with another comparable model and the product barely changes, investors may struggle to see a real technical moat.
On proprietary data, user count alone is not enough. A stronger signal is whether more usage produces data that makes the product measurably better over time, whether in accuracy, resolution speed or match quality.
Workflow depth offers another clue. If customers with deeper integrations consistently show stronger retention and expansion, founders have evidence of switching costs instead of just saying the product is sticky.
Then there is the business model. A product may show healthy margins at current usage levels and look very different once inference volume rises by 10x. The article says founders should work through that math before investors do, because it reveals whether scale strengthens the business or exposes a weakness.
It also notes how fast the market narrative has changed. Eighteen months ago, touching AI was often enough to strengthen a fundraising story in a material way. As more startups use the same pitch, investors have more reason to look underneath the surface. The AI label can still get a company into the room. Increasingly, the more valuable part is proving why the thing being built gets harder to copy as it grows.
Software may be heading toward a barbell structure
The article also looks at how AI could reshape the structure of software markets. Mike Vernal of Conviction draws a comparison with what the internet did to newspapers, arguing that software may be moving toward a similar barbell pattern.
Before the internet, regional newspapers benefited from expensive distribution. Once distribution became close to free, much of the middle disappeared. A small number of global publications became much larger, while thousands of independent newsletters and niche publishers emerged.
Vernal says AI could create the same shape in software. Traditional software moats usually came from three things: products were expensive to build, switching systems was painful, and integrations created strong ecosystems. AI weakens all three. Competitors can copy features faster, migration can become more automated, and AI can make integrations much easier.
The article describes a market with strength at both ends.
- Giant platforms: a small number of companies could own entire purchasing categories such as sales, marketing, finance, HR, legal or healthcare, expanding beyond a narrow tool until they become the system customers use for almost everything.
- Micro-software businesses: AI coding tools may let millions of people build highly specific software for themselves, their companies or a small group of clients. Some of those products could become profitable niche businesses.
- The uncomfortable middle: mid-sized point solutions could face the most pressure. If large platforms keep adding their features and small teams can cheaply rebuild narrow products, defending a stand-alone tool becomes harder.
The article uses Amazon as an analogy for what the next moat may look like. Amazon started with something relatively easy to copy, selling books online. Its defensibility came from decades of reinvestment in logistics, infrastructure, technology and distribution. In software, AI may push companies toward the same strategy: build faster, expand constantly, reinvest, and make the overall product system harder and harder to replicate.
That leaves venture-backed software startups with a strategic question. If building software is close to free, is the moat still just a good product, or everything built around it?
Seed valuations are often constrained by fund math before anyone debates the story
The article says founders often assume valuation mainly reflects traction, market size or how well a pitch lands. But another constraint can matter just as much: the economics of the VC fund on the other side of the table.
It offers a simple example. Suppose a founder is pitching a $28 million seed fund that usually writes checks of around $600,000 and wants roughly 6% to 8% ownership. At 7%, a $600,000 investment implies a post-money valuation of about $8.6 million. At 6%, it implies $10 million. If the company is raising at an $18 million valuation, that fund may simply be unable to make the investment fit its portfolio model.
The chain is laid out plainly: fund size → check size → target ownership → feasible valuation ceiling.
The article argues this matters even more now because seed valuations have risen. Carta’s cited median seed post-money valuation is $24 million. A $50 million fund investing $1.5 million at that price gets only 6.2% ownership, which may fall short of target.
That changes how founders should think about round size. If a company says it wants to raise $3 million and the lead investor wants 15%, the round already implies a $20 million post-money valuation before anyone explicitly says so.
Round design, dilution and the option pool can matter as much as the headline number
The more useful sequence, the article says, is to start with how much money the company needs to reach the next milestone with 18 to 24 months of runway. Add a buffer. That gives the round size. Divide that number by the dilution founders are willing to accept. Then target funds whose check sizes and ownership goals can support that valuation.
It flags another number founders often overlook: the option pool. Lead investors may ask for a 10% to 15% option pool to be created before the financing closes, which dilutes existing shareholders before the deal is done. The article says negotiating the pool based on a real hiring plan can sometimes matter more to founder ownership than fighting for another $1 million or $2 million in headline valuation.
To avoid spending weeks with the wrong investors, the piece suggests asking one question on the first call: what is the fund’s typical first check, and what level of ownership does it usually want? Those two numbers can quickly show whether the fund’s economics support the round.
Not every valuation gap should be read as a verdict on the startup itself, the article argues. Sometimes the company is not the issue. The round just does not fit the math of the fund.
If ChatGPT can do it for free, why would users keep paying for a separate app?
The analysis says this is becoming one of the biggest questions for consumer app founders. If users can open ChatGPT, Claude or Gemini and get a similar result in seconds, why would they pay $10 or $20 a month for a stand-alone product?
Daphne Tideman of RevenueCat argues the answer is not simply adding more AI. AI-powered apps are monetizing well. Revenue per paying user is 41% higher than in non-AI apps, and trial conversion is 52% higher. But 12-month retention is only 21.1%, versus 30.7% for non-AI apps.
The implication is blunt. AI can bring people through the door. It does not automatically give them a reason to stay.
The article points to the contrast between Chegg and Duolingo. The difference is not just the answer itself, but what exists around that answer. RevenueCat says apps can build six things that are harder for a blank LLM chat box to copy:
- Structure: turning an answer into a complete workflow or experience.
- Memory: learning the user over time so the app becomes more useful.
- Habit: bringing users back through streaks, reminders, accountability and gamification.
- Precision: producing better results through specialized data or expertise.
- Connection: building community, identity, personality or emotional attachment.
- Physical-digital links: connecting software with sensors, devices or real-world data.
Duolingo is presented as a strong example because it stacks several of those layers. ChatGPT can teach a language, but Duolingo spent years building streaks, leaderboards, progress systems, reminders and gamified loops. According to the article, introducing leaderboards increased learning time by 17% and tripled the number of highly engaged learners.
The piece also offers a simple “blank textbox test.” Open ChatGPT or another LLM, describe as accurately as possible the problem users come to your app to solve, and compare the answer with what your product actually delivers. If most of the value can be recreated with one prompt and one response, the app has a differentiation problem. If the product adds structure, cumulative data, habits, precision, connection or real-world components, that may be where the moat sits.
The article ends with a clear view. AI itself may not be the moat. The more important question for founders is what their product gives users that a blank chat box does not. RevenueCat’s position is that the winners will use AI underneath and compete on the experience built around it.
