The Harsh Truth of the VC Industry: Why Most Funds Haven't Returned Principal After Nine Years?

Author: Venture Curator

Compiled by: Deep Tide TechFlow

Deep Tide Introduction: This article uses two sets of hard data to expose two facts: that major VC firms are clustering to invest in homogenized AI projects, while emerging managers are more willing to bet on outliers; and that most VC funds have not returned even the principal to LPs nine years after their establishment. For founders currently raising funds, understanding the financial pressures behind the funds is far more important than the allure of their brand.

Emerging VCs are investing in more unusual startups, while AI spending varies greatly among over 70,000 companies.

Who is Investing in Outlier Venture Capital? What Does the Data Indicate?

Venture capitalists are often taught that this is a business about outliers. A few unexpected companies contribute to most of the returns, so institutions with the largest teams, broadest networks, and most capital should be best positioned to find them.

However, new research by Dan Grey of Odin, based on Dealroom data, shows that the largest institutions are now the least likely to invest in a company that looks different from all other funded companies.

Grey scored each company in the seed investment portfolios of five giant funds (a16z, General Catalyst, Lightspeed, NEA, and Accel) from 2023 to 2025, along with a select group of top emerging managers, based on rarity. Rarity was measured from two dimensions: business (industry, model, customers, geography) and technology, compared to all companies that received seed funding from 2018 to 2025. The least common fifth was classified as outliers.

Emerging managers directed 24.7% of their rated seed investments towards outlier profiles (49 out of 198). The giant funds only invested 11.5% (32 out of 278). This represents nearly a twofold gap.

This gap does not arise because every small fund is a contrarian investor. Many emerging managers are heavily invested in AI, just like the giant funds: Conviction is 100% AI, and South Park Commons is 93.5%. The difference comes from specialized funds with different missions, such as Lowercarbon, which only invested 18.8% in AI, and Multicoin, which only invested 15.4%.

This is why portfolio configuration is important. When combining the ledgers of five different emerging managers, the concentration typically decreases by 22.6%. When combining five giant funds, it only decreases by 2.0%, as all five are already between 82.7% and 89.5% in AI positions. Among 462 possible combinations of emerging managers, only 2 failed to outperform the giant fund group.

The significance of this goes beyond portfolio statistics. Research on financing risks has long suggested that whether investors expect subsequent funding availability determines which experiments can secure the first check. Odin cautiously states that its data does not prove this mechanism. But if the largest pools of capital only invest in projects that are widely considered fundable, then companies outside this consensus can only rely on others to place the first bet.

So does this mean emerging managers are better at picking winners?

Not necessarily. "Outliers" here refer to those that are rare among funded companies, not necessarily more innovative or more likely to succeed. When founder backgrounds are included in the scoring, the gap narrows to 19.8% for emerging managers versus 14.4% for giant funds. The emerging manager group is a selected sample and does not represent the entire market. Moreover, small funds also have their own herd moments: blockchain accounted for 46.7% of their early rounds, later dropping to 20.9%.

This is precisely where investors need to be cautious.

If your argument is that returns come from outliers, then pairing with a prestigious fund mostly just adds the same bets. A basket of truly different mission-driven funds can broaden the scope. But merely having a broad scope does not justify writing checks. Odin's own test is: what can a fund add to your existing portfolio that is missing, at what cost and with what ancillary rights, what is the diluted ownership percentage, and what are the risks of subsequent financing?

The real question is not which fund is the largest. The question is which funds are genuinely looking where others are not, because the data shows that this list is increasingly not the same group of people.

Nine Years Later, Most VC Funds Still Haven't Returned Money to Investors. What Will Happen Next?

Founders are often taught to view their VCs as patient capital. Venture capital is a long game; funds are set up to wait for big outcomes, and no one on the cap table is in a hurry for cash.

However, new data from Carta's Q2 2026 VC fund performance report indicates that many funds' patience is wearing thin, as most have not returned their investors' money.

Carta tracks net DPI, which is the ratio of cash actually returned to LPs to every dollar invested by LPs, covering about 3,000 U.S. venture capital funds that began investing between 2017 and 2026.

Funds that started investing in 2017 are now about nine years old, nearing the end of a typical ten-year fund lifecycle. The median fund has only returned 0.37 times cash. Even the top quartile funds have only returned 0.70 times. Only the top 10% have returned more than the amount invested by LPs, reaching 1.37 times.

Younger funds are faring worse. The median return for 2018 funds is 0.15 times, and for 2019 funds, it is 0.04 times, meaning that after seven years, only 4 cents are returned for every dollar.

Value exists on paper. The median book value for 2017 funds is 1.72 times, but only about one-fifth of that has been returned in cash. The rest remains locked in private companies that have not yet been sold or gone public.

Another detail explains why it is hard to see this from the outside. Most of these funds have indeed returned some money: 86% of 2017 funds and 70.4% of 2018 funds have returned at least some cash. So almost every fund can tell its LPs that distributions have begun. But very few funds can say they have returned money.

So why should founders care about their investors' fund books?

Because your investors' fund clock does not align with your company's clock. A fund that is close to expiration, still owes LPs real cash, and is raising the next fund has a strong incentive to achieve liquidity quickly. This pressure may manifest as pushing you to sell earlier, sell shares in the secondary market, or be more cautious about continuing to invest in your next round of financing.

And this is where founders need to be careful.

None of this means your investors will do something detrimental to you. Funds can extend their lifetimes, and many top institutions have ample reason to continue supporting their winners. Carta's data only shows the cash gap, not how each fund will act. But the year of the fund that writes you a check is now a real variable, not a footnote. A partner coming from a 2017 or 2018 fund is in a completely different situation than one coming from a brand new 2025 fund.

The real risk is not that your investors need liquidity. It is that you do not know which fund your money comes from, how much cash it still owes its investors, and how this will affect their advice in your next round of financing or exit.

Where is the Biggest Money in AI: Cutting-Edge Models or the Middle Market?

Most AI reports focus on the cutting edge: the latest, smartest, and most expensive models from each lab. But Tomasz Tunguz of Theory Ventures makes an argument: the truly important market is the middle tier, and spending data supports this.

The middle tier has taken about 40% of AI spending and 30% of token usage.

In contrast, the share of cutting-edge models is thin. Anthropic's strongest model, Fable 5.1, accounted for only 3.7% of gateway spending in the first 12 days after its release. Its predecessor peaked at 13.2%, and then dropped to 4.9% a month after the release of Opus 5 at half the price. Among large enterprise accounts, the token consumption share of cutting-edge models fell from 53% in early August to 45% in September.

Price wars are also unfolding at this level. In September, when Anthropic lowered prices on its new model, OpenAI followed suit about 90 minutes later. Before that price drop, the Opus series had maintained pricing at $5 per million tokens input and $25 output for four consecutive versions. At the low end, OpenAI slashed the price of Luna by 80% in July and then by 50% in September.

Three forces continue to push prices down in the middle tier:

Competition among labs. Labs now follow each other's price cuts within hours, rather than over several quarters.

Open-weight models. Open models run most of the token volume on gateways at prices 86% lower than the mixed prices of closed-source models.

Fine-tuning. Cursor Composer 2, fine-tuned based on the open-weight model Kimi K2.5, reduced total costs by 86% compared to previous self-developed models. Harvey reduced the cost per cell relative to Sonnet 5 by 55%, while scoring higher than Fable 5.

This situation is structurally persistent. The tasks that enterprises need AI to perform, such as summarizing a contract or classifying a work order, remain almost unchanged year after year. Yet the cost of the intelligence that meets this threshold is rapidly declining. Therefore, the level that can meet a fixed requirement is becoming cheaper every quarter, while most real work falls at this level rather than at the cutting edge.

Demand does not appear as a pyramid with a cutting edge at the top, but rather as a bell curve that is thick in the middle. Buyers care about how much intelligence they can get for each dollar, rather than peak capability. The unresolved question is whether this middle market will eventually become commoditized. If so, the economics of the entire AI market will change accordingly.

For anyone building products on these models, it is worth asking: which tier does your product truly need? Paying cutting-edge model prices for middle-tier work is now the worst way to lose profits.

How Much Should Your Startup Spend on AI? The Data Says This.

Ask most founders how much they spend on AI, and you will get a confident number: seats for ChatGPT and Claude, Cursor bills, and growing API costs, adding up to thousands of dollars a month. This sounds precise, but it is not. Because it combines three completely unrelated expenditures, with the only commonality being the category of supplier.

Ramp's payment data covering over 70,000 U.S. companies shows that "using AI" has become highly fragmented.

The median company spends $11.38 per employee per month. The top 10% spend $611.

The top 1% spend $7,449, nearly 650 times the median. When Ramp correlated these expenditures with the workforce records of 21,559 companies from Revelio Labs, it found that companies with high spending saw a 10.2% increase in employee numbers over two years, with a 12% increase in entry-level hiring.

Low-intensity users showed no significant changes in data.

So spend more. But where to spend?

Most founders compare AI bills with their software budgets because these costs appear on the same bill alongside Figma and Notion. This is the wrong denominator. Increasingly, AI spending is not replacing software but rather people. A customer service agent's counterpart is the customer service position, not your SaaS suite.

Take a typical seed-stage company: 12 people, with a monthly salary cost of about $180,000, and AI spending of $4,000 per month. Compared to the software budget, $4,000 seems significant. Compared to salary costs, it is only 2.2%. The same number leads to completely opposite conclusions.

The solution is to break down the AI budget into three parts, each evaluated against its own benchmark:

• Wage Line AI (customer service, SDR, and programming agents) is compared to the salary of the person you didn't hire for that position. If you can't name that position, it shouldn't fall into this category.
• Productivity Line AI (seats, note-taking, and writing tools) looks at usage rates. If usage declines by the second month, eliminate it.
• Cost Line AI (tokens and API calls within the product) examines gross margins at a 10x scale, and it must be included in your COGS.

Mixing the three categories together causes each signal to cancel each other out. Overspending on seats makes the total appear bloated, leading founders to restrain spending on agents, which is precisely where the returns lie. The top 1% of winners do not spend more; they categorize their spending well: they double down where AI replaces human labor, strictly control spending where it merely accelerates human work, and focus on gross margins in internal product areas.

We have broken down a complete ranking method, including a 30-minute auditing process that you can use this week to verify your bills. See "What Should Your AI Spending Benchmark Be?"

Why Venture Capital Due Diligence in Deep Tech Has Deteriorated?

You might think that the most difficult technologies should undergo the strictest scrutiny. A quantum computer, a nuclear fusion reactor, or a new chip architecture must either comply with the laws of physics and function or not, and verifying this requires real expertise.

But Marin Ivezic of PostQuantum believes the reality is quite the opposite: the larger the deal and the higher the technological content, the less likely it is that an independent person can verify whether the science holds up.

This is not because investors have become lazy. Several structural changes have compounded this issue.

The expert layer has disappeared.

In the past, investors could obtain independent technical judgments from bank research teams, which employed professional analysts. The 2018 European MiFID II regulations required research fees to be paid separately from trading fees, leading to a contraction of this model. Reports indicate that the scale of European equity research has decreased by about 20%, and the number of professional analysts at major banks has also declined.

Expert networks are now a $2.5 billion business, filling some of the gaps. However, they are only effective if you know which expert to ask and what questions to pose. In cutting-edge fields, this is precisely the knowledge that generalist investors lack.

Funding has flowed to generalists.

In 2025, private quantum investments reached $4.9 billion, a 192% increase. The largest investors are BlackRock and NVIDIA, not specialized quantum funds. Sovereign wealth funds and pensions are also increasingly participating in these rounds; a survey of sovereign funds found that 58% of institutions lack the resources to lead a deal. Capital that cannot conduct its own technical reviews ultimately has to rely on others' judgments.

The lead investor's name has replaced scrutiny.

A reputable lead investor provides a credit endorsement for all co-investors. Co-investors assume that the lead investor has verified the physical feasibility, but they do not see that review and cannot confirm it actually happened. Some rounds lack even this: Quantinuum's $600 million financing had no named lead investor.

Deadlines and herd mentality prioritize speed over scrutiny.

By the end of 2023, U.S. funds had a record $311.6 billion in uninvested capital. A fund manager under allocation pressure finds joining a hot deal more profitable than slowing it down. Moreover, being wrong alongside peers carries far less reputational damage than being wrong alone.

Deep tech lacks mandatory checkpoints.

Biotech serves as a good contrast. The FDA's approval process forces data disclosure at every stage, allowing the industry to establish a comprehensive set of due diligence standards: scientific advisory boards, professional reviewers, and phased checks.

The quantum field lacks equivalent checkpoints. The most rigorous technical review found for large quantum companies in this article comes from the Australian government, disclosed through a freedom of information request, rather than from any investor.

The results are clearly reflected in the numbers.

In 2025, private quantum investments will total $4.9 billion, while the total revenue of the entire quantum computing industry is about $1.4 billion. Three quantum companies that went public via SPAC in 2021 had predicted combined revenues of about $1.2 billion by 2025, but delivered only about $138 million. Moreover, those who conducted the most rigorous verification work were short sellers, not investors.

How this will end has precedents. Between 2006 and 2011, venture capital invested over $25 billion in clean technology, with losses exceeding half. Scholars studying that cycle warned at the time that pensions and sovereign funds without hardware experience would be the next to fund such companies.

The proposed solution is simple: before co-investing in a deep tech round exceeding a certain scale, LPs should require an independent technical feasibility review that cannot be commissioned by the fund or selected by the founders. It should be noted that the author himself invests in quantum startups and runs a company providing such review services. However, the core point remains valid: in the deep tech field, a mark on the equity table has quietly replaced technical verification, and in the face of such check amounts, a few weeks of independent review costs are negligible.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.