# 风投为何青睐估值超十亿美元的首轮融资？

- 来源：Deedy (@deedydas)
- 发布时间：2026-05-05 10:58
- AIHOT 分数：61
- AIHOT 链接：https://aihot.virxact.com/items/cmos2fdaa04eqslrjbhkkxfl6
- 原文链接：https://x.com/deedydas/status/2051496852076584988

## AI 摘要

当前约25家公司在首轮融资即获超十亿美元估值，其背后逻辑在于：AI领域潜在回报巨大，如Anthropic和OpenAI预示了万亿美元级别的私有市场流动性；同时，初创公司的核心约束是算力（GPU）和顶尖人才，大规模融资是获取这些稀缺资源的硬性要求。此外，有限的竞争提高了成功后的价值捕获概率。市场层面，大量资本追逐极少数顶尖团队推高了估值，而大型基金出于配置压力和追求规模回报的需要，也倾向于下大注。投资者对错过下一个AI巨头的恐惧以及创始人通过高估值减少稀释的动机，共同促成了这一趋势。

## 正文

http://x.com/i/article/2051490977140105217

# Are VCs dumb for investing in crazy $1B+ seed rounds？

There are ~25 companies with a ~$1B+ headline valuation in their first round of raised capital （TML， SSI， Ineffable， Ricursive， World Labs etc）. Here are 11 reasons this happens， and the case for it：

1. Outcomes are larger than ever before. Anthropic and OpenAI are ~$1T outcomes that are mostly liquid in private markets. Many of these startups are "neolabs"， big bold expensive research ideas pre-revenue and pre-product. If every single "neolab" raises at a $1B post at seed， if 1 out of a 100 "hit"， that could be a 10x return in 5-7yrs， an implied 40-60% IRR net of dilution.

1. Compute procurement. Today， capital is not much of a constraint as much as access to compute （GPUs） for many of these companies. This means getting， say， $100M， worth of GPUs can be a hard requirement. If you need $100M of GPUs and want low dilution （10%）， you might easily back into a $1B valuation.

1. Talent procurement. Talent is also more of a constraint than capital. AI researchers are expensive， both on cash and equity. You need to raise to pay enough cash. High valuations also correlate with a high strike price / 409a， even though it might be 10-20% of the preferred price. If you're valued at $1B and giving 0.25% to a researcher， they might still have to pay $250-500k to buy $2.5M of options / paper money. They might already have that money lying around but often could use cash to finance that and demand a commensurately higher base salary.

1. Limited competition increases success rate. p（success） = p（success | this idea works） x p（this idea works）. It's hard to measure p（this idea works）. But， because AI is compute， talent and capital constrained， p（success | this thing works） is much higher because you might be competing with <5 other plays vs 1000s. If the idea works， you'll capture the value.

1. Preference stack is a perceived downside cushion. If the value of the talent and IP is significant （remember， Meta pays 9 figure packages for top talent）， there's a belief that this is easily acquired for at least more money than you put in. If you invest $100M at $1B， you might believe there's no way the team doesn't get bought for >$100M， as evidenced by many such acquisitions （Cursor， OpenClaw， Windsurf， Vercept， Astral， Bun， Coefficient Bio - not all are neolabs） so investors get their money back （>1x）. That might lead VCs to believe the downside case isn't too bad.

1. Institutional investors often don't pay the sticker price. Rounds are done in multiple closes to reduce the blended cost basis for the investor. Institutional players might get their ownership in the first unannounced round and put a token amount in the next， with strategics like NVIDIA piling on in the final round. Multiple rounds counted together is called one "seed".

1. Pure market dynamics. A lot of capital is chasing few very strong teams （even with no product） and a bidding war can substantially inflate the valuation.

1. Large funds and large fund dynamics. Venture funds before have grown drastically in size from 10 years ago， which allow funds to even think about writing these large checks. If you have 10 partners on a $10B fund， each partner has to deploy $1B over， say， 3yrs. Deploying $300M/yr often happens in growth but in early， this is a very high deal volume. Deploying in one big swing is easier. You also need massive outcomes to move the fund， so large checks are structurally easier than hundreds of small early-stage positions.

1. Investor FOMO. Everyone wants to be in the next OpenAI / Anthropic and wants to fund things that resemble these labs at an early stage. LP pressure exists as well.

1. Founder greed. Occasionally， founders and employees take secondary even at the seed which incentives founders to push for a high valuation to reduce dilution. Pre-product pre-revenue secondary can be sizable （$10M+）， which leads founders to push for higher valuations.

1. Founder FOMO. Many talented people （professors， very senior researchers） see their peers who they think they're just as good at or better than raise large amounts so they start with a $50-100M ask out the gate. Investors who may have missed a competitor feels compelled to back a horse in the race. They feel "forced" to pay up because the company will not be competitive with much less cash.

The failure mode
In the best cases for these companies， you have S-tier founders， hire fantastic researchers， acquire compute and go after an important valuable problem. What can go wrong is

- Velocity stalls. After 1-2yrs， companies lose momentum and urgency when a） they're not hot anymore b） no clear research breakthroughs c） no clear revenue traction d） no easy fundraise e） no liquidity in secondary markets.

- Employees see their friends do well elsewhere， and begin to leave.

- They solve the important research problem， but it does not translate into a revenue generating business the way they thought.

- They solve the problem and make a lot of money， but continue to require so much capital that the investors get too diluted.

In 1/2/3， the company goes into sell mode and hopes they clear the preference stack to make money. In 4， the investors just don't make the return they were hoping for.

My take
Most of these will not turn out well， but at no less of a rate than venture overall. If even one hits， the returns will justify the ones that didn't work.