> ## Content Index
> Fetch the complete content index at: https://wire.fourthweb.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# VCs Split Into Two Species: Those Who Can Write $100M AI Checks and Those Who Can't
- URL: https://wire.fourthweb.ai/vcs-split-into-two-species-those-who-can-write-100m-ai-checks-and-those-who-cant/
- Published: 2026-08-04T04:00:06.000Z
- Updated: 2026-08-05T03:30:49.000Z
- Description: The VC world is fracturing into haves and have-nots, and the dividing line is whether you can write nine-figure checks for AI infrastructure. Felix Capital aimed for $600M, still $150M short — LPs want returns from old funds before committing new capital
- Author: Travis Wright
- Tags: AI Agent Economy, AI Agents, AI Infrastructure, OpenAI, Anthropic, a16z, Funding Rounds

**The VC world is fracturing into haves and have-nots, and the dividing line is whether you can write nine-figure checks for AI infrastructure.**

### The Summary

- [Felix Capital aimed for $600M, still $150M short](https://www.bloomberg.com/news/features/2026-08-04/big-ai-bets-divide-venture-capital-leaving-smaller-funds-behind?ref=wire.fourthweb.ai) — LPs want returns from old funds before committing new capital
- Mega-funds writing $100M+ AI infrastructure checks are hoovering up capital while traditional consumer/software VCs struggle
- The bifurcation creates a two-tier system: firms that can play in foundation models and [compute](https://wire.fourthweb.ai/tag/ai-infrastructure/), and everyone else scrambling for application-layer scraps

### The Signal

Venture capital is splitting along a fault line that has nothing to do with skill and everything to do with scale. Felix Capital's fundraising struggle tells the story cleanly: a firm with legitimate wins in Peloton and Deliveroo can't hit its $600M target because [LPs are demanding proof of returns](https://www.bloomberg.com/news/features/2026-08-04/big-ai-bets-divide-venture-capital-leaving-smaller-funds-behind?ref=wire.fourthweb.ai) before opening their wallets again. Meanwhile, Andreessen Horowitz, Sequoia, and Benchmark are raising multi-billion-dollar war chests specifically to chase foundation model companies that burn $50M a month on compute.

The math is brutal. AI infrastructure deals require $100M+ minimum checks to be relevant. Model training runs cost eight figures. If you can't write a check that size, you're not getting a board seat at the companies building the actual agent infrastructure. You're left fighting over application-layer plays where margins are compressing because everyone's using the same three foundation models.

> "The traditional VC model of writing $5M [Series A](https://wire.fourthweb.ai/tag/funding-rounds/) checks into promising startups is getting squeezed from both ends."

This isn't just a fundraising cycle blip. It's structural. The limited partners who fund VC firms are getting selective in a way they weren't during the 2020-2021 boom. They're concentrating capital with proven mega-funds who can access the AI infrastructure layer, and they're starving out mid-tier firms who made their names in consumer tech or SaaS. Felix's Peloton win doesn't matter when LPs want exposure to whoever's building the next Claude or the next inference optimization stack.

The downstream effects are already visible:

- Seed and Series A rounds in AI applications are getting \*more\* competitive, not less, as displaced capital hunts for deals
- Promising agent-layer companies are getting passed over because VCs can't commit to the follow-on rounds those companies will need
- Talent is consolidating at mega-funded AI labs, creating a winner-take-all dynamic in both capital and engineering

The application layer isn't dead, but it's getting commoditized faster than most VCs expected. When [OpenAI](https://wire.fourthweb.ai/tag/openai/) or [Anthropic](https://wire.fourthweb.ai/tag/anthropic/) ships a new capability, dozens of funded startups see their moats evaporate overnight. The only sustainable plays are infrastructure (too expensive for most VCs) or hyper-vertical applications with proprietary data moats (harder to scale, harder to exit).

### The Implication

If you're building in AI, know which game you're playing. Infrastructure requires patient capital and relationships with the handful of firms who can write $100M checks. Agent applications need to be so vertically specific that foundation model providers won't bother competing. The middle ground where most VC-backed startups have historically lived is vanishing.

For everyone else watching this unfold, the message is clear: the companies building Web4's foundation layer are being chosen right now, and the selection process is happening behind closed doors at a dozen mega-funds. The rest of the VC ecosystem is being forced into a supporting role, funding the tools and wrappers that sit on top of infrastructure they don't control.

### Sources

[Bloomberg Tech](https://www.bloomberg.com/news/features/2026-08-04/big-ai-bets-divide-venture-capital-leaving-smaller-funds-behind?ref=wire.fourthweb.ai)