Last month, global venture funding hit $189 billion.
Of that, $110 billion — 58% — went to a single company: OpenAI.
Can we still call this "diversified investing"?
Everyone's calling it a bubble. The data says otherwise.
"AI bubble," "it's about to pop" — you've probably heard this more than once lately. But when you actually dig into the performance data, the story looks different.
Carta, the cap table management platform, analyzed 2025 funding data from startups on its own platform and found AI startups captured 41% of the $128 billion in venture funding. That's an all-time high. The top 10% of startups took home half of all funding.
Normally, this kind of concentration talk leads straight to "so it's about to collapse" — but Peter Walker, Carta's head of insights, tells a different story. Younger funds formed after ChatGPT launched (late 2022) are starting off with strong IRR (internal rate of return).
"AI startups aren't raising huge rounds because they have more employees. It's because running AI models is just expensive. So round counts went down, but capital per round went up."
— Peter Walker, Head of Insights, Carta
PitchBook's numbers back this up — AI's share of total venture funding (trailing 12 months) climbed from 23% in Q3 2020 to 40% in Q3 2024, and now sits at 63%. That's nearly a 3x jump in three years.
But this pot is all riding on four names
Peel back that "great scorecard" one more layer, and you get a different picture.
Over the course of February alone, OpenAI raised a $110 billion round — one of the largest private rounds in history. That same month, Anthropic closed a $30 billion Series G at a $380 billion valuation. In January, xAI raised a $20 billion Series E. By some counts, four companies — OpenAI, Anthropic, xAI, and Waymo — took 65% of all venture funding in Q1.
Investment research outlet Angel Investors Network put it bluntly.
"VC assets in H1 2026 are no longer a diversified asset class. They're a concentrated bet on AI infrastructure wearing a 'diversified fund' costume."
— Angel Investors Network
By another count, OpenAI and Anthropic alone took 43% of all H1 venture funding — $217 billion. The fund names might read "Growth Fund," "Tech Ventures," "Strategic Fund" — but the servers all trace back to the same handful of data centers.
This concentration creates two completely different games depending on whether you're AI or not. Just look at seed-stage median valuations: AI startups sit at $17.9 million, non-AI at $12.6 million — a 42% gap.
| Metric | AI Startups | Non-AI (robotics benchmark) |
|---|---|---|
| Seed median valuation | $17.9M | $12.6M |
| ARR valuation multiple | 100x+ | 8–12x |
| Average time to exit | 4–6 years | 8–10 years |
At a glance, the table makes AI look like the clear winner — but what it's really saying isn't "AI wins," it's "AI and non-AI are playing by completely different rules". Lower valuations mean less dilution, and while exits take longer, you're burning less capital to get there.
Korea isn't immune either. In the first half of this year, 141 domestic deals worth over ₩10 billion accounted for 93% of all investment, and 91.5% of seed rounds went to AI and robotics. The VC data firm TheVC noted that "capital isn't flowing evenly across companies anymore — investment is increasingly concentrated around large deals and AI/robotics".
Heads Up
A better IRR doesn't mean cash actually hit anyone's bank account. Most of the recent venture market recovery is paper gains (unrealized value) — actual cash distributions (DPI) are still weak. Industry watchers describe the current moment as "not a valuation problem, but a realization problem".
So what should you actually check right now?
- Know the AI premium exists
Even at seed stage, slapping on an "AI" label gets you a 42% higher valuation. If you're positioning as non-AI, line up alternative moats AI can't easily copy — think proprietary datasets or regulatory barriers to entry — instead of algorithmic innovation. - Split capital around milestones
Use convertible notes and valuation caps so additional capital only unlocks once you hit specific technical or commercial milestones. This is especially critical for hardware businesses. - Rework your deal-sourcing network
Research-lab pipelines from Stanford, MIT, and CMU, plus spinout lines from big players like Lockheed Martin, Boeing, and Siemens, work far better for non-AI deal sourcing than public platforms. - Prove it with capital-efficiency numbers
Gross margins above 60%, CAC payback under 18 months, NRR above 100% — these three numbers prove you can grow without burning cash the way AI does. There are real robotics and defense-automation cases that hit $10 million in revenue on just $15 million in total capital raised. - Audit your own exposure
If you work at a startup or invest as an individual, figure out whether the equity or funds you're holding are exposed to infrastructure or to applications. Start by asking "what's my actual exposure" instead of trusting the label.
Deep Dive Resources
Original article TechCrunch's original piece on AI startup venture concentration and returns, based on Carta data techcrunch.com
AI mega-round concentration risk analysis An LP-focused deep dive on "concentrated bets wearing a diversified-fund costume" angelinvestorsnetwork.com
The non-AI founder playbook Strategy breakdown for the non-AI startups that took 19% of Q1 2026 venture funding angelinvestorsnetwork.com
Anthropic's official Series G announcement Investor list for the $30 billion round closed at a $380 billion post-money valuation anthropic.com
AI vs. non-AI valuation gap data A numbers-based breakdown of the AI premium at seed and Series A stages agentmarketcap.ai
Korea's investment polarization Data on large-deal and AI concentration in domestic startup investment this H1 sedaily.com
