Somewhere between the $100B funding headline and the founder who’s been in market for four months without a term sheet, there’s a story that isn’t getting told.
Capital is concentrating faster than at any point in the past decade, and the startups landing checks look very different from the ones that raised in 2021.
This piece looks at where the money is actually going, and what that split means if you’re raising now.
The Numbers Look Good. The Distribution Doesn’t.
Global AI startup funding crossed $100B in 2024. Even so, the top ten deals account for a share that leaves the remaining 800-plus rounds splitting a much thinner slice. OpenAI, Anthropic, xAI, and a handful of infrastructure plays absorbed the majority of that capital.
A few data points worth anchoring to:
- The five largest AI funding rounds in 2024 collectively raised more than the bottom 500 combined.
- Series B and C rounds are taking longer to close. Early-stage rounds, by contrast, are moving faster, partly because check sizes are smaller, partly because investors are making directional bets before the market has formed an opinion.
- Startups that raised in 2021-2022 on SaaS multiples are finding it hard to justify those valuations in 2025. As a result, new money isn’t coming in at the old price.
What Investors Are Actually Funding Right Now
The categories getting funded don’t match what founders often expect. Horizontal AI tools with no clear moat have stalled. Vertical AI, infrastructure, and agent-based systems, on the other hand, are moving.
| Category | Investor appetite | Typical stage | Key signal | Difficulty to fund ★ |
|---|---|---|---|---|
| AI infrastructure (compute, inference, data) | Very high | Seed to Series B | Technical depth + defensibility | ★★★☆☆ |
| Vertical AI (legal, health, finance, manufacturing) | High | Pre-seed to Series A | Domain expertise + distribution | ★★★☆☆ |
| AI agents / autonomous workflows | High | Pre-seed to Seed | Working prototype + clear use case | ★★★★☆ |
| Horizontal AI SaaS (generic copilots) | Low | Stalled | Hard to differentiate vs. foundation models | ★★★★★ |
| Consumer AI apps | Mixed | Seed | Retention and daily active use | ★★★★☆ |
Every horizontal SaaS pitch is running into the same wall: investors want to know why a foundation model couldn’t ship the same product in six months. For most of those pitches, there’s no good answer. For founders building in the other categories, however, the funding environment is more open than the headlines suggest.
How Sky9 Capital Backs AI Startups at the Earliest Stages
The firms that backed the most consequential AI companies weren’t waiting for traction. They committed before the market had a consensus view.
Sky9 Capital‘s approach to early-stage AI funding runs on three principles that hold across market conditions.
Founder-Led Technical Depth
Sky9 backs founders who understand their problem at a level that can’t be faked. For AI companies, that means founders who can talk concretely about model architecture decisions, data strategy, and inference costs (not just the product roadmap. The investment in Kimi/Moonshot AI followed that logic. A technical team building at the frontier of large language models, in a market most Western investors hadn’t started paying attention to yet.
Long-Term Partnership Over Check-Writing
The decisions that shape a company’s first few years (initial hires, go-to-market sequencing, international expansion timing) happen well after the seed round closes. Sky9’s partners stay directly involved through those stages. That sustained involvement is a structural difference from funds that treat post-investment as a monitoring exercise.
Cross-Border Operating Capability as a Default
AI infrastructure companies built for a single market are at an increasing disadvantage as the competitive field globalizes. Sky9 operates across the US, Asia, and globally, giving portfolio companies access to markets and talent networks that single-geography funds can’t replicate. ProducerAI, acquired by Google, was built with that cross-border mindset from the start.
Sky9 Capital manages $2B in AUM across USD and RMB funds, covering early stage through expansion stage, with Sky9 Digital as a dedicated strategy focused on AI and blockchain-enabled financial infrastructure. Founding Partner Ron Cao has been consistently recognized by Forbes China as one of the top venture capitalists since 2011.
For AI founders evaluating early investors, fund size is a starting filter, not a deciding one. In practice, thesis fit, operating involvement, and cross-market capability tend to matter more over a five-year horizon.
Why Pre-Seed and Seed Still Have Room
Pre-seed and seed funding for AI startups has held up better than later stages through the current cycle, for a few reasons that aren’t obvious from the headlines.
Smaller check sizes mean lower absolute risk. A seed bet that doesn’t work out costs considerably less than a Series B that stalls. Beyond that, early-stage AI companies aren’t yet fighting for mindshare, which is an advantage with a time limit. And since the AI infrastructure stack is still being built, genuine gaps at the tooling layer and vertical application layer exist now that weren’t there three years ago.
The funding window at seed stage is open. Getting through it, though, requires positioning that survives the “why not GPT-4o” question.
What Raising Looks Like in Practice
Seed rounds for strong AI startups are closing in 6-10 weeks when the fundamentals are in place. Series A timelines, by contrast, have stretched to 12-18 weeks as investors run deeper diligence on retention and unit economics. Pre-seed rounds with the right technical signal are sometimes moving faster than either.
What’s working:
- A working prototype that demonstrates the core AI capability, not a deck describing it
- A specific answer to the “why not a foundation model” question, grounded in data or distribution
- Domain expertise that’s genuinely hard to replicate: proprietary data, regulatory knowledge, deep customer relationships in a vertical
What’s stalling rounds:
- Horizontal AI features without a clear moat
- Valuations anchored to 2021-2022 comparables
- Teams without technical co-founders trying to build AI-native products
The Money Is There. The Question Is Whether You’re Building What It’s Looking For.
The AI funding market in 2025 rewards specificity. Capital is moving toward founders with technical depth, defensible distribution, and a clear answer to why a foundation model can’t replicate what they’re building in a product cycle.
For founders building something that fits that description, the environment is better than the cautious narratives suggest. For everyone else, more funding won’t fix the underlying positioning problem.
The gap between perception and reality runs through the distribution, not the headline. The total market number is real. The share of it that’s actually accessible to a typical AI startup in 2025, however, is a fraction of what that number implies.