When the 'God of VC' Says Money Is Leaving: What Should We Listen To?

I noticed an interesting detail: Neil Rimer said that line at the Tech Festival in Athens, and TechCrunch chose to write it out two months later. It's not that the news wasn't fast enough, but that the statement needed time to ferment—it touched upon the assumption the entire AI investment circle is least willing to face: Our perceived "infinite ammunition" might be turning.

[!abstract] Core Assumption

Capital density in the current AI sector is approaching historical extremes, while the marginal returns of technological breakthroughs are diminishing. Capital "flowing back" from AI is not a retreat, but a redistribution.

Let's Look at Three Data Points to Restore Rimer's Context

1. In 2024-2025, global AI startup fundraising accounted for over 40% of all VC funding, while AI companies' revenue contribution was less than 5% of global software spending.

2. Market cap increases of publicly traded AI-related companies (like Nvidia, Microsoft, Google) far exceeded their actual profit growth, showing typical P/E multiple inflation.

3. Within Index Ventures' own portfolio, the share of AI companies rose from 15% in 2022 to 35% in 2025, but the average exit time for late rounds stretched to 7.2 years.

Rimer is no amateur; he's a co-founder of Index Ventures, having invested in Skype, Dropbox, and Robinhood. He has witnessed firsthand how "tides change direction" in capital cycles. When he says "money will come back," he isn't bearish on AI, but bearish on the narrative of treating "AI as a panacea."

Deconstructing the Logic of "Capital Flow Back" from Three Dimensions

Dimension 1: Supply Side—Excess Capital, Scarce Quality Targets

Over the past two years, there has been too much money in the AI track and too few good projects. Massive funds flooded into "wrapper-type" applications (using GPT APIs with a layer of packaging), which lack moats. When capital costs rise (interest rates remain high), LPs demand more certain returns, not "AI concept" PPTs. Rimer's subtext is: "Junk projects" in AI will be drained first, and capital will flow to companies with real data flywheels and vertical industry know-how.

Dimension 2: Demand Side—Enterprise Procurement Enters "Value Verification" Phase

In my circles with CIOs, starting in H2 2025, enterprise clients' procurement decisions for AI products slowed noticeably. They are no longer moved by "what AI can do," but ask "how much cost reduction does your product offer compared to existing solutions." When AI shifts from a "stimulant" to a "cost-reduction tool," only products that directly reflect in financial metrics will survive. Capital flowing back is essentially the market punishing "fake AI" and rewarding "real AI."

Dimension 3: Exit Side—IPO and M&A Windows Narrowing

The number of AI company IPOs in 2025 plummeted, even falling below 2022 levels. The reason is simple: Public market investors no longer pay for "losses for growth." Benchmarking against the Web3 bubble, when liquidity recedes and the last bag holders disappear, capital withdraws from tracks with "high valuations but no profits." Rimer's words sound more like a leading indicator: Surviving AI companies must prove positive unit economics.

Core Contradiction: Technology Cycle vs. Capital Cycle

AI technological breakthroughs are still on a downward cost curve (inference costs drop 90% every 12 months), but the capital cycle has entered a "tightening" phase. The mismatch between these two cycles is the biggest structural risk currently.

• Technology optimists believe: AI will bring 10x efficiency gains to all industries; investment returns are just a matter of time.
• Capital pessimists believe: Even if the tech works, the "last mile" from lab to commercialization takes 5-10 years, while current LP patience lasts only 2-3 years.

Rimer stands in the middle, leaning pessimistic. His "money will come back" doesn't mean capital disappears, but that capital will shift from "betting on the future" to "betting on the present"—withdrawing from infrastructure layers (GPUs, Cloud) and flowing to application-layer enterprises that can generate cash flow quickly.

Benchmarking Overseas Cases: 1999 Internet Bubble vs. 2025 AI Bubble

Using Porter's Five Forces framework, the supplier bargaining power in the AI industry (from Nvidia, Microsoft) has reached historical peaks, while buyer bargaining power (enterprise clients) is rising. This is a typical characteristic before a bubble burst.

• 1999: Infrastructure providers like Cisco and Microsoft made fortunes, but many .com companies went bankrupt.
• 2025: Nvidia and Microsoft remain the biggest winners, while AI startups face "survivorship bias."

Rimer's subtext might be: Rather than investing in AI application companies, invest in the "water sellers" of AI infrastructure. But even so, infrastructure valuations have already discounted the next 3-5 years.

A Clear Trend Prediction

Within the next 18 months, total venture capital in the AI sector will experience its first year-over-year decline, with a drop of 15-25%. But this isn't the end of AI; it's the squeezing out of bubbles. The real opportunities lie in "AI + Vertical Industries" among companies that can provide ROI data, not generic "AI assistants."

There are three directions for capital flow back:

• Return to "Hard Tech": Shifting from Generative AI to AI for Science, AI...

Original Link: https://techcrunch.com/2026/07/17/neil-rimer-thinks-the-ai-money-is-coming-back-out/