Renowned tech investor Gavin Baker: Everyone is worried about over-building AI, but the real risk may be under-building.

In a recent podcast episode released by a16z, tech investor Gavin Baker had an in-depth conversation with a16z partner David George. Baker frankly stated that he visited the entire industry this summer, asking everyone the same question: "Can you tell me even one quantitative data point in your business that is deteriorating? Just one." The result—he found none.

While everyone is worried about the over-investment in AI, Baker believes that the market's assessment of AI demand is severely underestimated, and the real risk to be wary of right now may be insufficient investment . Currently, there are probably fewer than 10 million heavy AI paying users globally, while there are 1.5 billion knowledge workers worldwide. Demand diffusion is only just beginning, and supply is already severely constrained.

The fundamentals are accelerating, but the stock price is falling.

Baker observed a clear divergence: while the fundamentals of the AI industry continued to accelerate in July and August, AI stocks in the public market experienced a significant pullback during the same period.

"Overall, AI accelerated in July and accelerated again in August," Baker said. "Strangely, publicly traded stocks have plummeted over the past two months."

He specifically named the dynamics of several companies: OpenAI is clearly accelerating, open source is accelerating even faster, and Grok has also experienced a "rather dramatic acceleration" after launching Grokbot. He believes that Anthropic is in a pre-IPO quiet period, and its growth rate data is not yet transparent, but the rest of its operations are "accelerating."

Baker used an analogy: "You know the saying, 'You can drown even in a river that's only two feet deep on average.' There hasn't been much movement at the index level, but some AI stocks have already seen significant pullbacks—while the fundamentals are generally accelerating."

With fewer than 10 million heavy users, the spread has only just begun.

Baker believes that the current support for hundreds of billions of dollars in AI revenue comes from an extremely small user group.

"These companies currently have about $80 billion in revenue, but how many heavy paying users are behind them? I'd guess no more than 30 million, maybe even less," David George said.

Baker's assessment was more conservative: "Perhaps actually less than 10 million."

He cited data from his own company, Atreide, as an example: from March to August, internal token spending increased 100-fold. With only two people using Grokbot Enterprise Edition, token spending is expected to increase another 10 to 20 times within a month.

" There are 1.5 billion knowledge workers worldwide. We seem to be just getting started on the demand side, but we are already severely constrained by supply, " Baker said.

He also observed that AI-native companies are spending over 10% of their monthly workforce on tokens, while even the better-performing traditional companies are reaching 1%. "When I look at supply and demand characteristics, the supply-side question is 'Is this sustainable?' But when considering the demand side, I think it's quite clear."

The real risk: Insufficient construction

This is where Baker's biggest divergence from the mainstream market narrative lies.

"Everyone is worried about oversupply, but I'm more worried about a severe shortage," Baker said. "If that's the case, you might not see a decrease in the cost of accessing AI, but rather a significant price increase."

David George further pointed out that this supply shortage may continue until 2028, and political obstacles will further delay the original construction plan.

Baker cited a seemingly absurd point by Marc Andreessen— that token costs could increase tenfold. "It sounds ridiculous, but we do live in a supply-demand driven world. If demand expands dramatically and supply can't keep up, the premise changes completely."

What worries him even more is the social consequences of the supply shortage. " That could ironically lead to real inequality in computing power—large companies and the wealthy can afford it ," George said, "and those 'data center degradation growth theorists' will be complaining about it two years later, without realizing that they are the ones who caused it."

Baker responded directly: "It's precisely because they won't let us build data centers."

The payback period for computing power investment is less than one year, demonstrating an exceptionally strong economic logic.

Baker believes that the current economic logic of computing power investment is extremely rare in his investment career.

He cited data disclosed by Nebius and CoreWeave: launching 1 gigawatt of computing power costs about $50 billion, and customers can prepay 50% to 60%, or $25 billion to $30 billion. The remaining portion can be recovered more quickly after entering the spot market, with the overall investment recovery period being about 9 to 10 months.

"In my investment career, there have been very few opportunities to see a company deploy tens of billions of dollars and achieve a return in less than a year," Baker said. "It's quite rare."

He also pointed out that Nvidia GPUs can be funded, and the current funding costs are quite low, with institutions such as Blackstone, KKR, and Apollo participating at relatively low costs. "As these models improve and their lifespan extends, the monetization rate per gigawatt continues to increase, and the actual equity payback period may be far less than one year."

Baker cites Microsoft as a cautionary tale: Satya Nadella slowed down after pledging $80 billion in capital expenditures at Davos last year, "and now they regret it." In contrast, OpenAI opted for aggressive investment, "and the high returns have clearly proven it was the right decision, both in the short and long term."

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.