In June last year, Sequoia’s David Cahn published a provocative post titled ‘AI’s $600B Question’. The post was a follow up to a previous blog, pointing out ‘a big gap between the revenue expectations implied by the AI infrastructure build-out, and actual revenue growth in the AI ecosystem’. One of the responses to that piece, the author said, ‘was that “GPU CapEx is like building railroads” and eventually the trains will come, as will the destinations—the new agriculture exports, amusement parks, malls, etc.’

However, as Cahn suggested, that rebuttal was revealing. With railways, the infrastructure can be valuable for the operator that builds it. Owning the tracks between important destinations could give you a degree of monopoly pricing power, given the limitation on how many rival tracks could easily built. By contrast, “AI clouds continue to flood the market”. And without a monopoly or oligopoly (as with CPU clouds) “high fixed cost + low marginal cost businesses almost always see prices competed down to marginal cost”.

The analogy goes further. In the case of actual railroads, as with many new technologies, “speculative investment frenzies often lead to high rates of capital incineration”. Rail networks proved to be fundamentally transformative technology, but speculating in new rail companies was a risky business that frequently led to investors losing their shirts.

Overhyped or the next technology revolution?

Fast forward to January 2025 and the assessment seems prescient – both in its healthy scepticism but also, perhaps, in its optimism.

The launch of DeepSeek, developed by a Chinese AI company, sent US tech stocks into a tailspin. By providing advanced AI technology – at apparently a fraction of the cost, and using less sophisticated chips – raises questions about the prevailing assumption that scaling AI depends on massive ongoing investment.

It’s no wonder that investors in Nvidia and other technology providers are feeling a little shaken (though the actual implications of DeepSeek’s launch are far from clear at this point). But what do these developments augur for the future of AI?

The cost of computation

Consider another document that created as sceptical buzz in June last year: a Goldman Sachs report titled ‘Gen AI: too much spend, too little benefit?’

The report features an interview with Goldman’s Head of Global Equity Research, Jim Covello. According to Covello, the bank estimated that AI infrastructure buildout would cover over $1 trillion over the next several years. “So, the crucial question is,” Cavello asked, “What $1tn problem will AI solve?”

More subtly, he cautions against facile comparisons to the development of the internet:

Many people attempt to compare AI today to the early days of the internet. But even in its infancy, the internet was a low-cost technology solution that enabled e-commerce to replace costly incumbent solutions. Amazon could sell books at a lower cost than Barnes & Noble because it didn’t have to maintain costly brick-and-mortar locations. Fast forward three decades, and Web 2.0 is still providing cheaper solutions that are disrupting more expensive solutions, such as Uber displacing limousine services. While the question of whether AI technology will ever deliver on the promise many people are excited about today is certainly debatable, the less debatable point is that AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn’t designed to do.

To those who assume technology costs will automatically get lower, Cavello warned that we shouldn’t take for granted that AI costs will become substantially lower over time. Indeed, lower computing costs were the result of intense competition, whereas Nvidia’s dominance as a monopoly seemed secure.

Who benefits?

Now the obvious irony: investor confidence in the AI project has been shaken by the very thing that AI pessimists were worried wouldn’t happen. That is, AI has (apparently) gotten cheaper than expected, more quickly than expected.

If DeepSeek’s promise is real, then the cost of AI compute has come down dramatically. That sounds like a positive for the development of AI as a technology that will transform the way we do business and manage real world problems. Indeed, it sounds like the kind of innovation that drives significant long term economic growth.

Of course, if you’re an investor in US tech stocks, you may have more immediate concerns about your portfolio. And, to be fair, the health of US tech stocks is not a niche concern, considering the concentration risk apparent in the tech-heavy S&P500.

No wonder so much internet traffic is being hastily directed to the Wikipedia entry for the Jevons Paradox.

All opinions, news, research, analysis, prices or other information is provided as general market commentary and not as investment advice and all potential results discussed are not guaranteed to be achieved. The information may have been derived from publicly available sources, company reports, personal research, or surveys. Past performance is not indicative of future performance. Trading carries risk of capital loss. Service available to professional clients only.