The AI Scaling Law: Plateau Ahead? Implications for Chinese and US investors

Exploring the Risks and Asymmetric Opportunities in U.S. and Chinese AI Investments.

The AI Scaling Law: Approaching a Plateau?

If you’ve followed AI circles, you’ve heard of the “scaling law.” It’s the idea that the bigger your dataset, the more parameters you train, and the more compute you use, the better your model becomes. Recently, some experts have suggested we’re reaching a plateau. It’s like going to the gym: at first, each extra hour on the treadmill pays off, but eventually, your gains start to level off. If (and that’s still a big if) these diminishing returns are real, it could upend the current assumption that anyone can just throw GPUs at a problem to dominate.

This could have huge implications for investors around the globe. As an investor, my role isn't to identify the best companies but to find the best investments. This means seeking out mispricings — situations where the potential upside far outweighs the downside risk. These opportunities often arise when the market is overly pessimistic, creating a scenario where even marginal improvements can deliver significant returns. Conversely, I avoid investments that are priced for perfection, as even the slightest deviation from expectations can lead to substantial losses.

If the AI scaling law really is hitting a plateau, it might just swing the door wide open for companies that focus on resourcefulness and smart innovation, rather than merely throwing mountains of data and computing power at the problem.

USA: CapEx and Multiples

Nowhere is the excitement about AI more palpable than in the United States. Tech giants like Microsoft, Google, Amazon, and Meta are shoveling billions into AI. Microsoft alone plans to invest over $100 billion in cloud and GPU resources by 2027, while Nvidia’s chips remain indispensable.

Yet these high-dollar commitments come with high expectations. Valuations have ballooned, partly because the market treats AI growth as inevitable. Capital expenditures for Microsoft, Google, and Meta exceed their depreciation expenses by over three times, which will eventually show up in growing depreciation down the road. That means less free cash flow—and if AI scaling hits a plateau or becomes commoditized, these investments could feel a bit like sinking money into a black hole.

As we kick off the new year, I’m making a small but exciting change to this newsletter. I’m introducing paid subscriptions! Free content will still be here for everyone, but starting now, subscribers can unlock exclusive insights, more in-depth analysis, and full access to all articles. This step allows me to dedicate even more time and energy to creating thoughtful content for you.

If you’ve enjoyed reading so far and want to support the work I do, consider subscribing. It’s a great way to stay connected and get even more value. Thanks for being part of this journey, and here’s to an amazing 2025!

Will There Be Money to Be Made? How About First-Mover Advantage?