The AI Rally

In 2023 something began that markets hadn't seen since the dot-com boom: a rally driven by a single technology that promised to reshape the entire digital economy. Two years on, the question is no longer whether artificial intelligence will transform business, but how long it will take — and who will capture the value.

The S&P 500 rose more than 20% in 2024. But if you strip out the "magnificent seven" from the rest of the index, the result is far more modest. The AI rally is not a broad-market phenomenon: it is a concentration of capital in a handful of companies that convinced the market they own the infrastructure of the future.

The trigger was ChatGPT, but the real engine of the rally is the demand for compute. Training a model like GPT-4 took thousands of GPUs over months. Inference at scale — answering millions of queries a day — takes even more. This insatiable demand for chips and data centers created a virtuous cycle for infrastructure providers.

Nvidia went from a video-game company to the most valuable company in the world at certain points in 2024. Its CUDA architecture, built over 15 years, turned out to be exactly what language models needed. It is not easy to replicate.

The rally isn't only about hardware. The companies building on that infrastructure — and with the distribution to monetize it — captured enormous value too. Meta, Microsoft, Google and Amazon are investing hundreds of billions in AI capex. They wouldn't do it if they didn't believe the returns are coming.

Each has a different thesis. Microsoft bet on integration into the enterprise workflow. Meta on open source and mass distribution. Google on multimodal models and data dominance. The diversity of bets is, in itself, a sign that the space is large enough for multiple winners.

The prices above refresh every 30 seconds during market hours. What they show is not just a number: it is the synthesis of millions of investment decisions about who they think will win the AI race. And the market, for all its short-term irrationality, tends to capture real signals about the long run.

No rally lasts forever, and the AI one has real risks. The first is valuation: many companies are pricing in growth scenarios that assume a mass adoption not yet visible in their results. The second is regulatory: Europe already has the AI Act, and the U.S. is building its framework. The third is competition: China is not out of the race, and open-source models erode the incumbents' edge.

The fourth risk is the most underestimated: concentration. If 80% of the rally's value sits in seven companies, one disappointing quarter at any of them can drag the whole index down. We've seen this before. Diversification within the theme matters.

That said: the demand for AI compute infrastructure is structural, not cyclical. Models will keep getting bigger, more capable, cheaper to run. The hardware that runs them will stay scarce. And the companies with the distribution to monetize that capacity will keep capturing value. The timing is hard. The direction, less so.