AI Hyperscalers Worth-A-Buy?

AI Returns On Investment

In our January 2025 post, we outlined our concerns about the returns on capital expenditures for AI and AI-related stocks.

We noted that while AI will be more significant than the Internet, that alone would not drive mega-cap performance in 2025.

That investors would want to see returns on investment by the end of 2025 or early 2026.

More specifically, investors would want to see business models and lifestyle changes driven by AI that deliver returns on investment.

Mega-cap Stocks

We are not surprised to note, therefore, that the Roundhill Magnificent Seven ETF posted its 2025 high on 29 October and closed 11% down from that high at the close of trade last week.

In the same period, Amazon has fallen 23%, and Microsoft has fallen 27%. Both are in technical bear market territory.

In contrast, the S&P 500 is 2% off its 2026 all-time high. (That said, the balance of probability does point to a correction in the first quarter or early second quarter).

While the market is likely in a late cycle and tech stocks underperform in late-cycle markets, the current underperformance of the hyperscalers more likely reflects growing concern about the long-term sustainability of AI-related capital expenditures.

Capital Destruction

Data centres are negative free cash flow: a low-margin, low-return-on-capital business.

The return on that investment will come from what the chips facilitate and, on that basis, the monetisation of AI is unlikely before 2027/28.

A significant amount of capital will be destroyed in the AI race, and not all hyperscalers will be winners. The challenge for investors is in identifying the winners.

The cost of utilising AI could collapse due to overcapacity, and it may make more sense to invest in companies that ultimately benefit from the capital expenditure rather than the companies that incur it.

While Nvidia’s performance is broadly in line with the Roundhill ETF, it faces its own challenges.

AI Race

The Trump administration’s decision to allow exports of the Nvidia H200 to China has not been fully embraced by Beijing.

China is determined to be more self-reliant. At the start of the year, China announced $70 billion in chip-sector incentives, in addition to an existing $50 billion fund.

China has instructed its tech companies to stop buying Nvidia chips for AI use and banned foreign AI chips from state-funded data centres.

The challenge for Nvidia wasn’t the restrictions imposed on their chips. The challenge is that the Chinese sector has moved on.

China’s focus is not on building a Chinese Nvidia. It believes it has that already in Huawei.

The focus is on building an alternative AI proposition for markets that want AI now, not necessarily perfect performance.

That could impact Nvidia’s ability to scale globally.