Big Cloud? Small Cloud? Which one should you be optimistic about?

👆Click the blue text to follow us

The Big Three Clouds

In the first phase, the market rewards companies that make chips and provide scarce computing power. Now, as massive capital expenditures gradually turn into revenue, which companies can truly emerge as the winners in the second phase? We looked to institutions for some strategic inspiration.

According to Bloomberg, Wellington Management, which manages about $1.3 trillion in assets, and Capital Group, which manages about $3.6 trillion, are both planning to increase their allocation to hyperscale clouds like Microsoft, Amazon, and Google while continuing to hold semiconductors. Bloomberg argues that the logic behind this barbell strategy is that AI investment is shifting from one-off capital expenditure to a phase of validating business returns.

The most recent earnings reports reinforced this view. Cloud businesses at the Big Three Clouds accelerated again, and supply of computing power remained tight, which temporarily eased market concerns that massive capital expenditures would not generate returns. Brian Barbetta, head of tech investing at Wellington Management, noted that hyperscale cloud providers remain core holdings in their investment portfolio, and their team has recently increased allocations to several such companies.

Now that you have semiconductors in the left hand and the dominant cloud giants in the right, what about the smaller cloud companies in the middle? Both CoreWeave and Nebius have seen impressive gains this year, so why are large institutions reluctant to allocate to smaller clouds?

Bloomberg believes the core reason is the shortage of computing power. Clients are eager to get GPUs and datacenter capacity, and whoever has compute power available for immediate delivery can command a higher leasing price. However, this raises a question: how long will this scarcity of computing power last?

Noah Weisberger, head of US equity strategy at BCA Research, proposed a straightforward pair trade: go long the Big Three Clouds and short the small clouds. The main idea is that once new data centers and GPUs come online at scale, compute leasing prices will quickly return to normal, and the high-price window that smaller cloud providers rely on will narrow.

On top of that, small cloud companies face another layer of pressure. As he points out, many projects rely on debt, leasing, and external funding, meaning they have to shoulder ongoing interest, equipment depreciation, and construction costs. If computing prices fall, revenues will be pressured first, but debt and depreciation costs won't fall at the same time. At that point, small clouds will be squeezed from both ends, and that's the real stress test.

By comparison, the Big Three Clouds not only have large-scale infrastructure, but also self-developed chips and long-term enterprise relationships, and can enable clients to choose models, deploy applications, manage data, and control costs. Customers aren’t just buying GPUs, but also the cloud platform, security services, software tools, and ongoing operations support. These advantages constitute the moat of the Big Three Clouds, and will be more evident over a longer period.

Richard Clode, fund manager at Bankers Investment Trust, has made Amazon the largest overweight position in his fund. Clode expects that, from later next year through 2028, the profit and cash flow growth rates of the major cloud providers could gradually outpace their capital spending growth rates. Today's capital expenditure will become tomorrow's revenue.

Alright, that's enough about clouds for now. The next question is: should we give up on semiconductors? Is it an either-or choice between chips and cloud?

John Lamb, Director of Equity Investments at Capital Group, believes there is no need to choose sides. This is an ecosystem that is expanding: semiconductor companies provide the foundation of computing power, while the hyperscalers turn that computing power into long-term services. As long as AI demand keeps growing, both types of companies can be beneficiaries. That’s why the barbell strategy of holding both makes more sense.

Of course, the profit cycles of the two types of companies are not the same. Lamb reminds us that it usually takes 12 to 18 months from a data center’s groundbreaking to generating revenue, so cloud providers turn capital expenditure into revenue much later than the semiconductor makers.

However, not everyone is bullish on AI. Alberto Conca, CIO of Swiss boutique wealth manager LFG+ZEST, estimates that AI's commercial revenue would need to increase 5 to 13 times from current levels just to support today's capital expenditures. This suggests that the market has already priced in very high AI adoption rates and customer willingness to pay. If corporate demand materializes slowly, or compute utilization starts to decline, the cloud giants will also face triple pressures from capital expenditure, depreciation, and valuations.

Moreover, as the AI market matures, the ultimate winners will be far fewer than today’s players. Real winners need to have technology, client relationships, and control over infrastructure—not those who mainly rely on temporary scarcity pricing, as their competitive advantage may gradually weaken as supply increases.

Jason believes this article is well worth referencing. For instance, the barbell strategy of holding both semiconductors and cloud giants is excellent; in the words of the Capital Group director, there's no need to pick sides.

We are still in the mid-stage of AI development, and both camps have a long way to go. The latest earnings results somewhat proved that previous capital expenditures can be recouped, and the sustainability of investments and cash flow has eased concerns about semiconductors' viability to some extent.

However, regarding the point in the article about going long the big clouds and short the small ones, I think the risk is too high. You can go long on both, or you can make money shorting small clouds, but their share prices are highly volatile in the short term. If the market goes the other way, losses can be huge, even leading to forced liquidation. For individual investors, this is an extremely risky strategy.

I also agree with the last point: scarcity pricing will diminish as supply improves. For example, small clouds are now shifting contracts from long-term to short-term to capture short-term leasing premiums, because compute rental prices are high right now. If you can earn five years’ revenue in one year, it’s hard to sign a five-year deal. So, if short-term compute prices rise further, small clouds will be among the biggest beneficiaries.

By contrast, the Big Three Clouds are moving contracts to longer terms with relatively lower compute pricing. The relationship between the big and small clouds is a bit like storage versus TSMC: one locks in profits via short-term premiums and long-term contracts, and the other balances interests via reasonable price increases. Which one values long-term customer interests and relationship stability more? The answer is obvious.

[Meitou Pro]

Regardless of the size of the cloud provider, Meitou Pro closely tracks and provides in-depth analysis.

Meitou Pro recently analyzed the financial reports of Microsoft and Google, updated Microsoft’s valuation model, and compared which among CoreWeave, Nebius, and IREN is the better investment among small cloud providers, along with related risks.

If you are interested in cloud, do not miss these in-depth analyses. The link is in the pinned comment. New users get 7 days free—feel free to check it out.

Invest in US stocks with Meitou! The number one Chinese US stock research brand!