Artificial Intelligence Has Lifted Global Markets. It May Now Be Their Largest Concentrated Risk

The A.I. investment cycle has rewarded shareholders across technology, energy, industrials and international markets. The same breadth that made the rally powerful has also made genuine diversification increasingly difficult.

Category: Strategy. Written by Jaime Garcia, Founder, SnowRock. Published July 21, 2026. 18 min read.

In short

• A portfolio can hold technology, industrial, energy and international stocks and still depend on one assumption: that A.I. spending keeps rising fast enough to justify today’s prices.
• The rally broadened without diversifying. Utilities, chipmakers, turbine builders and foreign indexes are increasingly expressions of the same capital-spending theme.
• The practical defense is to look through fund labels and sector names to find what is actually driving earnings, and to hold assets that do not depend on the A.I. cycle.

Artificial intelligence has become one of the most important forces in global financial markets, and its influence extends far beyond the companies building large language models or manufacturing advanced semiconductors.

The construction of data centers has created new demand for electricity, turbines, cooling systems, construction equipment, networking infrastructure, natural gas and industrial materials. Companies that once appeared largely disconnected from software have become indirect expressions of the same investment thesis.

Caterpillar is one example. The company is best known for heavy construction and mining machinery, but it also manufactures equipment used in power generation, and as data-center developers seek enormous quantities of dependable electricity, that segment has become more strategically valuable. Investors have responded by treating Caterpillar not only as an industrial company but as a beneficiary of the A.I. infrastructure build-out. The same repricing has occurred across utilities, energy producers, semiconductor suppliers, electrical-equipment manufacturers and data-center contractors. This has broadened the A.I. rally without necessarily diversifying it.

An investor may own technology companies, industrial businesses, international stocks and energy producers while remaining heavily exposed to the same underlying assumption: that artificial-intelligence spending will continue rising fast enough to justify extraordinary capital investment and elevated asset prices. That dependence is becoming one of the defining risks in modern portfolios.

The Market Has Become an A.I. Supply Chain

The first phase of the rally was concentrated in a small group of dominant technology companies. Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla attracted enormous amounts of capital as investors sought direct exposure to artificial intelligence, cloud infrastructure, advanced computing and automation. The trade then expanded outward. Semiconductor designers and manufacturers rose because every significant A.I. system depends on advanced processors; memory-chip producers benefited from the need to move and store vast quantities of data; and semiconductor-equipment companies gained because chipmakers needed more sophisticated factories.

The next phase reached physical infrastructure. Data centers require power plants, turbines, transformers, cooling systems, transmission equipment, backup generation, land and construction, and the expansion therefore lifted companies that would not ordinarily be categorized as technology investments. Utilities became A.I. beneficiaries. Natural-gas producers became A.I. beneficiaries. Construction-equipment companies became A.I. beneficiaries. Electrical-component manufacturers became A.I. beneficiaries. The market increasingly began to treat any company positioned near the build-out as part of the same economic cycle. This created the appearance of a broad rally, but beneath the surface, much of the performance is connected to one dominant capital-spending theme.

Market concentration risk: strong returns, narrow foundations

Investors in U.S. equity funds have enjoyed exceptional recent gains. For the three months ending June 30, the average domestic stock fund returned approximately 14.8 percent, the strongest quarterly performance since the recovery following the initial pandemic decline in 2020. Over the twelve months through June, domestic stock funds gained approximately 23.2 percent. Smaller growth companies performed even more strongly, with small-cap growth funds returning about 24.4 percent for the quarter and 33.6 percent over the year.

Those numbers suggest a healthy market expanding beyond its largest companies, but the underlying drivers are more concentrated than the category labels imply. A smaller industrial supplier selling equipment to data centers may be counted as a small-cap stock. A utility expanding generation for A.I. campuses may sit inside a defensive sector fund. A chip supplier based outside the United States may appear in an international allocation. The classifications differ; the economic exposure may not.

A portfolio can contain many securities and still depend on a narrow group of assumptions. In the current market, those assumptions include continued growth in computing demand, sustained capital spending by large technology companies, adequate access to electricity and a future stream of A.I.-related profits large enough to validate present valuations. If those assumptions weaken together, the number of holdings may provide less protection than investors expect.

A Rally That Rotates

The A.I. trade has not advanced in a straight line; it has moved through the market in waves. At first, investors favored the companies developing models, cloud platforms and processors. Then attention shifted toward semiconductor manufacturing and equipment. Later, capital moved into utilities, energy producers and industrial companies expected to support data-center construction. Some software companies suffered during the same period, as investors began to question whether A.I. agents could replace functions currently sold through conventional enterprise applications; businesses such as Salesforce and Oracle faced concern that intelligent systems might reduce customer dependence on established software products or compress the pricing power of subscription platforms.

Microsoft occupied both sides of the trade, benefiting from its position in cloud computing and artificial intelligence while also facing questions about whether new interfaces could disrupt parts of the conventional software market. This illustrates the reach of the theme. Artificial intelligence can raise a company's valuation because it creates demand for its products, and it can lower another company's valuation because it threatens to make those products less necessary. Avoiding A.I. exposure is therefore difficult, since even businesses that do not develop or supply the technology may be judged according to whether it will strengthen or undermine their economic model. The market is doing more than investing in artificial intelligence. It is repricing much of the corporate economy around it.

International Diversification Is Less Diversified Than It Appears

Investors often turn to foreign stocks when the U.S. market becomes expensive or concentrated, and international equities have recently delivered strong returns: international stock funds gained approximately 12.8 percent in the second quarter and 26.8 percent over the twelve months through June, outperforming domestic stock funds over the longer period, while emerging-market funds rose approximately 22.4 percent for the quarter and 45.9 percent over the year. Those gains might appear to provide diversification away from the American A.I. boom. In many cases, they extend it.

The largest holdings in broad international index funds include Taiwan Semiconductor Manufacturing, Samsung Electronics, SK Hynix, ASML and Tencent, and each occupies an important position in the artificial-intelligence ecosystem. Taiwan Semiconductor manufactures many of the most advanced processors. Samsung and SK Hynix supply memory essential to high-performance computing. ASML produces the lithography equipment required to manufacture leading-edge chips. Tencent participates in cloud computing, software and large-scale A.I. development. An investor may move capital outside the United States and still remain deeply exposed to the same technological cycle.

Since April, more than half of the return of the FTSE All-World Index reportedly came from companies directly associated with artificial intelligence, and when the broader technology sector is included, the contribution rises substantially further. Global diversification has therefore become geographically broader without becoming economically independent. The companies are located in different countries; their earnings are increasingly tied to the same source of demand.

Correlation Is Rising Across Markets

Diversification works best when assets respond differently to economic events, so that weakness in one market is offset by stability or strength elsewhere, because different countries, industries and asset classes are influenced by different earnings cycles, policies and risks. That structure has weakened. Global markets are increasingly synchronized by shared exposure to artificial intelligence, semiconductor demand, U.S. technology spending and geopolitical events. A change in the capital budgets of several large American technology companies can affect chipmakers in Taiwan, equipment manufacturers in Europe, memory suppliers in South Korea, utilities in the United States and commodity producers around the world. A disruption in the Persian Gulf can affect energy prices, inflation, bond yields and the cost of operating data centers. A restriction on semiconductor exports can alter valuations across multiple continents.

A portfolio like this is diversified by name and concentrated by narrative. The market remains geographically global, but its principal economic drivers are becoming more concentrated. An investor may own hundreds or thousands of companies through broad index funds while receiving less risk reduction than the number of holdings suggests.

Why the Current Boom Is Not the Dot-Com Bubble

Comparisons with the late-1990s technology boom are unavoidable. Then, as now, investors assigned enormous value to a technology expected to transform the economy, companies raised capital aggressively, new business models appeared rapidly, and traditional valuation measures were often dismissed as inadequate. There is one important difference: many of the companies leading the present market are extraordinarily profitable. Microsoft, Alphabet, Meta, Amazon and Nvidia generate substantial cash flow, their businesses are established, their balance sheets are strong and their customer relationships are global. The market is not relying entirely on speculative companies with no meaningful revenue, which makes the current cycle more economically credible. It does not eliminate valuation risk.

A profitable company can still become a poor investment when its share price assumes a level of future growth that cannot be achieved. Strong earnings support a valuation; they do not justify any valuation. The critical question is not whether the leading A.I. companies are real businesses, because they are. The question is whether the profits generated by artificial intelligence will eventually be large enough to support the total amount of capital committed throughout the supply chain, including not only model developers but chips, data centers, energy systems, transmission infrastructure, cooling equipment and the expanding group of businesses whose valuations now depend on A.I. demand.

The Capital-Spending Problem

The artificial-intelligence economy is being built through one of the largest capital-spending programs in corporate history. Technology companies are committing hundreds of billions of dollars to data centers, processors and power contracts; utilities are planning new generation; industrial suppliers are expanding factories; and governments are offering incentives for semiconductor and infrastructure projects. The spending itself creates revenue. A company selling turbines can benefit before the data center produces a dollar of A.I.-related profit, and a chipmaker can record extraordinary demand while the company purchasing those chips is still searching for a sustainable business model. This creates a gap between infrastructure revenue and end-market economics.

During the construction phase, the cycle can appear self-validating. Technology companies spend more, suppliers report stronger earnings, their share prices rise, and higher valuations make further capital raising easier, a process that continues as long as investors believe future demand will justify the investment. The danger emerges if commercial adoption fails to produce sufficient returns. Customers may use artificial intelligence extensively without paying enough to cover the enormous cost of providing it; competition may drive model prices downward; efficiency improvements may reduce the need for additional hardware; and enterprise adoption may take longer than expected. If any of those conditions develop, capital budgets could be revised sharply, and the businesses farthest from the end customer may discover that their strongest demand was driven by a temporary construction cycle rather than a durable earnings base.

The Backdoor Exposure Problem

Many investors know when they own Nvidia or Microsoft. They may be less aware of their A.I. exposure through industrial, utility and international funds, and this indirect exposure matters because portfolio construction often relies on sector labels. An investor may deliberately reduce technology holdings and add utilities for stability, but if those utilities have been repriced primarily because they are expected to serve data centers, the portfolio may remain tied to the same A.I. investment cycle. The same is true of energy companies: oil and gas stocks may appear to hedge technology exposure because they respond to commodity prices, but when part of the investment case depends on growing electricity demand from A.I. facilities, the diversification benefit becomes less complete.

Industrial funds may include equipment manufacturers, electrical-component companies and engineering businesses whose growth forecasts now assume sustained data-center construction, and international funds may be dominated by semiconductor supply-chain companies. The result is hidden concentration. The investor has reduced direct ownership of A.I. developers while increasing ownership of the infrastructure supporting them. The form of the exposure changes; the dependence remains.

Bonds Have Provided Stability, but Not Immunity

Bond returns have been more modest than equity returns, as would normally be expected. Taxable domestic bond funds gained approximately 1.8 percent during the quarter and 5.1 percent over the twelve months through June, while municipal bond funds rose roughly 2.4 percent for the quarter and 6.7 percent for the year. These returns have helped balanced portfolios, because bonds can provide income, reduce volatility and preserve capital during equity-market declines. Their value as protection, however, depends on the reason stocks are falling. When economic weakness causes equity prices to decline and interest rates to fall, high-quality bonds often rise; when inflation drives both interest rates and discount rates higher, stocks and bonds can fall together, as occurred in 2022.

The present environment still contains meaningful inflation risk, because energy disruptions, trade restrictions, infrastructure spending and tight supply chains can push prices higher, and if central banks respond with higher interest rates, longer-duration bonds may decline even as equity valuations come under pressure. Bond investors should therefore distinguish between long-term allocation and near-term liquidity. An investor who can hold high-quality bonds for many years may be less concerned about temporary price changes, while someone who expects to need the money soon may be better served by shorter-duration instruments such as Treasury bills, money-market funds or certificates of deposit. Bonds remain a critical source of diversification. They are not a guaranteed hedge against an A.I.-related market decline, particularly if that decline occurs alongside inflation or rising interest rates.

Balanced Funds Have Benefited From the Same Cycle

Funds combining stocks and bonds have also generated strong returns. Asset-allocation funds holding between 50 and 70 percent in equities returned approximately 8.7 percent for the quarter and 14.5 percent over the year; target-date 2035 retirement funds gained roughly 9.4 percent for the quarter and 16.5 percent over twelve months; and retirement-income funds, which generally hold larger bond allocations, returned approximately 5.1 percent for the quarter and 10 percent over the year. These results demonstrate the strength of the broader market environment, but they should not be interpreted as proof that portfolio risk has disappeared.

Target-date and balanced funds frequently own broad index portfolios, and those indexes inherit the concentration of the markets they track. When the largest companies rise, their weights increase automatically, so the better a group of companies performs, the larger its share of the index becomes. A retirement investor may therefore become more exposed to the dominant A.I. companies without making an active decision. This is one of the paradoxes of market-capitalization-weighted investing: success itself creates concentration.

Valuation Is the Central Constraint

The strongest argument for caution has little to do with whether artificial intelligence creates economic value, which it clearly does. The concern is that the market may already be pricing in a great deal of that value before it has arrived. Stock prices reflect expectations about future earnings, and when expectations become exceptionally high, even strong results can disappoint: a company may increase revenue by 30 percent and still fall sharply if investors expected 40 percent. The present market contains numerous signs of extraordinary optimism. Businesses connected to data centers have received premium valuations, semiconductor companies have been priced around sustained demand, large technology firms are expected to convert infrastructure spending into durable, high-margin revenue, and some public offerings have reached valuations difficult to support through conventional earnings analysis.

That does not guarantee an imminent decline. Expensive markets can become more expensive, momentum can continue for years, and investors who left the market early during previous technology cycles often missed substantial gains. Valuation is a poor short-term timing tool. It is an important long-term risk indicator: the higher the starting price, the more future success must already be assumed.

What Could Break the Trade

The A.I. investment cycle could weaken through several channels, any one of which could slow it, and several of which occurring together could reverse it.

1. Slower commercial adoption Companies may experiment heavily with artificial intelligence without integrating it deeply enough to justify large recurring spending.
2. Pricing compression Competition among OpenAI, Anthropic, Google, Meta, Chinese laboratories and open-source developers may drive the price of model access downward.
3. Efficiency gains New processors, smaller models and improved software could reduce the computing power required for a given task, weakening forecasts for endless infrastructure expansion.
4. Energy constraints Data centers may be delayed by electricity shortages, transmission bottlenecks, permitting disputes or public opposition.
5. Capital discipline Shareholders may pressure technology companies to reduce spending if A.I. revenue fails to grow quickly enough.
6. Regulation Governments may impose restrictions on model development, data-center construction, chip exports or the use of A.I. in sensitive industries.
7. Geopolitical disruption Conflict involving Taiwan, the Middle East or major trade routes could damage semiconductor and energy supply chains.
8. Higher interest rates Inflation could keep borrowing costs elevated, reducing the present value of future profits and making infrastructure projects more expensive.

A.I. Risk Is Not Limited to Technology Stocks

A correction in the A.I. trade would not remain confined to model developers or semiconductor companies. Utilities could fall if projected electricity demand is revised downward. Industrial companies could suffer if data-center orders are canceled. Energy producers could lose part of the premium associated with long-term power demand. International markets could decline through semiconductor exposure. Private-credit funds could face losses on infrastructure financing. Commercial real estate tied to data-center development could weaken. Even government finances could be affected where tax incentives and public infrastructure were committed to projects that no longer appear economic. The expansion of the trade has increased its systemic importance, and a narrow technology disappointment could become a broad capital-spending contraction. This is why the market’s greatest source of recent strength may also be its largest emerging vulnerability.

Diversification Requires Looking Through the Labels

The solution is not necessarily to abandon artificial-intelligence investments. The technology may create enormous long-term economic value, and avoiding every company connected to it could exclude many of the strongest businesses in the world. The more practical objective is to understand total exposure, looking through fund names and sector classifications to identify what is actually driving earnings.

These questions reveal economic concentration that conventional allocation charts may hide. A portfolio can appear balanced across sectors while remaining dependent on one investment regime.

The Case for Broad Ownership Still Holds

None of this creates a reliable argument for exiting the stock market. Short-term market movements cannot be predicted consistently, and investors who sell because valuations appear high may miss further gains and struggle to decide when to return. Broad equity ownership remains one of the strongest long-term methods of participating in economic growth. The discipline lies in avoiding dependence on a single narrative, which may require holding assets whose recent performance appears unexciting: high-quality bonds, cash, short-term government securities and companies whose earnings do not rely primarily on the A.I. infrastructure cycle. It may also require accepting that international equities provide less diversification than they once did.

Diversification is rarely most appealing when it is most necessary. The assets offering the greatest protection often appear inferior during a powerful bull market: cash trails equities, bonds seem dull, and unrelated industries underperform the dominant theme. Their value becomes clearer only when the theme reverses.

The Investor’s Real Challenge

Artificial intelligence has created a difficult portfolio problem. The companies benefiting from it include some of the most profitable, innovative and strategically important businesses in the world; their momentum is strong and their influence continues to expand. Excluding them entirely would be a substantial active bet against one of the most consequential technological developments of the era. Owning them through broad market funds is reasonable. Failing to recognize how much of the portfolio now depends on them is not.

The market's concentration is no longer visible only through the weight of several giant technology companies. It runs through semiconductor supply chains, power generation, industrial equipment, energy markets and foreign indexes, and it has spread faster than most investors have moved to diversify away from it.

The technology may continue exceeding expectations; infrastructure spending may remain strong; corporate adoption may create profits large enough to validate today’s prices. But the market has reached a point where merely meeting optimistic expectations may not be enough. A.I. has rewarded investors by becoming the dominant story across global markets. It may now threaten them for the same reason.

The SnowRock read

We are not investment advisers, and this is not investment advice. We publish it because the failure mode it describes, a system that looks diversified but rests on a single assumption, is the exact one we spend our days preventing inside companies. Swap portfolios for AI stacks and the warning is nearly word for word. A business can run six vendors, three clouds and a dozen models and still have bet everything on one provider’s pricing, one model family’s availability, or one assumption about how cheap inference will stay.

So the discipline we press on operators is the one this analysis presses on investors: look through the labels. Map what your AI actually depends on, not what the architecture diagram says it depends on. Keep your data and evaluation in your own hands, keep a real alternative for every critical model, and make sure the thing that would hurt most if it changed is something you chose on purpose rather than something you inherited from a diagram that looked reassuringly spread out. The concentration you can see is a decision you can manage. The concentration you cannot see is the kind that does the damage.