Your stock portfolio may have more AI overlap than you realize
It's time to stress test your AI portfolio.
Owning Nvidia, CoreWeave and Vertiv can look like three different ways to invest in artificial intelligence. One sells chips, another rents computing capacity and the third supplies power and cooling equipment. Each has a distinct business and its own reason to succeed.
But all three can benefit from the same decision: somebody choosing to spend more money on a data center. Investors spreading money across those businesses may be diversifying a company's risk while concentrating exposure to one investment cycle.
Last week's selloff made that distinction worth revisiting. Concerns about data-center debt and higher borrowing costs unsettled AI-related shares. That does not mean demand is collapsing, or that these stocks will always move together. It does, however, raise a useful question: How many holdings in your portfolio rely on a continuation of the same spending boom?
The overlap is easier to understand by following a hypothetical project. A cloud operator raises money, secures power and installs computing equipment. Its suppliers receive orders for processors, networking, power systems and cooling. Customers eventually pay to use the computing capacity.
Those transactions produce revenue for different companies at different times. They also share a vulnerability. If funding becomes more expensive or a customer delays expansion, the effects can travel through the project. A supplier might face postponed orders while the operator must carry costs longer before seeing the expected revenue.
Nvidia (NVDA) shows how much a successful supplier can benefit. In its fiscal second quarter, which ended July 26, data-center revenue reached $89 billion out of $96.2 billion in total revenue. By my calculation, that is roughly 92%. Data-center revenue more than doubled from a year earlier.
That category extends beyond generative AI, so the percentage should not be mistaken for a precise measure of AI exposure. Still, the business mix makes the importance of infrastructure spending clear. Nvidia also reported a 75% gross margin, demonstrating substantial profitability at its current scale. A slowdown in orders would reach a business with very different economics from a customer borrowing heavily to install its equipment.
CoreWeave (CRWV) illustrates that difference. Its second-quarter results showed $2.575 billion in revenue and $640 million in net interest expense. The latter was equivalent to about a quarter of revenue. Its approximately $104 billion revenue backlog offered visibility into future business, but the company noted that booking it as revenue depends on delivery and service availability requirements.
The gap between spending and collections matters. CoreWeave generated $679 million in operating cash flow during the quarter while spending $6.422 billion in cash on property and equipment, including capitalized internal-use software. That comparison excludes other investing activities and is not a complete measure of its financing needs. It nevertheless shows why access to capital matters alongside demand.
Vertiv Holdings (VRT) provides another set of economics. The power and cooling supplier reported second-quarter sales of $3.274 billion, up 24% from a year earlier. Operating cash flow reached $1.1 billion, and the company said it ended the quarter with a net cash position.
Those figures can distinguish Vertiv from a heavily indebted operator. They do not remove its exposure to customer spending. Vertiv attributed minor revenue timing shifts to temporary supply-chain congestion and the phased execution of increasingly large projects. Even during strong demand, deployment schedules can affect when suppliers record sales.
The investment case for owning different parts of this market remains credible. A supplier can serve competing operators. A well-financed customer can continue building when a weaker competitor retreats. Equipment upgrades can create demand within existing facilities. Choosing a power supplier also avoids having to predict which particular AI application will win.
But reducing one uncertainty does not eliminate another. A business can serve many customers whose expansion plans depend on similar assumptions about future computing demand. Customer diversity and spending-cycle diversity are different protections.
Don't miss: Could an Anthropic IPO put your 401(k) at risk? The AI safety debate hits a new dimension.
For a portfolio, I would turn that distinction into three questions.
First, what pays for each holding's growth? Write down the spending decision behind expected sales. If several answers are some version of "more data-center construction," add those positions together when considering exposure. Include relevant holdings inside funds, while recognizing that a broad index also contains businesses with other sources of earnings.
Consider a hypothetical $100,000 portfolio with $10,000 in each of the three companies. No individual position exceeds 10%, yet $30,000 sits in businesses exposed to the build-out. If each position fell 30%, those holdings alone would subtract $9,000, or 9% of the starting portfolio value, assuming everything else stayed unchanged. That is a stress test, not a prediction. The stocks could perform very differently. The test's purpose is to make a shared vulnerability visible before deciding whether the rest of the portfolio provides enough protection against it.
Second, who needs fresh financing? Compare operating cash flow with investment spending, then examine debt maturities and commitments. Companies that can fund expansion internally have more flexibility than those dependent on lenders or new shareholders. That difference can affect both survival through a downturn and the share of future profits existing owners retain.
Third, what happens if growth merely slows? An investment does not require a collapse in demand to disappoint. A share price built around rapid expansion can struggle when revenue keeps rising at a less impressive rate. The price paid matters alongside the quality of the underlying business.
These questions will not produce a precise correlation forecast or a universal allocation limit. They can reveal whether a new purchase adds a different source of returns or increases an existing bet.
An investor may reasonably decide that AI infrastructure deserves a large place in a portfolio. The important step is to make that concentration deliberate. Before buying a third beneficiary of the build-out, I would ask what it contributes if the spending cycle weakens. A different product provides useful diversification only to the extent that it changes the risks you actually own.
The real test is whether your portfolio can still hold up when its biggest shared assumption turns out to be wrong.
See also: Time for a Micron mindset shift? Here's how to navigate the new AI landscape.
-Jurica Dujmovic
Comments