Investors in artificial intelligence may see the same sell-off across a wide group of stocks and assume the risk is evenly shared. James Demmert, who has worked in the investment industry for more than 35 years, argues that the data-center debate creates a more useful divide: Companies that already operate major data-center networks may be better positioned than businesses whose growth depends on building many more of them.
Demmert, chief investment officer at Main Street Research, favors putting new money into the broader AI trade, including semiconductors and major cloud operators. He places Alphabet and Amazon near the top of his Magnificent Seven ranking because their stocks have lagged despite healthy cloud results and because each already has substantial data-center capacity.
That strategy suits investors who are seeking long-term AI exposure, can tolerate volatility, and evaluate balance sheets. It is less suited for investors looking for short-term trades tied to data-center construction, where capital requirements can matter as much as demand for computing power.
Here is how Demmert, who spoke in an interview on Street Talk with TheStreet’s Caroline Woods, separates established AI operators from more speculative data-center growth stories.
Note: The market musings expressed herein are those of the interviewed guest and should not be treated as professional investment advice. Readers should consult a certified financial planner (CFP) for individualized advice based on their available capital, goals, risk tolerance, and time horizon.
Why AI data-center fears can create a split market
The central worry is straightforward: AI companies are spending heavily on computing infrastructure, while investors question whether the eventual revenue will justify the investment. Demmert said strong results from cloud businesses at Microsoft, Amazon, and Alphabet haven’t fully eased investor nerves. In his view, investors remain particularly focused on pressure surrounding Google Search rather than the cloud results that he described as strong.
That skepticism has weighed on valuations. Demmert said the semiconductor industry was trading at 15 times earnings, down from a multiple of 30 a year earlier. A price-to-earnings ratio compares a stock’s price with its annual earnings; a lower multiple can make a stock look cheaper, but it doesn’t guarantee that its price will rise.
The data center build-out problem, which I frankly think is completely overdone, I think that we can’t slow that down if we want to keep up ahead of China.
James Demmert, when asked why some AI-linked companies could be hurt by a slowdown in data-center build-outs
Demmert’s case rests on the idea that AI demand will continue to require more computing capacity and that the U.S. will have a strategic reason to keep building it. That is an investment view, not a certainty. His stated risk is that demand for electricity and data centers could slow more than expected, which would weigh especially heavily on companies tied to the power required for AI computing.
Why existing data centers may offer a safer AI route
Demmert distinguishes between companies that already own large data-center footprints and companies trying to acquire or build enough capacity to scale. He described Alphabet and Amazon as safer AI exposure because existing facilities can serve customers that need computing resources now. He still sees potential returns in faster-growing operators, but said investors must be prepared to wait through a few quarters of uncertainty.
However, there’s a trade-off. Established operators may not offer the same upside as a smaller company that successfully expands quickly. Yet Demmert’s framework puts financial durability first. A company that needs substantial new capacity must also finance that expansion, and a weak balance sheet can turn a favorable industry trend into a difficult stock-specific situation.
We love strong balance sheet companies, dominant market share, but the balance sheet there just is very concerning.
James Demmert, when asked whether CoreWeave is an opportunity or a trap
For instance, Demmert called CoreWeave a trap, citing its need for more data centers and balance-sheet concerns. His conclusion is a company-specific judgment, rather than a verdict on all newer AI infrastructure businesses. But it explains why a data-center theme should not be treated as a single bet: Owning capacity, needing capacity, and financing capacity are materially different positions. Coreweave’s stock dropped about 40% in the year to date at the time of the interview.
How to build AI exposure without making one concentrated bet
For investors who want AI exposure, Demmert favors a quality-focused semiconductor group that includes Nvidia, Advanced Micro Devices, Taiwan Semiconductor Manufacturing Company, ASML Holding, and Micron Technology. While their specialties range from chip design and memory to fabrication equipment, all five provide the foundational hardware essential for AI computing.
Demmert is more cautious on the electrification trade, naming GE Vernova and Siemens Energy as companies whose performance could be vulnerable if power demand worsens. This caveat underscores a broader reality: the AI investment theme spans well beyond large technology stocks. Over-indexing to one vertical risks misdiagnosing where supply-chain constraints or execution disappointments will actually emerge.
Demmert adds a word of caution for investors loaded up on tech: Don’t make AI your only bet. He sees viable opportunities across healthcare, energy, and financials, though he recommends waiting for pullbacks in energy and healthcare rather than buying at peak momentum. Diversification doesn’t wipe away risk, but it cushions the blow if market expectations for AI fail to pan out.
The takeaway for AI investors weighing data-center risk
For your portfolio, first decide whether you want long-term exposure to established AI leaders, a more speculative position in emerging infrastructure providers, or both. Keep those roles separate. Demmert’s approach favors dominant companies with existing capacity and durable finances over businesses that must raise capital to make their expansion plans work.
Then revisit the evidence as cloud results, capital spending, power demand, and balance sheets change. If your goal and time horizon support AI ownership, a diversified group of established operators and semiconductor suppliers may fit better than a single all-or-nothing data-center bet.
Sources
James Demmert on Street Talk · watch the full segment
Cited segments:
Demmert said the semiconductor industry was trading at 15 times earnings, compared with 30 times earnings a year earlier. (5:37–6:26)
Demmert said Alphabet Inc. and Amazon.com, Inc. were safer because they already have data centers. (7:52–9:01)
Demmert called CoreWeave a trap because of its data-center needs and concerning balance sheet. (9:16–9:30)
Demmert named Nvidia, Advanced Micro Devices, Taiwan Semiconductor Manufacturing Company, ASML Holding, and Micron Technology as high-quality semiconductor exposure. (6:28–6:59)
Some or all of this content may have been generated by AI. While we strive for accuracy, AI can occasionally produce incorrect information. If you buy something via one of our links, we may earn a commission.