Could AI Change Everything—and Investors Still Lose?
Most investors have followed the extraordinary share-price gains of Nvidia and other beneficiaries of the AI boom, so enthusiasm for the industry is understandable. But will shareholder returns from here match the excitement, and how much of the anticipated growth is already reflected in today’s valuations?
The technology is transformational. Demand for compute is real and growing. And the productivity improvements are tangible. Nevertheless, investing in a theme hasn’t always been a sound strategy. Gus and I are wired cautiously, which tends to make us skeptics and contrarians (sometimes to my wife’s dismay).
The capital committed to AI infrastructure has been eye-popping: Goldman Sachs estimates that global AI-related investment will exceed $1 trillion in 2026 alone.1 Suppliers to this buildout have benefited accordingly. But how long can this level of spending continue?
History suggests skepticism isn’t unfounded: past buildouts of transformational technologies have often ended in overinvestment. After the internet boom, an estimated 95% of installed fiber sat unused in 2001,2 and excess capacity persisted for years. In the 1880s, U.S. railroad mileage increased about 75%,3 and by 1894 roughly a quarter of it was in receivership.4 In 1909, there were 272 U.S. automobile manufacturers, but by 1941 only nine remained.5
Railroads transformed commerce, automobiles transportation, and the internet communication, while fracking revolutionized American energy production. Yet in each case, considerable economic value accrued to consumers or users of the technology, but that value did not always translate into attractive returns for the investors footing the bill.6 The excess fiber laid during the internet buildout, for example, helped drive wholesale bandwidth prices down more than 95% between 1998 and 2007,7 providing cheap “plumbing” that allowed companies like YouTube and Netflix to flourish over the following decade.
Consider Cisco, a leading supplier of internet-related hardware in the late 1990s. Following three years of roughly 44% annualized revenue growth, Cisco briefly became the world’s most valuable publicly traded company in March 2000, with a market capitalization of $555 billion, trading at more than 200 times earnings. Investors who purchased at its $80.06 high watched the stock decline nearly 90% by October 2002. Its share price didn’t regain that high until December 2025—nearly 26 years later.8 The internet proved as transformative as investors anticipated, and Cisco remained a strong business. The problem was the price: investors had paid in advance for decades of success.
Today’s AI buildout has important differences, and we could still be in the early-to-mid innings. Rising usage and new applications could create enough demand to justify much of today’s investment. And the largest infrastructure investors—Microsoft, Amazon, Alphabet, and Meta (the “hyperscalers”)—are among the most profitable businesses in the world, with robust balance sheets.
Industry spending is also remarkably concentrated: the four hyperscalers alone plan to spend more than $700 billion on capital projects in 2026 (based on April 2026 guidance),9 much of it on AI, and have largely funded the buildout with cash on hand and from operations. But that cushion is thinning.
As the chart below shows, the four companies’ combined capital expenditures have risen from about half of operating income (EBIT) in 2015 to 106% for the twelve months ended June 2026. Meanwhile, their net cash fell from roughly 6.3 times annual capital spending in 2015 to 0.17 times—roughly two months’ worth. They can still borrow or issue shares, and increasingly are,10 but the numbers suggest today’s pace of spending may be difficult to sustain.

And if these financially sound companies become more disciplined with spending, what happens to the businesses throughout the AI infrastructure supply chain that depend on continued investment? Some infrastructure operators have borrowed heavily to finance expansion,11 leaving them vulnerable if spending slows or access to capital weakens. Even a modest pullback by the hyperscalers could have a disproportionate impact on the industry.
Skepticism and optimism aren’t mutually exclusive. One can be wildly optimistic about the technology yet remain skeptical about the investment prospects.
Specifically, we question:
- Whether today’s supply-demand imbalance will persist years from now as new capacity comes online.
- How demand will change once compute is priced to allow infrastructure providers to earn an adequate return.
- Whether current rates of infrastructure spending can be sustained.
- Whether today’s valuations leave room for attractive returns for investors.
Investing is an art that requires balancing aggressiveness with conservatism, humility with confidence, and patience with gumption. When it comes to investing in AI, we think caution is warranted.
For us, caution means valuation discipline: we’re wary of owning businesses whose prices already assume years—or decades—of uninterrupted growth. As Cisco showed, a great company and a great investment aren’t always the same thing.
SOURCES
- Goldman Sachs, “Global AI Investment Is Forecast to Exceed $1 Trillion in 2026,” August 7, 2026. goldmansachs.com
- New York Times News Service, “Economy’s slump idles 95% of fiber-optic lines,” Deseret News, June 18, 2001. deseret.com
- U.S. railroad mileage, 1880 (93,267) and 1890 (163,597), per John F. Stover; American-Rails.com, “Railroads in the 1880s.” american-rails.com
- “Panic of 1893,” Wikipedia, citing W. N. Leonard, “The Decline of Railroad Consolidation,” Journal of Economic History, 1949. wikipedia.org
- Steven Klepper, “The Evolution of the U.S. Automobile Industry and Detroit as Its Capital,” Carnegie Mellon University, 2001. zew.de
- Argus Media, “Shale industry write-downs may top $300bn: Deloitte,” June 22, 2020 (shale industry: $300 billion cumulative negative free cash flow and more than 190 bankruptcies since 2010). argusmedia.com
- DrPeering.net (William B. Norton), “Internet Transit Pricing: Historical and Projected,” 2010 (U.S. internet transit prices per Mbps, 1998–2007). drpeering.net
- Cisco: revenue, Cisco Systems Form 10-K for fiscal 1999; market value, NPR, “Cisco Soaring,” March 28, 2000; valuation, AJ Bell, “Is 2025 more bubbly than 1999?”; price history, Sherwood News, “After a quarter century, Cisco surpasses dot-com bubble closing high,” December 11, 2025. sec.gov · npr.org · ajbell.co.uk · sherwood.news
- Sherwood News, “Alphabet, Amazon, Microsoft, and Meta plan more than $700 billion on capex this year,” April 30, 2026. sherwood.news
- International Monetary Fund, Global Financial Stability Report, October 2026, Background Analysis 1.1, “Rising Leverage in the AI Ecosystem.” imf.org
- The Next Web, “IMF warns $1.4tn in circular AI deals could deepen a market correction,” October 2026 (on the IMF’s October 2026 Global Financial Stability Report). thenextweb.com
IMPORTANT DISCLOSURES
Latticework Investment Management (“Latticework”) is an investment adviser registered with the State of Kansas. Registration does not imply a certain level of skill or training. More information about Latticework, including our Form ADV Part 2A brochure, is available upon request or at adviserinfo.sec.gov.
This commentary is provided for informational and educational purposes only and reflects Latticework’s views as of the date of publication, which are subject to change without notice. It is not intended as, and should not be construed as, investment, tax, legal, or accounting advice, or as an offer or solicitation to buy or sell any security. Please consult your own advisors before making any investment decision.
References to specific companies, including Nvidia, Cisco, Microsoft, Amazon, Alphabet, Meta, and Netflix,[C1] are for illustrative purposes only and do not constitute a recommendation to buy, sell, or hold any security. Latticework and its clients may hold positions in securities mentioned.
Certain information is obtained from third-party sources believed to be reliable, but Latticework has not independently verified it and makes no representation as to its accuracy or completeness. The chart reflects Latticework’s calculations based on company filings. Forecasts and estimates, including those attributed to third parties, are inherently uncertain and may not be realized. Links to third-party websites are provided for convenience; Latticework is not responsible for their content.
Past performance is not indicative of future results. All investing involves risk, including the possible loss of principal.
Recent Posts





