Understanding the AI CapEx Boom | Five Lessons from Kai Wu
How to Think Clearly About the AI CapEx Boom, Capital Cycles, and the Future of Innovation
Artificial intelligence may become the most transformative technology of our lifetimes. It is already changing how companies operate, how capital is allocated, and how investors think about growth. Yet beneath the excitement is a more complicated story that is rooted in capital cycles, competition, and the history of technological booms.
Our good friend Kai Wu of Sparkline Capital wrote an excellent paper, “Surviving the AI Capex Boom” that used a data driven approach to relate the current AI boom to previous transformational periods. He joined us to discuss the big ideas of the paper.
Whenever we sit down with Kai, we know we will come away with a deeper, more balanced understanding of innovation. His research connects intangible assets, market history, and the behavior of firms under competitive pressure. Our recent conversation with Kai on the AI CapEx boom was one of his most thought-provoking yet.
Here are the five lessons that stand out.
Lesson 1: Technology Changes Fast, but Capital Cycles Endure
Investors love to tell themselves that each new technology breaks the rules. Kai reminded us of a simple truth: the physical and financial cycles behind innovation are as old as railroads and telegraphs.
In every major boom, companies rush to win the land grab. This happened with railroads in the nineteenth century, telecom infrastructure in the late 1990s, and it is happening again with AI.
“We are seeing record levels of spending,” Kai told us. “In annualized terms, the AI buildout is now larger relative to GDP than railroads at their peak or the fiber boom in 2000.”
Even though the technology changes, the pattern repeats. Capital floods in. Competitors expand aggressively. Investors reward aggressive spending. Supply overshoots demand. Prices fall. Firms with heavy capital intensity struggle or fail.
Kai summarized the mindset that gets investors into trouble: “The four most dangerous words in the English language are this time is different.”
The lesson is timeless. The technology may be novel, but the capital cycle is not.
Lesson 2: The Magnificent Seven Are Asset Light No More
For more than a decade, the Magnificent Seven dominated markets because they were asset-light, intangible-intensive, and enormously profitable. They printed cash while spending relatively little on physical infrastructure.
That world is changing fast.
In their pursuit of AI leadership, several of these companies have transformed their businesses. Kai laid out the numbers in the interview: Alphabet’s CapEx is now 21 percent of revenue, Microsoft’s is 28 percent, and Meta’s is 35 percent.
“Meta is spending more on CapEx than the average utility today,” Kai said. “More than AT&T did at the height of the dot-com bubble.”
Companies that once thrived on intangible advantages now look more like industrial firms, with high fixed costs, rapid depreciation, and constant reinvestment requirements. GPUs depreciate in two to five years, not thirty years like railroad steel. Firms cannot build these data centers once and harvest profits indefinitely. They must continuously refresh them.
As Kai put it, “These companies need to be making trillions of dollars in revenue five years out in order to justify these investments.”
The implication is profound: business models that succeeded because they were capital-efficient are becoming capital-intensive. History shows that this lowers returns.
Lesson 3: High Investment Does Not Guarantee High Returns
One of the most important insights in the conversation was Kai’s discussion of the long-term underperformance of firms with rapid asset growth or heavy capital spending.
Fama and French documented that companies that aggressively grow their asset base underperform significantly over the long run. Kai emphasized that this is not just a sector effect. It appears within sectors as well.
“It is evidence of the capital cycle,” he explained. “Companies that are aggressively trying to grow their businesses tend to underperform those that are more conservative.”
The same pattern holds for firms with large increases in CapEx. In the dot-com era, telecom companies dramatically outperformed during the boom, then collapsed more than 90 percent when the cycle turned. Heavy investment creates excess capacity. Excess capacity crushes margins. It is the oldest story in industrial organization.
This does not mean AI is unimportant. It means investors must distinguish between the users of AI and the builders of AI infrastructure. History suggests the users often earn the economic returns.
Lesson 4: The Ultimate Beneficiaries of Technology Booms Are Not the Builders
Kai walked us through the railroad boom of the 1800s and the fiber boom of the late 1990s. In both cycles, the builders experienced significant problems while society enjoyed enormous gains.
After the fiber bust, 85 percent of bandwidth went unused. Prices fell 90 percent. But that collapse made Netflix, Google, YouTube, and Facebook possible. Infrastructure investors lost large sums of money, but the next generation of companies flourished because of the infrastructure left in their wake.
Kai drew the parallel clearly: “You see this dynamic over and over. The companies that build the infrastructure struggle. The companies that use that infrastructure thrive.”
This is an evergreen lesson. The value created by a platform rarely accrues to the platform builders. It accrues to those who apply the new technology to real problems.
That distinction is likely to matter again.
Lesson 5: You Can Be Long Innovation Without Being Long Capital Intensity
Kai is not bearish on AI. Far from it. He believes it will be a transformative general-purpose technology. The lesson is not to avoid AI, but to be thoughtful about where in the value chain the returns will accrue.
“You still want to be long innovation. You still want to be long AI,” he said. “But you want to do so in a way that is less exposed on the capital intensity side and less exposed on the valuation side.”
This is the core of Sparkline’s intangible value framework. The goal is to invest in innovative companies, but to do so with a focus on their actual value relative to their intangible assets and not just promises of future gains.
Roughly 20 percent of an AI-oriented portfolio today built using Kai’s intangible value framework would be infrastructure firms that remain reasonably valued. The other 80 percent would be early adopters across sectors such as financials, communications, industrials, healthcare, consumer staples, and energy.
These companies are beneficiaries, not builders. They ride the wave without paying for the wave machine.
Progress Is Real, but So Are the Cycles
AI may reshape entire industries, just as the internet and railroads did. But the forces that determine investment returns are older and more predictable than any model architecture.
Capital intensity matters. Competition matters. Valuation matters. Incentives matter.
Kai’s work reminds us that technology cycles reward both optimism and discipline. Optimism helps us imagine what is possible. Discipline keeps us grounded in the economic realities behind that possibility.
If history is any guide, the winners of the AI era will be the firms that harness this technology, not necessarily the firms that build the infrastructure beneath it.
Kai Wu's framework for thinking about the AI CapEx boom is really valuabe. The historical parallels to railroads and fiber are spot on. What striked me most is how the Magnificent Seven have shifted from asset-light to capital-intensive models. The point about users of AI outperforming builders resonates with what we saw in previous technology cycles. Looking forward to seeing how this plays out over the next few years.