The Durable Investment Opportunity Beneath the AI Infrastructure Boom

Why AI’s most durable cash flows may belong to the businesses behind the machines

Much has been written about whether the artificial-intelligence investment cycle resembles the dot-com boom—and eventual bust—of the late 1990s. The parallels are understandable. Capital is flowing rapidly, valuations have expanded, transformative claims are being made and investors are again attempting to determine which companies will emerge as the enduring winners of a technological revolution.

But historical comparisons can obscure as much as they reveal and this cycle is notably different and much, much more fundamental.

The central premise of the dot-com era—that the internet would fundamentally reorganize commerce, communication and daily life—proved correct. What failed were many of the business models, valuations and financing structures built around that premise. The technology endured and ultimately transformed the global economy, even as much of the capital deployed during the initial investment frenzy was lost.

Artificial intelligence may follow a similarly uneven path. Some expectations will prove excessive. Some investments will fail to earn their cost of capital. There will be overbuilding, consolidation and a narrowing of today’s crowded field. But none of that necessarily invalidates the scale or durability of the underlying transformation.

That tension became even more pronounced this last week, as Anthropic CEO Dario Amodei and other leaders of frontier AI laboratories publicly called for slowing the pace of model development and introducing stronger safeguards. OpenAI CEO Sam Altman also endorsed greater caution, while Microsoft introduced its own framework emphasizing human oversight. These statements have intensified the debate over whether voluntary restraint will be sufficient—or whether regulation will be necessary to manage the technology’s development. The Washington Post

The warnings arrive at an extraordinary moment in the capital markets. Anthropic is preparing for a potential public offering, with some investors discussing a valuation approaching $2 trillion—although the company has not confirmed either that valuation or a final timetable. OpenAI, meanwhile, will not pursue an IPO in 2026. Altman described a listing at this moment as ill-advised and cited the unresolved safety risks surrounding rapidly advancing AI. Fortune

This apparent contradiction deserves attention. The companies racing to develop the world’s most powerful models are simultaneously attracting unprecedented valuations and warning that the technology may be advancing too quickly. That does not necessarily mean the investment cycle is about to reverse. It does mean investors should distinguish carefully between enthusiasm for the technology, the economics of individual AI companies and the enduring value of the infrastructure supporting the broader ecosystem. One of my colleagues recently published a thoughtful essay on the sheer amount of “vapor-watts” in the market. Before you think about investing in the scale of opportunities— he suggests unearthing the hype.

This is why I am writing now. The most useful question is not simply whether AI is a bubble. It is which parts of the investment cycle are supported by durable demand, who is financing the buildout and where lasting economic value is most likely to accrue.

Matt King a thoughtful Financial Times contributor and macro and credit strategist, recently offered a persuasive framework for considering those questions. He contends in a recent article that markets tend to focus on what is being built while paying too little attention to how it is being financed.

That distinction matters. Unlike many companies at the center of the dot-com boom, today’s largest investors in AI infrastructure—Microsoft, Alphabet, Amazon and Meta—entered the cycle with enormous cash reserves, highly profitable core businesses and some of the strongest balance sheets in the world. Their financial strength allowed them to initiate an unprecedented capital-investment cycle without initially becoming dependent on outside financing.

I agree with King’s macro premise, that the hyperscalers’ financial strength has been fundamental to the pace and scale of the buildout. But the effects of their spending extend well beyond their own balance sheets. Their capital expenditures have become revenue for an increasingly broad ecosystem of companies supplying the computational power, memory, fabrication capacity, electrical equipment, cooling systems and physical infrastructure that the AI economy requires.

Nvidia is the most visible beneficiary, but it is hardly alone—TSMC (Taiwan Semiconductor), Micron Technology and a growing universe of semiconductor, power-management, cooling and infrastructure companies have seen their earnings capacity and balance sheets strengthened by the hyperscalers’ investment cycle. Capital is not simply being spent. It is moving through a complex industrial supply chain—creating profits, funding additional investment and expanding the capacity of the digital economy.

This is one reason we believe the AI infrastructure cycle must be understood as more than a technology trade—it is an industrial transformation.

Every additional unit of computing power requires a physical ecosystem around it— land, electricity, water, cooling, connectivity, security and sophisticated facilities maintenance, capable of operating continuously. The cloud may feel intangible to its users, but the infrastructure behind it is among the most tangible—and increasingly resource-intensive—parts of the modern economy.

That physical reality is central to our investment thesis.

Every bubble is explained twice. In the build-up, the emphasis is on what investors have been buying. Only afterwards do people come to recognise that the more important point is how it was all being financed—Matt King: FT

In the public markets, we have invested in companies positioned to benefit from the expansion of computing capacity and the infrastructure required to support it—especially in the bottlenecks. We remain attentive to valuation and to the risk that enthusiasm may periodically run ahead of near-term earnings. But we also believe it would be a mistake to dismiss the entire cycle as speculative simply because valuations have expanded or financing conditions are evolving.

Unlike the dot-com boom, this cycles demand is being supported by some of the world’s most profitable companies. The infrastructure being built has useful economic lives measured in decades. And the benefits are spreading across companies occupying essential positions in the technology and industrial supply chains.

Our private-market strategy approaches the same opportunity from a different direction.

Rather than attempting to predict which artificial-intelligence application will ultimately dominate, we have focused on the “boring” yet essential services required to keep digital infrastructure operational. My partners and I have assembled a data-center services and maintenance company that now possesses global scope and scale. Its work sits behind the technology itself, supporting the critical environments on which the digital economy depends.

These are the “picks and shovels” of the AI infrastructure boom—but even that familiar phrase understates their importance.

A data center is not merely built and then left to operate. It must be continually decontaminated, maintained, monitored, repaired and adapted. Contamination must be controlled. Mechanical and electrical systems must remain reliable. New equipment must be installed without interrupting existing operations and air flow optimized. Aging facilities must be modernized as power densities rise and cooling requirements become more demanding.

The economic consequences of failure can be enormous. As computing becomes more valuable and concentrated, the services protecting that infrastructure become more critical.

I

nvesting in the Bottlenecks

Nvidia CEO, Jensen Huang has described the challenge as anticipating the “critical pinch points” in the AI supply chain. As he recently explained, “Each one of these bottlenecks gets a great deal of attention. Now we’re prefetching the bottlenecks years in advance.” When asked to identify the hardest constraint, his answer was not another chip architecture or software model. It was far more tangible: “Plumbers. Plumbers and electricians.”

That observation captures an important part of our investment thesis. The next phase of the AI economy will not be determined by computing power alone. It will also depend on whether the physical infrastructure surrounding that computing can be built, supplied and maintained. Reliable power, water, cooling, electrical capacity, environmental monitoring and skilled technical services are no longer peripheral to the technology—they are among its most important constraints.

For investors, the implication is clear—follow the bottlenecks. Where demand exceeds available capacity, essential-service providers can acquire greater strategic importance, stronger pricing power and deeper customer relationships. Impala Ventures has been investing against precisely this premise—identifying the critical services without which the digital economy cannot scale or remain operational.

This creates a different kind of investment opportunity from owning the most celebrated semiconductor or software company. Mission-critical data-center services can produce recurring demand, durable customer relationships and attractive cash flow. They are often difficult to replace because providers accumulate institutional knowledge, operate inside sensitive environments and become integrated into their customers’ risk-management processes.

In other words, the value is not limited to constructing the infrastructure or the rakes of GPU’s they house. Considerable long-term value will be created by maintaining it. We are focused on monetizing the installed base of datacenters that we support from construction, through operations, to reconfiguration as GPU’s get more advanced and need refreshed.

The Known & Unknown Financial Risks

There are extraordinary financial risks. Hyperscaler spending could slow. Certain markets may develop excess capacity. Power constraints, permitting delays and community opposition could limit development. Some participants will undoubtedly overbuild, overborrow or pay valuations that assume uninterrupted growth.

The financing structure is also changing. As the scale of capital required exceeds even the hyperscalers’ prodigious internal cash generation, more of the burden will move toward debt markets, private credit, infrastructure funds and outside equity partners. That shift deserves close attention. The cost of capital ultimately matters, even when the underlying technology is transformative.

But a changing capital structure does not invalidate the underlying demand. It instead places greater importance on selecting the right point in the value chain.

For the past nine years, Impala Ventures has pursued an intentional investment thesis centered on the infrastructure enabling the digital economy. We have expressed that conviction across both our public-equities portfolio and our private-market investments. Our perspective has been shaped by a simple observation: regardless of which AI models, applications or platforms ultimately prevail—and regardless of how regulation influences their development—they will depend on an expanding foundation of physical infrastructure and the essential services required to keep it operating.

Regulation or a voluntary slowdown could alter the timing and composition of AI investment. It could moderate the pace of frontier-model development, increase compliance costs and concentrate more power among companies with the financial resources to satisfy stringent safety requirements. Public-market expectations may adjust accordingly. But the physical infrastructure already constructed or under development will not disappear. Data centers will always require reliable power, water, cooling, environmental control and continuous maintenance. Indeed, a more regulated and safety-conscious AI economy may place an even greater premium on resilience, monitoring, compliance and operational reliability.

That is the distinction at the heart of our investment thesis: the eventual winners among AI models remain uncertain, but the essential requirements of the infrastructure supporting them are becoming clearer. The AI boom will inevitably move through cycles of optimism, constraint and recalibration. Public-market valuations will fluctuate. Financing models will evolve. Some ambitious projects will fail to earn their cost of capital.

But the infrastructure already being created will still need to operate tomorrow, next year and for decades to come.

That is where we see the more enduring opportunity. Not only in the companies designing the chips or financing the next generation of hyperscale campuses, but in the essential, often overlooked businesses that keep those facilities functioning every hour of every day.

The excitement may surround artificial intelligence. The durable cash flows may belong to the companies quietly keeping its physical foundations running.