Artificial intelligence has triggered a capital spending boom unlike anything the technology industry has experienced in the last twenty years. For more than a decade, the world’s largest cloud computing companies—Microsoft, Amazon, Alphabet (Google), Meta and others—grew steadily by building global data centre networks to support cloud adoption and digital transformation. Investments were large, but predictable. Generative AI didn’t just continue this trend, but rather, it has exacerbated it, fundamentally changing the scale, urgency and economics of their investment decisions.

‍

According to research by McKinsey, global data centre investment is expected to reach $6.7 trillion by 2030. Of that total, $5.2 trillion will be required to build new AI-capable data centres, while only $1.5 trillion will be allocated to traditional IT infrastructure. AI will consume nearly 80% of every incremental dollar spent on data centres globally over the next decade. This represents a structural break from the past: spending is no longer driven by organic user growth; it is driven by the explosive computational demands of training and running increasingly complex AI models.

‍

AI investment is not discretionary. Without sufficient computing and data centre capacity, hyperscalers risk falling behind in a winner-takes-most market. This dynamic has created the conditions for the most intense corporate spending race in recent history.

‍

Yet, this surge in capital expenditure introduces a profound investment challenge.

‍

‍

Capex is growing faster than revenue

‍

The scale of spending required means that hyperscalers must deliver revenue growth at unprecedented levels simply to justify their investment. Based on their current capital expenditure trajectory, hyperscalers must generate approximately 16% year-one revenue growth, equivalent to $242 billion of new revenue, just to sustain their existing return structures and dividend expectations. And this assumes that returns on capital remain unchanged, which is a generous assumption in a period of sharply rising capital intensity.

‍

McKinsey estimates that AI-related data centre spending will need to increase by almost 30% per year through the end of the decade to meet infrastructure requirements. Revenue, however, is not scaling at the same rate as capital commitments. This widening gap between required capex and achievable top-line growth introduces a new form of valuation risk: even if the AI revolution succeeds, the economics for the hyperscalers may be less attractive than markets currently assume.

‍

‍

Where does $242 billion in new revenue come from?

‍

There are only three logical sources of new revenue for hyperscalers: increased software spend, labour productivity (as companies replace or augment human work with AI), or displacement of spend that traditionally flowed to other enterprise software vendors.

‍

However, even under optimistic adoption assumptions, the numbers fall short. If hyperscalers were able to capture ten percent of the global technology budget, they would generate roughly $10 billion of incremental revenue. If AI adoption allowed organisations to reduce people costs by only half of the average attrition rate, the resulting efficiency gain would be equivalent to a 2.5% reduction in global labour costs, which translates to approximately $25 billion. Together, these two sources represent around $35 billion—a fraction of the $242 billion required in the first year alone to sustain return structures.

‍

This mismatch between required revenue and realistic achievable revenue demonstrates the magnitude of the challenge. The spending race is being driven not by immediate commercial return but by strategic fear, of losing platform control, market share, or technological leadership.

‍

That is only for year one, what about when it increases for the next four?

“I’m willing to go bankrupt rather than lose this race.”

‍

No modern investment cycle has been characterised by language like this. In the last year, two of the most influential founders in technology have openly acknowledged that overspending is a risk they are willing to take.

‍

Mark Zuckerberg recently stated that if his company ends up “misspending a couple hundred billion dollars, that would be unfortunate, but the bigger risk would be not spending aggressively enough”. Larry Page, co-founder of Google, went even further, saying he was “willing to go bankrupt rather than lose this race.”

‍

Their words reveal the strategic mindset of the companies driving AI adoption: preserving leadership is seen as more important than preserving margins.

‍

‍

Investment implications: capex pressure means returns must be monitored

‍

In investment terms, when capital intensity rises as sharply as it is today, it challenges a company’s ability to maintain its return profile. Higher capex often forces management teams to make difficult trade-offs: dividend growth may slow (bad idea), capital return programs (such as buybacks) may be reduced or suspended, and valuation multiples may compress as investors reassess the risk-adjusted return outlook.

‍

We own four companies that benefit from AI infrastructure investment regardless of which hyperscaler ultimately “wins” the race: TSMC, ASML, Amphenol, and Applied Materials. These companies represent the physical backbone of the AI capex cycle. AI cannot happen without leading-edge semiconductors, without the equipment required to manufacture those chips at ever smaller geometries, nor without the connectors and interconnects that allow power and data to flow within data centres.

‍

TSMC produces the world’s most advanced semiconductor manufacturing technology. ASML is the sole provider of EUV lithography systems used in the most sophisticated chip production. Applied Materials designs and manufactures equipment essential to semiconductor fabrication. Amphenol supplies the connectors and interconnects that enable data transmission and power distribution in the high-performance environments of modern data centres. These companies profit from the capex wave without having to fund it.

We also own two hyperscalers that are central to the AI arms race. Their scale, technological capability, and access to capital make them critical to the AI ecosystem. However, we will monitor both positions carefully.

The required revenue growth targets are clear, and we will track whether those indicators are being delivered. If the economics no longer justify the capital commitments, we will exit.

‍

In short, we own the beneficiaries of the spending, not just the spenders.

The real investment insight

‍

AI will reshape industries, business models, and productivity. Over the long term, the economic impact will be significant. But in the near term, investors must distinguish between companies that will benefit from AI adoption and those that must fund the infrastructure required to enable it.

‍

The companies spending the most are not necessarily the ones that will earn the highest returns.

‍

‍

Yes, capital discipline still matters

‍

As the AI cycle accelerates, the most attractive opportunities may come not from the companies building the largest data centres, but from those enabling the buildout—those that generate revenue every time a hyperscaler decides to spend another billion dollars on capacity.

‍

Ultimately, return on capital will matter more than the volume of capital deployed.