The Fertile Bubble: AI Infrastructure, the Capital Cycle and the Question of Who Reaps the Rewards in the End

Read the market analysis and fund positioning

The physical side of an abstract revolution

When people talk about artificial intelligence, they think of models and software – something almost weightless. Those who finance it think of concrete, silicon and electricity. Behind the intelligence revolution lies a construction boom of historic proportions: data centres with the energy appetite of entire cities, built with sums of money never before raised in the industry. And everything about this story adds up. The demand is real, the models deliver on their promises, and those who supply the infrastructure stand to profit from it.

It is precisely this coherence that should make investors sit up and take notice, for a true story does not necessarily translate into a sound investment case. The more solid the operational reality becomes, the more readily the market extrapolates today’s growth far into the future. Whether this extrapolation holds up or whether a familiar pattern repeats itself is not a matter of conviction, but of scrutiny; and past cycles provide the benchmark here.

The ever-recurring pattern: build-up, crash, exploitation

Infrastructure-driven technological revolutions follow a recurring pattern. The trigger is followed by a phase of land-grabbing. Companies secure market share before the boundaries are drawn. In this phase, capital flows in to secure the best position and out of fear of being too late – returns are only considered later. The equipment suppliers benefit from this. The ‘shovel sellers’ record explosive sales, and their valuations soar the most.

Then capacity outstrips demand. Returns fall, yet the enormous competitive pressure keeps spending high, and this is precisely how overcapacity arises. Eventually, demand disappoints or funding dries up; prices and returns on capital plummet, and value corrections follow. The equipment suppliers are then hit hardest, because demand does not gradually wane but collapses abruptly.

What happens next is crucial. The infrastructure is real and permanent, so it will be used. This can be seen, for example, in the ‘dark fibre’ laid in 1999 – the vast quantities of fibre-optic cable laid but left unused. It was only over the following decade that it was put into operation, and not by the original builders. The productivity gains flow to a new group of companies that build on the now-cheap infrastructure, whilst many of the original investors have disappeared from the market.

This pattern has a name: the economist William Janeway calls it a ‘productive bubble’. In this process, capital is destroyed, but the economy is permanently transformed. The shale gas boom is one of the most recent examples. Between 2010 and 2020, North American producers generated over 189 billion US dollars in negative free cash flow. Investors lost out, whilst cheaper energy eased the burden on household incomes. A systemic crisis was averted because the losses were spread across many small producers. This is the central theme of this text: what is good for the economy can ruin individual investors.

Cisco is a cautionary tale that illustrates this lesson in a single share price chart. The company was the dominant, highly profitable ‘shovel seller’ of the internet revolution. Anyone who bought the shares in March 2000 at their peak took twenty-six years to recoup their nominal cost price ; it was not until early June 2026 that Cisco finally surpassed its dot-com high; in real terms, investors are still in the red today. The irony of the story is that the catalyst that closed the decades-long price gap is, of all things, the next capex supercycle – this time for AI.

The bottom line is that being right about the technology is not the same as making money from it.

Four markers of a bubble

Rather than simply claiming that we are living in a bubble, we propose a diagnostic tool: observable markers that distinguish healthy growth from severe overvaluation. Three markers assess supply, one assesses demand; only the last one can trigger the downturn.

Three markers on the supply side

Marker 1 – Financing structure

Who is paying, and can the money tap be turned off?

In the case of the telecoms bubble, the core of the problem was debt: over two thousand network operators were reliant on borrowed capital and supplier financing, and fell by the wayside when the markets consolidated between January and April 2001.

Today, by contrast, the core consists of a handful of financially strong, large hyperscalers who finance their investments predominantly from current cash flow (Amazon cites around 200 billion US dollars for 2026, whilst Alphabet and Meta are in the high three-figure range). The tap cannot be turned off as easily for them as it was in 1999.

Preliminary assessment: the core is green.

The deterioration in issuers’ financing structures is a peripheral issue here. In 2025, the five largest hyperscalers raised more than three times their usual level of debt; free cash flow is shrinking, whilst investment is doubling. Above all, debt is being shifted off-balance-sheet: more than US$120 billion in infrastructure debt has been offloaded from balance sheets via special purpose vehicles (SPVs). One telling example is a case where such a company received an investment-grade rating, even though the operator behind it is in the high-yield segment. And the first cracks are appearing: a major private lender has withdrawn from a project worth billions, and a high-profile site expansion has been cancelled. These are still isolated cases, but this is precisely where the chain broke last time.

Result: a green core, creeping debt at the edges; the early stages of fragility that engulfed the periphery during the telecoms bubble.

Marker 2 – Revenue quality

Does the cycle close with external, cash-generating cash flow?

During the telecoms bubble, equipment manufacturers lent their customers money to buy their own technology and recorded this as revenue. Lucent, for example, committed around 8 billion US dollars in customer financing, whilst at Nortel, financing offers even reached 130 per cent of the purchase price. Money flowed from the financing side into the equipment manufacturers’ revenue figures, and the growth looked spectacular.

The modern echo of this is the circular interdependence of chip manufacturers, AI laboratories and cloud providers, in which one party’s investment becomes another’s revenue.

Yet not every such interdependence is dangerous. The litmus test that distinguishes legitimate concern from scaremongering is this: does the circle ever close with external, paying cash flow? If the loop ends with a profitable customer, it acts as an accelerator; if it ends in euphoria and a balance sheet that never sustains itself, it is ruinous. Which scenario occurs is determined by the demand side.

Assessment: yellow; the structure exists, but its level of risk depends on Marker 4.

Marker 3 – Cost side and key figure

Does the valuation take account of the true costs and the true useful life of the asset?

EBITDA excludes depreciation as if it were non-cash. This may apply to a building that stands for forty years, but not to graphics processors, which become obsolete in two to three years and are depreciated over five to six years. For an asset on the depreciation conveyor belt, the annual loss of value is a real, recurring cost item: the chips must be replaced to stay in the race. Two factors exacerbate this.

Firstly, purchases are made at the peak of the cycle. For hyperscalers, not investing seems more dangerous than over-investing – so they buy, paying almost no heed to the price. And the price is high: producer prices for semiconductors are at an all-time high. Electricity is also becoming more expensive, as grid prices have skyrocketed in US regions with a dense concentration of data centres.

Inflation becomes apparent at the end of the value chain: in June 2026, Apple raised the prices of Macs and iPads by up to 300 US dollars, explicitly citing the cost of memory chips required for AI expansion; Microsoft followed suit with its consoles. A cost base built up at the peak permanently reduces the return on investment and increases the scale of any subsequent write-downs.

Secondly, today’s profits primarily reflect the scale of investment – not what these investments actually cost. This distortion is symmetrical: whatever drives prices up will just as quickly drive them down. A leading storage manufacturer recently saw its turnover jump by almost 200 per cent, yet 85 to 90 per cent of this was price-driven.

For a heavily debt-financed cloud provider, the return on investment is already close to zero simply because of the generous depreciation period. As soon as a shorter, more realistic lifespan is applied, it turns negative. And the market is already shifting: rental prices for the penultimate generation of chips fell by around 28 per cent in one year.

Assessment: poor performance in terms of financial discipline, most pronounced at the periphery.

The demand-side marker

Marker 4 – Demand

The actual trigger.

One thing links supply and demand: the current shortage. It drives up costs but limits short-term overcapacity, as you cannot build capacity that you cannot source. Nor does it resolve the problem; it merely shifts it.

This is first evident on the supply side: the three dominant storage manufacturers are currently expanding their capacity simultaneously and on a massive scale; if the plants reach volume production from 2027 onwards, it will be the largest simultaneous expansion in the industry’s history.

Every decision in this context is rational, because capacity is sold out. Yet, collectively, this behaviour in the storage industry has invariably led to overcapacity and a price collapse, most recently in 2022/23, and before that in 2018/19. It is the same dynamic as when ‘dark fibre’ suddenly came online: scarcity today, overcapacity tomorrow.

However, only demand can trigger the downturn. The telecoms crash began with the debunking of a demand myth: the claim that data traffic would double every hundred days was by about an order of magnitude too high, and in the end only a fraction of the fibre-optic cable was in use.

The same question arises with AI: how much of the demand represents real, paying, sustainable cash flow, and how much is capacity based on mere assumption?

Depreciation acts as the stopwatch here – it starts ticking from the moment a chip is purchased, regardless of whether it is already generating revenue. And it ticks quickly: every new investment in chips immediately incurs further depreciation costs, whilst AI revenues grow only slowly. Ongoing investments alone generate more depreciation costs than current AI revenues can cover. To catch up, revenues would need to roughly double.

An external analysis confirms this sobering reality: taking the consensus estimates up to 2030 and assuming no costs whatsoever, the implied return on investment for almost all major hyperscalers is strongly negative, with the notable exception of Amazon. To achieve a ten per cent return, they would need to generate an additional two to five trillion US dollars in annual revenue, compared with around one and a half trillion today.

And it is precisely this exception that is revealing. Amazon is not in a better position because of superior technology, but because of the model behind it: with AWS, investments feed into a business that has long been leasing its capacity to paying customers outside the AI cycle – precisely the external, paying cash flow from Marker 2. It is therefore about the business model and the capital cycle, not the technology.

The strongest counter-argument must be met head-on: the relevant market size is not companies’ IT budgets, but the global market for cognitive labour, which is many times larger. In that case, AI is not selling the tool, but the labour itself, and the benchmark is no longer the price of the software, but the salary that would otherwise have to be paid.

Yet this version has its limits. Substitution only takes effect when the cost per result falls permanently below the price of human labour. That is not the case. What has been evident so far is supplementation, not replacement: AI boosts the productivity of specialists, but beyond the realm of reliable tasks, it worsens the results. Recognising this requires precisely the expertise that runs counter to the substitution thesis.

The real question is a race: will the cost per outcome fall below the price of human labour, and will this happen before the depreciation schedule forces write-downs?

Assessment: As long as this remains open, Marker 4 remains the yellow signal, and the only one that can trigger a downturn.

Conclusion

Taken together, the markers do not paint a clear picture of sentiment, but rather a diagnosis: healthy at the core, yet cause for concern at the periphery. The technology itself will prevail, and the systemic core is on firmer ground than in 1999 – namely, debt-free and self-financed. Yet returns are unevenly distributed, and the areas of risk can be clearly identified:

• developers relying on debt financing who stake everything on a single, rapidly depreciating asset and finance it against merely projected turnover;
• valuations that ignore actual depreciation;
• capacity based on assumed rather than actual demand.

Anyone wishing to be part of the revolution would be best advised to invest in the part of the capital that can weather a dip in demand: the self-financed core and the genuine bottleneck suppliers, not the debt-financed periphery. And they should not pay the prices of long-life assets for a technology that is, in reality, on the depreciation conveyor belt.

The infrastructure that has been built will ultimately be used; the question is not whether the technology will prevail, but whether the investor can weather the period during which the market realises that they have got ahead of themselves.

The technology is real. The capital cycle is the danger. And this danger is concentrated precisely where it can be identified.