9 October 2026
The artificial intelligence (AI) investment boom continues to gather momentum, with technology companies committing enormous sums to data centres, computer chips and supporting infrastructure. Yet an increasingly important question is emerging: will these investments generate enough revenue to justify their cost?
The concern is not whether AI will transform the economy. It is whether the infrastructure being built today will produce sufficient financial returns for the companies and investors funding it.
Our assessment is that the AI infrastructure boom remains in a late-cycle expansion. Demand for computing capacity is still strong, but the financial risks associated with sustaining this investment are becoming more apparent.
AI Spending Shows Little Sign of Slowing
The world’s largest technology companies continue to invest heavily in AI infrastructure.
Microsoft, Alphabet, Amazon and Meta have collectively committed hundreds of billions of dollars to building data centres, purchasing advanced semiconductors and expanding computing capacity.
Recent company disclosures illustrate the extraordinary scale of this investment:
- Microsoft reported quarterly capital expenditure of approximately US$41 billion, including finance leases.
- Alphabet spent US$80.6 billion in the first half of 2026, more than double the amount spent a year earlier.
- Amazon’s quarterly cash capital expenditure reached US$53.1 billion, up from US$31.4 billion a year earlier.
- Meta raised the lower end of its 2026 capital expenditure guidance to US$130 billion, from US$125 billion previously.
Much of this spending reflects the belief that demand for AI computing services will continue to grow rapidly.
Importantly, there is no convincing evidence that the major technology companies are collectively preparing to scale back investment.
The semiconductor industry tells a similar story. Nvidia reported data-centre revenue of US$89 billion for its fiscal second quarter ending July 2026, more than doubling from a year earlier. Meanwhile, Taiwan Semiconductor Manufacturing Company’s September revenue rose by 54.6% year on year, although its business extends beyond AI chips.
These figures suggest that demand for advanced computing hardware remains strong.
However, strong spending does not necessarily mean that the investments will ultimately be profitable.
Are AI Investments Generating Enough Revenue?
This is becoming the most important question surrounding the AI boom.
There are encouraging signs that companies are generating revenue from AI-related services.
For example, Google Cloud’s second-quarter revenue rose 82% year on year, while Amazon Web Services recorded growth of 37%. Oracle also reported exceptionally strong growth in its cloud infrastructure business.
Although these businesses benefit from AI adoption, their reported revenue includes traditional cloud services. It is therefore difficult to determine exactly how much additional revenue is being generated specifically by AI.
More importantly, companies generally do not disclose sufficient information to calculate the profitability of their AI infrastructure investments separately.
This creates a significant gap in our understanding of the investment boom.
Consider a simplified example. A company increases its annual AI infrastructure investment by US$20 billion and generates an additional US$5 billion in annual revenue. That additional revenue may appear encouraging, but it tells us little about profitability until operating expenses, depreciation, financing costs and the useful life of the equipment are considered.
The relevant question is not simply whether AI revenue is growing.
It is whether additional AI revenue and profits are growing sufficiently to justify the additional capital being invested.
The distinction matters because technology infrastructure is expensive to build, and some equipment can become technologically outdated relatively quickly.
An investment boom can therefore continue even while its underlying financial returns are deteriorating.
The First Signs of Financial Strain
One of the more concerning developments is the growing reliance on external financing to support AI infrastructure expansion.
Historically, the largest technology companies have generated enormous amounts of cash from their existing businesses. This has enabled them to finance substantial investments internally.
However, as infrastructure spending accelerates, an increasing share of operating cash flow is being absorbed by capital expenditure.
Amazon, for example, reported negative free cash flow of US$7.6 billion over the 12 months ending in the second quarter of 2026, compared with positive free cash flow of US$25.9 billion a year earlier.
This does not imply that Amazon is facing financial difficulties. Nevertheless, it illustrates how rapidly infrastructure investment can consume internally generated funds.
The situation is potentially more challenging for companies without similarly profitable core businesses.
Oracle is a particularly important example. The company has substantially increased its investment in AI infrastructure while pursuing large-scale debt and equity financing.
This raises questions about the financial risks being transferred to lenders, infrastructure investors and other providers of capital.
There are also early signs that investors are becoming more selective.
In October, Australian data-centre developer Firmus abandoned a proposed US$5 billion initial public offering amid valuation concerns and insufficient investor demand.
Meanwhile, developments in corporate credit markets suggest that investors are increasingly scrutinising the borrowing requirements associated with AI infrastructure.
These are not signs of a widespread funding crisis. Capital remains available, particularly for established borrowers with strong balance sheets.
However, they suggest that investors are beginning to distinguish between companies with sustainable funding models and those whose investments depend heavily on continued access to external capital.
This could become an important turning point.
AI infrastructure spending does not have to become unprofitable across the entire industry for investment to slow. Financing becoming more expensive or difficult for weaker borrowers could be enough to delay or cancel projects.
Is the World Building Too Many Data Centres?
At present, the physical data-centre market does not suggest widespread oversupply.
According to property consultancy CBRE, vacancy rates across major US data-centre markets fell to just 1.4% in the first half of 2026, compared with 1.6% a year earlier.
Furthermore, approximately 80% of capacity under construction had already been pre-leased, up from around 74% previously.
These figures indicate that demand continues to absorb newly developed capacity.
Electricity availability, grid connections and equipment shortages are also restricting the pace at which new data centres can be completed.
However, these constraints create an important distinction.
A project delayed because electricity is unavailable does not necessarily indicate weakening demand. It may simply mean that infrastructure development cannot keep pace with customer requirements.
By contrast, a company cancelling a project despite having secured sufficient power would be a much more concerning development.
Such cancellations could suggest that expected financial returns have deteriorated or that customers no longer require as much capacity as originally anticipated.
For now, the evidence points more towards infrastructure shortages than excess capacity.
But this could change if the large volume of projects currently under development eventually becomes operational.
Falling AI Computing Costs: Good News or Bad News?
Another risk to the infrastructure boom comes from technological progress itself.
AI chips and models are becoming more efficient. Improvements in semiconductor performance, software optimisation and model design mean that certain AI tasks can be completed using less computing power.
This is positive for businesses and consumers because it makes AI services cheaper and more accessible.
However, it creates uncertainty for companies investing heavily in AI infrastructure.
If the amount of computing power needed to perform a particular task falls substantially, the value of older equipment and the revenue it can generate may also decline.
For example, an investor purchasing expensive graphics processing units (GPUs) today faces the possibility that newer chips or more efficient AI models could significantly reduce the market price of comparable computing services.
The important issue is whether falling costs encourage enough additional AI usage to offset the efficiency improvements.
If demand expands faster than computing efficiency improves, total infrastructure requirements could continue growing.
But if efficiency improves faster than demand, the industry could experience slowing investment even while AI adoption continues to accelerate.
This is why falling prices for AI computing services should not automatically be interpreted as evidence of weakening demand.
Investors need to distinguish between prices declining because technology has improved and prices declining because there is too much capacity.
How Might the AI Boom Eventually Peak?
The end of the AI infrastructure boom may not begin with an obvious collapse in semiconductor orders or a sudden reduction in technology company spending.
A more plausible warning could emerge gradually through deteriorating investment economics.
Consider the following sequence:
- AI infrastructure investment grows faster than the revenue and cash flow it generates.
- Companies increasingly depend on borrowing, leasing and other financing arrangements.
- Lenders demand higher interest rates or stronger protection against losses.
- Infrastructure developers become more selective about new projects.
- Companies postpone investments that no longer offer sufficiently attractive returns.
- Semiconductor orders and data-centre construction eventually slow.
There is some early evidence of financial pressure, but the complete sequence has not yet developed.
Importantly, the first warning may come from financial markets rather than technology companies themselves.
Credit spreads — the additional interest lenders demand above safer government bonds — could become particularly useful indicators.
If borrowing costs for AI infrastructure projects rise persistently relative to the wider corporate bond market, it could indicate that lenders are becoming increasingly concerned about financial returns and repayment risks.
Another important indicator is the rental price and utilisation of advanced AI chips.
If computing capacity expands faster than demand, equipment utilisation could fall, forcing operators to reduce rental prices.
Such a development would directly undermine the financial returns generated by existing infrastructure.
For now, available pricing evidence is mixed, and there is insufficient reliable utilisation data to conclude that widespread oversupply is emerging.
What Would a Slowdown Mean for Investors?
Not all AI-related investments would necessarily suffer equally if infrastructure spending slows.
Companies supplying advanced chips, semiconductor manufacturing equipment, specialised memory and data-centre electrical systems are particularly exposed to the pace of new infrastructure construction.
A meaningful reduction in spending could weaken their revenue growth and earnings expectations.
Data-centre operators and infrastructure lenders may also face pressure if customers defer projects, renegotiate contracts or encounter financing difficulties.
However, the implications for major technology companies could be more complicated.
If infrastructure spending slows because demand for AI services disappoints, it would likely be negative for both equipment suppliers and technology companies.
On the other hand, if spending slows because AI becomes significantly more efficient, the outcome could be quite different.
Companies such as Microsoft, Alphabet and Amazon might benefit from lower infrastructure costs, improving cash flow and potentially higher profit margins.
Businesses adopting AI could also benefit from cheaper computing services and stronger productivity gains.
In other words, a slowdown in AI infrastructure investment would not necessarily mean that AI adoption or its economic benefits are slowing.
It could instead mark a transition from an investment-heavy phase towards one in which businesses capture more of the financial benefits of the technology.
For investors, distinguishing between these two scenarios will be essential.
How Likely Is an AI Investment Slowdown?
Our current assessment is that there is a high, although highly uncertain, probability — broadly 50% to 75% — that AI infrastructure capital expenditure growth will slow materially over the next 12 months.
This is a judgemental assessment rather than a statistical forecast.
It is also important to distinguish between slower growth and an outright contraction.
A company increasing investment by 15% next year rather than 50% this year would represent a substantial slowdown in capital expenditure growth, even though the amount invested would still increase.
Given the extraordinary pace of investment in recent years, some moderation would not be surprising.
An outright decline in infrastructure spending would require stronger evidence than is currently available.
We would become significantly more concerned if several developments occurred together:
- Major technology companies begin lowering their forward capital expenditure guidance.
- GPU rental prices decline persistently alongside weakening utilisation.
- AI infrastructure borrowing costs rise significantly relative to the broader corporate bond market.
- Developers begin cancelling projects despite having secured electricity supply and customers.
- AI-related profitability fails to improve despite continued investment.
Conversely, stronger evidence of AI-related profits, healthy computing demand and improving free cash flow would reduce the likelihood of a damaging downturn.
Conclusion: Strong Demand, Increasing Financial Risks
The AI infrastructure boom is not over. Demand for advanced semiconductors remains strong, data-centre occupancy is high, and major technology companies continue to invest enormous sums.
Nevertheless, the financial risks are becoming increasingly important.
The key concern is not that AI will fail to transform the economy. It is that the investment required to support this transformation may be growing faster than the profits it can realistically generate.
History offers numerous examples of transformative technologies accompanied by excessive infrastructure investment. Railways, telecommunications and the internet all delivered substantial long-term economic benefits, but not every company financing their expansion earned attractive returns.
AI could follow a similar pattern.
For investors, the most important question is therefore shifting.
It is no longer simply how much companies will spend on AI. It is how much financial return they can earn on that spending.
That may ultimately determine when the AI investment boom peaks, which companies emerge stronger and which investors are left bearing the costs.
This commentary reflects information available as of 9 October 2026. The probability assessment is illustrative and judgemental, not a statistically derived forecast. It should not be interpreted as a prediction that AI adoption or absolute infrastructure spending will decline.
Sources and further reading
(Assisted by AI)
