|
“The decisive test will be whether utilisation, monetisation and productivity gains develop quickly enough to justify the scale of investment before cash buffers erode and financing conditions become less forgiving.” |
 |
The AI investment cycle has become a major force in the global economy
Artificial intelligence is driving one of the largest capital expenditure cycles in modern history. The investment is already having a material economic impact, with the AI capex cycle contributing meaningfully to US GDP growth each quarter since early 2025.
The implications extend well beyond America's technology sector. Semiconductor production, data centres, power infrastructure, communications equipment and software are connecting the AI investment cycle to manufacturing and economic activity around the world.
This extraordinary scale inevitably raises a question for investors: is the AI boom becoming another dot-com bubble?
Our analysis suggests the comparison is useful, but incomplete.
Is today's AI boom the same as the dot-com bubble?
There are clear similarities. Both periods involve investors attempting to value the future benefits of a potentially transformative technology before its full economic payoff can be observed. Both have generated exceptional infrastructure investment, strong investor enthusiasm and increasingly concentrated equity-market leadership.
But there are also fundamental differences.
The late-1990s boom included many businesses with limited earnings, untested business models and a heavy dependence on continued access to capital markets. In contrast, today's leading AI investors are predominantly highly profitable, cash-generative businesses operating within an established digital economy. AI is being added to an existing ecosystem of cloud computing, enterprise software and data infrastructure rather than building that ecosystem from scratch.
The question today may be less about whether AI succeeds and more about whether the returns generated by AI ultimately justify the enormous amount of capital being invested in it.
Stronger foundations do not eliminate the risks
The financial starting point for today's AI leaders is considerably stronger than it was for many telecom and technology companies during the dot-com era. Earnings and cash generation have provided substantial support for investment, while technology-sector earnings have grown faster than share prices during much of the current cycle.
However, the financing picture is changing.
As AI investment accelerates, capex is likely to absorb an increasing proportion of operating cashflow. The next phase of the buildout is therefore becoming more reliant on external sources of capital, potentially including investment-grade bonds, leveraged finance, structured products, private and alternative capital, leases and joint ventures.
This does not make today's AI cycle identical to the dot-com boom. But it does shift the debate towards capital discipline, transparency and, critically, return on investment.
Why AI infrastructure demand could remain resilient
There are also reasons why AI infrastructure demand may prove more durable than during previous technology investment cycles.
AI adoption is already spreading through businesses, while moving from experimentation towards integration within everyday workflows could substantially increase computing requirements. Meanwhile, physical constraints across advanced semiconductors, networking, data centres, electricity generation, transmission and cooling mean additional capacity cannot necessarily be created quickly.
Cheaper AI models could complicate the outlook for individual frontier-model providers, but they could also make AI economically accessible to a much larger universe of businesses and consumers. Falling model costs therefore do not necessarily imply falling demand for compute or AI infrastructure.
Where could the AI capex cycle go wrong?
We see several risks that deserve particularly close attention.
- Circular financing: AI companies increasingly operate as investors, customers, suppliers, tenants and guarantors within the same ecosystem. These arrangements can support investment, but can also mean revenues, financing and demand are less independent than they initially appear.
- Political and geopolitical constraints: Local opposition to data-centre construction, pressure on power and water infrastructure, US-China technology competition and potential restrictions on frontier-AI development could all disrupt investment plans.
- Equity-market concentration: Technology companies and hyperscalers now account for more than half of the MSCI World Index's market value and more than 40% of its earnings. This concentration means changes in expectations for a relatively small number of businesses could have unusually broad market consequences.
- The US wealth effect: Rising equity wealth has become an important support for US consumption. A significant reversal in AI-related equities could therefore weaken confidence and spending as well as the assumptions underpinning future AI investment.
- Tighter financial conditions: History suggests that investment bubbles do not necessarily end simply because valuations become expensive. Higher interest rates, wider credit spreads or reduced risk appetite can provide a catalyst that causes crowded investors to reduce exposure simultaneously.
AI can succeed even if some AI investments fail
This may ultimately be the most important lesson from previous technological revolutions.
A technology can transform the economy while simultaneously producing periods of overinvestment, consolidation and substantial losses for some investors. The internet achieved its early promise despite the collapse of the dot-com bubble. There is no contradiction in believing that AI could prove similarly transformative while questioning whether every project, company or financing structure created during the investment boom will generate an adequate return.
The current AI capex boom therefore appears more financeable and commercially grounded than the late-1990s technology cycle, but it is not immune from excess.
The critical test will be whether utilisation, monetisation and productivity gains develop quickly enough to justify the extraordinary investment being made before cash buffers decline and financing conditions become less forgiving.
