The AI revolution: learning from past bubbles

Artificial intelligence is transforming industries and reshaping how we work, communicate and invest. History reminds us that every great technological leap attracts speculative excess. The railway mania of the 1840s and the dot-com boom of 2000 both began with genuine innovation, drew in enormous capital, and ended in overbuild and collapse. Portfolio Manager JC Xue considers whether the current AI cycle will follow the same pattern.

Faster feedback, fewer missteps

A key difference between today’s AI surge and the dot-com era lies in adoption speed. In the 1990s, internet use was still novel and profitability distant. Consumers had to buy a computer, go online, learn to navigate the web, and then trust a new e-commerce model before spending a cent. The long lag between investment and returns made it easy for capital to pour in unchecked.

That lag has disappeared in the AI speculation. Billions of people already have internet-connected smartphones, and can use tools like ChatGPT instantly. Once the computing infrastructure — GPUs for training and servers for inference — is in place, user adoption and monetisation happen almost immediately. Faster feedback on investment decisions means fewer opportunities for the kind of prolonged overbuilding that defined earlier bubbles.

A self-limiting supply side

The supply side also looks different. The fibre-optic buildout of the dot-com era was virtually unconstrained, with multiple firms laying overlapping networks. In contrast, the AI boom depends on a handful of critical chokepoints — most notably TSMC’s advanced chip-making capacity. You cannot simply hire more builders to ramp up chip production: these natural bottlenecks limit how quickly capital can flood into the system.

Builders and users are the same

Another important distinction is that today’s largest investors in AI infrastructure are also its first and biggest users. Microsoft, Google and Meta are integrating AI directly into their existing businesses — enhancing ad targeting, improving productivity and embedding new AI tools like Microsoft Copilot. Unlike the internet infrastructure firms of the 1990s, these companies see clear and immediate returns on their investment.

The risk of repetition

Even so, the ingredients for a bubble are present. The race towards artificial general intelligence has become an arms race among tech titans. Ambition and competition can override caution, leading to redundant investment as each player builds its own vast data-centre network. At the same time, valuations for early-stage AI firms are beginning to stretch credulity — some AI startups command multi-billion-dollar valuations before earning a cent of revenue.

Staying disciplined

For investors, the lesson is familiar. Revolutionary technologies can create immense long-term value, but they also breed overconfidence. At Foord, we prefer AI beneficiaries whose valuations remain anchored in reality — companies such as TSMC, Google, Alibaba and Tencent. The AI revolution may prove more grounded than past manias, but the discipline to distinguish promise from speculation remains as critical as ever.