A growing amount of capital is flowing into companies involved in AI infrastructure, namely data centers and semiconductors, which has massively increased the valuations of these companies on the stock market. However, to justify these valuations, investors will want to see more successful applications of AI in the coming months and years.

As we will show, there has already been progress in applying AI to improve the productivity of existing business operations, but the key question is whether we will also see new types of business models based on AI. Amid historically high valuations of AI infrastructure companies, the current lack of new types of AI-based businesses should be a cause of concern for investors. After all, AI could end up resembling the telecoms industry in the early 2000s, when investors lost confidence in the companies building the telecoms infrastructure because the expected use cases failed to materialize.

In recent months, there have been several signs of AI’s potential to boost the productivity of existing business operations:

• According to Morgan Stanley, the personal care company L’Oreal reports that an AI diagnostics machine uplifts the sales conversion rate of its products at retail counters from 10% to 73%.
• AI for advertising and promotion at L’Oreal is generating productivity gains of up to 15%.
• The head of Indian IT company Tata Consultancy Services has said that AI will result in ‘a “minimal” need for call centers’ in as soon as a year.
• Payments company Klarna revealed that its AI assistant, powered by OpenAI, is doing the equivalent work of 700 full-time agents and has helped reduce repeat inquiries by 25%.

The boost of productivity does not make everybody a winner. Already, investors’ confidence in call operators such as Teleperformance and Concentrix is waning because AI is disrupting their call-center business, leading to significant share price pressure.

Besides boosting the productivity of existing business operations, AI could create new business models. However, clear examples are still few, and even when they do emerge, the history of the 2001 telecom bubble suggests that we should temper our expectations. In the early 2000s, high expectations about the use cases of 3G network technology and fiber-optic networks led to massive investments in infrastructure and network licenses (such as the 3G license auction in Europe) for telecommunications. Nevertheless, the high expectations did not materialize, leading to a stock market crash, particularly affecting telecom companies worldwide.

It is worth noting, however, that the wave of investment in telecoms infrastructure facilitated cost reductions, paving the way for a wave of innovations in the long term. Indeed, a similar scenario is possible in the case of AI. Although we may have overestimated the potential use cases of AI in the short to medium term, it is hard for us to imagine the possibilities of use cases in the long term.