From baggy pants to bucket hats, Y2K fashion isn’t the only trend making a comeback. Growing familiarity with artificial intelligence (AI) is quickly echoing the hysteria when the internet was introduced back in the late 1990s, creating an obsession for anything that had a dotcom angle and funding speculative hype that later misallocated capital.
The determination to become a major AI player is evident in the amount of money pouring in. Between US$700 billion and US$1.5 trillion is expected to flow into AI-related infrastructure this year alone, with total spending expected to reach US$3 trillion by 2029, spurring unprecedented demand for semiconductors, data centres, and reliable energy sources to power the industrial transformation. It’s not just cash. Valuation multiples are reaching historical highs, inflating the paper value of new companies.
[See more: AI education is now compulsory at public schools in Beijing]
Even as cheques are being signed, the fevered pursuit of AI may be prompting unrealistic expectations. The release of ChatGPT-5 provoked an almost immediate backlash for not materially narrowing the generative AI gap despite costing three to five times more to train than its previous models, adding to the chorus of criticism that few commercially viable uses for the AI technologies have emerged. An MIT NANDA paper published this past August found that 95 percent of generative AI pilots did not improve corporate profitability.
Why is there so much attention on spending?
These findings have yet to prompt concerns as investors continue to reach into their wallets. Unlike the speculative heyday of the early internet, tech enthusiasts argue that companies are using proceeds from proven business models rather than relying on bank loans, making them less vulnerable to interest rate increases that compressed balance sheets before the dotcom burst. Another consideration is the legislative passing of the One Big Beautiful Bill Act, enabling companies to defer AI-related capital expenditures by recognising those costs later.
However, these conditions can turn. Sudden inflation spikes could force central banks to aggressively normalise interest rates while the overproduction of hardware components could depress their market values. Even semiconductor chips designed for today’s AI models may become obsolete as architectures evolve. Still, the potential of being at the forefront of new multi-trillion dollar markets is justifying these risks, comments Andrew Pearson, managing director of Intelligencia Limited, an analytics and AI consultancy based in Macao.
“Those deploying resources are looking at current market values and simply adding a zero to them,” he tells Macao News, adding that fear of missing out on any AI related opportunities is also a factor in the decision making. “The general bet is that the gains from one or two winners will compensate for the potential losses when other investments go sour. ”
But the latest scaling strategies are making some nervous. Trillions of dollars in cross-investment partnerships between advanced hardware and AI software companies are implementing vendor finance models that supercharged the dotcom bubble. Others involve buying AI infrastructure and renting it out, a version that Jefferies equity strategist Christopher Wood describes as the digital equivalent of the WeWork model.
“If there is going to be an unravelling, it is likely to show up there first,” he writes in a note to clients.
What happens next
If markets question the ability to monetise these large bets, the valuation gains could go in reverse. With cheaper AI models like China’s DeepSeek becoming widely available, subsequent retreat from AI capital expenditures eliminates a growth engine that major industries have become reliant on. Underscoring the links between investments and the economy, JP Morgan estimates that AI expenditure accounted for 1.1 percentage points of first half GDP, even though the sector only accounts for four percent of output.
January’s market sell-off provided a dress rehearsal if hyperscalers and tech companies suddenly decided to stop spending, Woods describes, arguing that it will be the adoption of cheaper open sources like DeepSeek that will inevitably generate commercially viable use cases for AI technology.
[See more: DeepSeek is rushing to release its next AI model]
Beyond the bubble
As that transformation unfolds, calls for global coordination and a recognised governance framework become increasingly urgent, with multiple public stakeholders working with private companies to pinpoint overlapping interests to ensure that benefits of AI technology can be shared without hampering innovation.
Such tropics are bound to come up at the UNU Macau AI Conference 2025, which convenes today at the Galaxy Arena. Under the theme “AI for Humanity: Building an Equitable Digital Future,” the conference aims to bridge divides that ensure that the benefits of AI are within reach.
More information about the UNU AI Conference can be found here.
