The rapid expansion of artificial intelligence is entering a more complicated financial phase, with investors increasingly questioning whether the enormous sums committed to data centres, chips and advanced AI models can generate returns large enough to justify the spending. That debate is at the centre of a new episode of Reuters Breakingviews’ The Big View, in which economist and author Felix Martin speaks with host Peter Thal Larsen about the difficulty of assigning a reliable financial value to useful technology.

Published on November 17, the episode, titled “Where the artificial intelligence boom goes next,” examines mounting pressure in the financing system supporting the AI race. The discussion considers data-centre developers’ borrowing needs, the growing use of debt and equity by major technology companies, Nvidia’s position at the centre of the AI hardware economy and the challenge facing private AI laboratories seeking public-market investment.

Rows of server racks inside a data centre, illustrating the infrastructure investment behind artificial intelligence.
Server racks inside a data centre. Photograph via Unsplash; illustrative image, not a specific facility discussed in the Reuters podcast.

Martin, a former fund manager and author of Money: The Unauthorized Biography, argues in the conversation that the financial structure behind the infrastructure boom deserves closer attention. He describes how large cloud and technology companies that once relied heavily on internal cash generation are increasingly turning to borrowing and equity issuance to support capital-intensive expansion. The result is a financing network that extends beyond conventional bank loans and includes substantial commercial relationships between companies supplying, financing and purchasing AI infrastructure.

One of Martin’s central concerns is what he characterises as a web of trade credit and circular financing. Companies across the AI ecosystem can be customers, suppliers, investors or financiers to one another. Such arrangements can help projects move forward quickly, but they also make it harder for investors to understand where the ultimate financial risk sits and how much demand is supported by independent end-user revenue.

Nvidia occupies a particularly important position in this system because its processors are essential to many large-scale AI operations and the company has also participated in financing arrangements involving AI businesses. The Reuters discussion examines the implications of a market in which the leading chip supplier is connected not only to the sale of hardware but also to the funding and expansion of customers and partners. That concentration creates questions about how resilient demand would be if financing became more expensive or customers slowed their infrastructure plans.

Recent reporting has added to those concerns. The Wall Street Journal reported that a high-profile proposal by major financial firms to mobilise as much as $500 billion for Nvidia-related chip financing had not yet produced deployed capital. The arrangement was described as a non-binding memorandum rather than a firm commitment to provide the full amount. The distinction illustrates why headline financing pledges should not automatically be treated as money already available to build data centres or purchase processors.

The pressure is also visible among specialist data-centre operators, sometimes called “neoclouds”. These companies build or lease computing infrastructure and rent access to customers developing AI models and applications. Their expansion plans depend on access to large amounts of capital, continued demand for computing capacity and the ability to service debt over long periods. If customers delay projects or the economics of AI services disappoint, heavily leveraged operators could face refinancing difficulties.

Anthropic’s planned public offering provides another test of investor confidence. Reuters’ episode points to uncertainty around the company’s IPO ambitions as part of the wider debate about whether private AI developers can convert rapid technological progress into sustainable business value. However, the exact size and timing of an offering have been described differently in subsequent coverage. The Associated Press reported on November 17 that Anthropic was pressing ahead with a potentially record-setting IPO that could raise more than $60 billion. The company’s plans should therefore be understood as evolving rather than as a confirmed withdrawal from the public markets.

The financial challenge is not simply whether AI products are useful. It is whether the economic benefits can be captured by the companies paying for the infrastructure. A technology can transform work, improve productivity and attract millions of users while still failing to generate enough revenue for every supplier, landlord, lender and model developer involved in its production. The distribution of value across the supply chain matters as much as the technology’s overall usefulness.

That distinction is especially important because the AI build-out requires spending years ahead of some expected returns. Data centres need land, electricity, cooling systems, networking equipment and expensive processors before their owners can earn income from customers. Investors must estimate future utilisation, pricing and the lifespan of rapidly evolving hardware. If demand grows more slowly than expected, facilities may struggle to earn enough to cover their financing and operating costs.

Political opposition is adding another layer of uncertainty. Communities and elected officials in parts of the United States have challenged new data-centre projects over electricity consumption, water use, local infrastructure and the effect on household utility bills. The Associated Press has reported that political resistance and possible restrictions on construction are among the pressures facing Anthropic and other AI companies. Delays to permits or power connections could affect the timetable and economics of projects that have already attracted significant investment.

Market volatility has reinforced the questions raised in the podcast. Reuters reported on November 20 that the Nasdaq fell sharply as investors reassessed AI-related valuations, data-centre spending and the outlook for technology earnings. Nvidia and other companies linked to the AI investment cycle were among those facing selling pressure. The market movements do not, by themselves, prove that the AI industry is in a bubble, but they show that investors are becoming more sensitive to the gap between ambitious spending plans and demonstrable returns.

Martin’s broader point is that the value of a technological breakthrough and the value of the businesses built around it are not necessarily the same. Investors may agree that AI will have a lasting impact while disagreeing sharply about which companies will capture the profits, how much infrastructure will be needed and how quickly customers will pay for the services. History offers examples of transformative technologies whose early investors and infrastructure providers did not always enjoy the gains that followed.

The conversation leaves the AI industry with a difficult question: can the sector turn extraordinary technological progress into durable cash flows before the cost of financing and infrastructure becomes unsustainable? For data-centre operators, chipmakers, cloud providers and AI laboratories, the next stage of the boom may depend less on announcements about computing power and more on evidence that customers are willing to pay enough for the technology to support the vast investment behind it.

Listen to the episode: Reuters Breakingviews – Where the artificial intelligence boom goes next.

Support independent journalism: Help HeadlinesXtra continue delivering timely news, business reporting and independent journalism. Support HeadlinesXtra.