THE AI BUILDOUT: FOLLOWING THE MONEY
This article has been prepared by Fundhouse and is reproduced by LifeMap Financial Planning for general information. It reflects Fundhouse's views as at the date of publication and those views may change. It does not take account of your individual circumstances and should not be treated as personal financial advice or a personal recommendation.
Artificial Intelligence (AI) continues to dominate market headlines, focusing on the remarkable capabilities of AI models, the race for computing power, and whether the technology will transform industries and impact employment. But there is a question that gets much less attention: how is this infrastructure being funded?
Recent analysis suggests that the world's largest technology companies have built up over $1.5 trillion of future purchase commitments related to data centres, chips, energy contracts and computing capacity. To put that into perspective, capital heavy industry like “Big Oil” (think Exxon, Chevron, Shell, BP and Total) spend around half this number a year in aggregate on drilling, field equipment and oil and gas extraction. In other words, $1.5tn is a staggering figure for an industry that has prided itself on being ‘capital light’ (meaning it doesn’t need to spend huge amounts of money on physical assets).
Positively, we know these hyperscaler companies (think of hyperscalers as the mega-landlords of the internet, such as Microsoft, Alphabet, Amazon and Meta) generate enormous amounts of profits and cashflow from their existing businesses. Their balance sheets remain relatively healthy, and many still enjoy strong credit ratings, meaning they are considered reliable borrowers. But there is a catch. In many cases, their capital spending requirements are growing faster than their operating cashflow. As a result, they have started to borrow money to fund these commitments. And they are using complex financial engineering to achieve it.
While hyperscale companies still own many of their data centres directly, they are increasingly using outside investors and financing partners to help fund new projects. Why? It’s not because they can’t afford it, but building a data centre is incredibly expensive. If a company owns all of these assets itself, its balance sheet grows rapidly and it can become harder to maintain the strong financial metrics that investors like to see. Instead, they turn to financial engineering to help. One way is by setting up a separate legal entity (called a Special Purpose Vehicle - SPV), to own the data centre, so it sits on the SPV’s balance sheet rather than the hyperscalers. The hyperscaler may own a small stake in the SPV, with the remaining ownership held by the data-centre owner. The SPV then borrows money to fund the data centre, and the hyperscaler pays a rental fee (over 20 years, for example). This approach helps technology companies continue expanding their AI infrastructure while managing the impact on their balance sheets.
Although the SPV is officially on the hook for all the debt, it is often implied that the hyperscaler is ultimately responsible to repay the debt. This is because the hyperscaler is often required to continue to make the rental payments under any circumstances, even if the data centre is damaged, the tech market crashes, or the hyperscaler decides they no longer need the capacity. What is called rent is in effect a debt obligation.
These companies are continuing to increase their borrowings. According to Vanguard1, “between 2020 and 2024, the five hyperscalers combined issued roughly $35 billion of debt each year, on average. In 2025, that figure jumped to $93 billion. Year to date, they have issued approximately $132 billion.” This is a challenge because a lot of investors are taking the same risk twice. They may own shares of these technology companies through their equity funds, while also lending money to those same companies through corporate bond funds. We have concerns that if something happens to the AI narrative that it could have a wider impact on both the equity and corporate bond market.
At Fundhouse, we continue to admire the technological achievements of the companies leading the AI revolution. These businesses have transformed industries and continue to innovate at an astonishing pace. However, our investment process requires us to assess not only the quality of a company, but also the price being paid and the expectations embedded within that price. We believe that many of the hyperscalers are expensive and have therefore shifted your portfolios away from both the investments and lending in these companies. Our positioning is not driven by a forecast that AI spending will fail, nor by concerns over any single headline. Rather, it reflects our long-standing preference to allocate capital towards areas where expectations appear more modest and valuations more attractive.
Fundhouse is the trading name of Fundhouse Bespoke Limited. Fundhouse provides investment management services and does not provide financial advice. Importantly, this note does not represent investment advice and any reader should always speak to their financial adviser before making any investment decisions. Please note that the value of any investment may go down as well as up and you may lose capital when investing and the value of your investments may not always increase. Please ensure that you are comfortable bearing financial losses and that you are comfortable taking a long-term investment view of five years or more.
