The ‘Warren Buffett of Japan’ just shone a $9b light on AI’s vulnerabilities
The so-called “Big Five” hyperscalers (Amazon, Alphabet, Meta, Microsoft, and Oracle) will spend more than $375 billion on AI chips and infrastructure this year, more than $500 billion next year, and by some estimates more than $600 billion in 2027; Roughly half will be financed from existing cash flows, but increasingly from external capital.
Although startups like OpenAI are valued at $500 billion in recent equity raises and plan to go public at $1 trillion next year, they don’t have meaningful cash generation relative to their expenses and are reliant on constantly raising equity capital and lately uninvestable (and therefore quite expensive) debt.
Nvidia’s shares fell on this news, but the sale may have a positive outcome. Credit: Getty Images
The development of the sector is too expensive to be financed entirely by stock markets. Companies will have to tap nearly every corner of the global financial system to purchase chips, build data centers, and finance the energy and water infrastructure needed to operate them.
Given the risky nature of underlying assets and their dependence on chips with relatively short half-lives that require constant reinvestment, banks are unlikely to be major direct, long-term financiers.
Private credit and private capital are already being utilized. Financial engineering, the use of off-balance sheet vehicles, short-term data center leases with guarantees against losses (to avoid them being classified as debt), and other “creative” techniques are emerging.
Vendor financing (such as Nvidia agreeing to invest $100 billion in OpenAI in $10 billion tranches, matched by OpenAI’s commitment to purchase $100 billion of Nvidia chips in similar tranches) is an increasingly common feature of the industry, one that emphasizes interdependence.
AI’s demand for computing power, or “computing” as the industry defines it, is astronomical, so the question of whether sufficient financial capacity is available to fund it is an important one.
Similarly, there is a question mark over whether the development of infrastructure (data centres, power plants and water supply) can keep up with computing demand.
JP Morgan recently estimated that approximately 122 gigawatts of global data center installations will need to be built at an accelerating pace between next year and 2030, requiring approximately 150 GW of power.
It takes three to four years to build a new gas-fired power plant and more than a decade for a new nuclear power plant; So can infrastructure construction keep pace with industry needs, and can these new energy sources effectively finance the credit risk of a single industry that has yet to prove its economics sound?
The development of the sector is too expensive to be financed entirely by stock markets.Credit: Bloomberg
In the near future, hyperscalers will be able to finance their share of the financing burden thanks to their strong cash flows and access to debt markets; however, this may require cuts to some non-AI-related investments.
As well as equity capital and possibly some limited bank lending for institutions, access to investment-grade bond markets, leveraged loan markets, securitization markets and even government financing will also be required.
What is unknown is not whether there is demand for AI products, for it is whether companies and consumers will be prepared to pay enough to get returns from their AI investments commensurate with the risk.
There are now billions of AI users worldwide, but less than 5 percent of OpenAI’s ChatGPT users, the dominant chatbot in the industry, pay for its services.
Loading
AI can be transformative, but its pioneers will crash and burn if they cannot generate the revenues and returns to deliver on their commitments and do so in the time periods and at levels that justify the scale and risk of their investments.
In the 1990s, telecommunications companies and cable companies invested heavily in new networks whose capacity exceeded demand and users’ willingness to pay.
The telecommunications and dot-com boom ended in spectacular failure, but eventually left a useful legacy of infrastructure and intellectual capital as demand for that capacity grew. It also laid the foundations for today’s dominant technology companies.
While this isn’t inevitable for an industry still in its nascent stages, it is a possible outcome for the AI boom.
Capital constraints on capital demand and the question mark over whether all the infrastructure required to meet the demands of all industry participants can be built will likely determine the near-term outlook for the industry, separating the financially weak from the strong in the process.
The Business Briefing newsletter delivers big stories, exclusive news and expert insights. Sign up to receive it every weekday morning.
