AI is, at the same time, the most amazing and dangerous technology ever conceived by man, given its potential impact on humanity. On the plus side, it will increase human productivity, remove mundane tasks from human work, enhance creativity, and educate us. On the minus side are the fear of automated control systems, the invasion of personal privacy, and the loss of jobs.
Among the dangers, the most important at the current moment is the financing and construction of AI infrastructure. We’re a long way from full implementation.
As a 40-year veteran of the information technology industry, I have witnessed the creation of amazing technological advances: the personal computer, networking, email, the internet, and the smartphone. Just comprehend the fact that the processor in today’s cell phone is a million times more powerful than the processor in an IBM mainframe computer that cost $3 million in 1970. All of these technologies I listed were driven forward by smart engineers, entrepreneurs who could sell the concepts to investors, and the finance people who put up the money to launch them.
AI represents the most challenging technology yet for several reasons. Number one, enormous funding is required to build AI infrastructure. Number two, the construction of the datacenters will take a decade or more. Number three, processor technology must advance significantly to handle the compute workload required by AI. Number four, the new AI data centers will require an enormous amount of electrical capacity. Number five, these new systems will require thousands of engineers to maintain them.
In this article, we will discuss the funding, which carries enormous risk. If those providing the financing get into trouble, we may face another Great Recession, which could cause great harm to the American people.
Today, the leading players in the AI buildout are household names, like Oracle. Meta, Microsoft, Google, and Amazon. These companies are called hyperscalers, and they are the ones building the AI data centers. The hyperscalers have contracts with AI software providers to install their software in the new centers. They have spent $1 trillion on these contracts so far, and estimates suggest $6.7 trillion will be needed to reach all potential AI users worldwide.
The scale of this project is a problem. All the hyperscalers have cash, but only Oracle has used its own cash to pay for AI infrastructure. Their cash drain caused Oracle’s stock to drop 40% recently, proving that corporate money cannot possibly satisfy the need for AI capital.
Where will the AI financing come from? Commercial banks are not an option because they don’t have enough money. Selling bonds is not an option because the risk associated with AI would require high interest rates (making them junk bonds), which would negatively impact the company’s balance sheets.
The only source remaining is private equity.
For those of you who are not familiar with the terminology, private equity loans come from “private” equity investment companies. These firms make money by lending, just as banks do. The problem and risk associated with their operation is that private equity firms are not regulated by the government the way banks are.
Banks are required to hold cash for emergency use and are subject to tests that demonstrate their ability to remain solvent in an economic crisis. Losses incurred by private equity firms are passed on to their clients. Private equity firms operate in secret, and the data behind the transactions they create is unknown. That means the average investor has no way of knowing how much risk they are taking.
Private lending has exploded in recent years, accounting for 65% of all lending in the United States this year. The idea of unregulated companies making billions of dollars by making risky loans is downright dangerous. Congress has done nothing to regulate or limit the expansion of private equity because they don’t understand the problem, and they are bought off by its lobbyists.
The 2008 housing crisis was caused by banks selling mortgage-backed securities as investments, even though the mortgages were high-risk. Home buyers without proper credit were given loans they couldn’t repay. When foreclosures began to rise, many of the companies that sold the securities went bankrupt, and the investors who bought them lost all their money. Before it was over, the American people had lost $15 trillion in home equity.
This time, the financing is more complicated. Private equity firms are buying insurance companies and pension funds to serve as sources of loans to hyperscalers. Those firms will create loans backed by the data center leases they have with the hyperscalers. The private equity firms will turn those loans into investments that the public can purchase. The investors will do well as long as the value of the data center leases remains unchanged.
What are the risks of owning these investments? If the structures are not completed in a timely fashion, or the hyperscalers opt out of the leases without a replacement tenant, their value will decrease. The leases may decrease if more data centers are built than the AI market can support.
Overbuilding these special-use facilities will render the excess capacity worthless. If any of these things happen, the investments will decline and could become worthless. There is resistance from municipalities that don’t want drains on their electrical grids. Each AI infrastructure facility consumes as much electricity as a small city!
The financing we are discussing here is the normal way entrepreneurs operate in an unregulated capitalist system. They find a project that will make them money and invest in growing that market. The market takes off, and they make the easy money. As the market matures, companies doing business in it consolidate. Consolidation occurs when larger ones acquire smaller players. Most of the rest go out of business. The average investor may make or lose money depending on the investments they hold.
The most important difference between betting on an exploding market versus a mature market is risk. Nothing is known about the exploding market because it has no track record. These companies (AI software developers) do not make a profit, so there is no way to assess the risk of investing in them. OpenAI (ChatGPT) is not forecast to be profitable until 2030. Contracts negotiated between hyperscalers and private equity firms protect each party, but not the investors served by the investment vehicles they create.
The potential damage to investors and the economy as a whole if the reality of the AI datacenter buildout and the vision presented for it do not align could be in the trillions of dollars and another recession. No one discusses the likelihood of this scenario or has plans to deal with it, because the market is too focused on exploiting the euphoria of the moment and making quick money.
Investors should be cautious, and the public should demand greater visibility into this latest and greatest technology paradigm.
Wrong Speak is a free-expression platform that allows varying viewpoints. All views expressed in this article are the author’s own.
 | A guest post by
|