Yves here. Wolf Richter highlights a crucial aspect of the AI investment bubble that has often been overlooked: the role of leverage, which is not primarily among investors but rather at the company level where AI spending is being financed.
Currently, there is no indication of leverage compounding upon itself, reminiscent of the conditions that led to the disastrous financial crisis of 2008. This crisis is frequently mischaracterized as a mortgage crisis, when in truth, it was rooted in derivatives, specifically credit default swaps, which amplified exposure to the riskiest segments of subprime real estate debt to levels four to six times higher than the real economy could bear. Overleveraged institutions like AIG, Citibank, and several European banks bore the brunt of these inflated exposures.
What we face today is a concerning degree of investment round-tripping lacking transparency. In the lead-up to the Great Crash, layers of trusts created intricate financial webs, with investors borrowing heavily to invest in them. Although current situations are not as extreme, the circular investing creates an illusion of inflated asset values and potential misrepresentation of equity for borrowing companies. A recent BBC report captures some of this opacity:
AI-related enterprises accounted for a staggering 80% of the gains in the American stock market this year. Gartner predicts that global spending on AI is set to reach an astounding $1.5 trillion (£1.1 trillion) before the end of 2025.
OpenAI, which popularized AI with ChatGPT in 2022, is at the center of controversial financing deals. Last month, it secured a $100 billion agreement with chip giant Nvidia, the most valuable publicly traded company globally.
This deal expands Nvidia’s existing investment in OpenAI, anticipating the creation of data centers powered by Nvidia’s advanced chips.
On Monday, OpenAI disclosed plans to acquire billions in equipment from AMD, another chipmaker, potentially making it one of the largest shareholders in that company.
As a private entity recently valued at half a trillion dollars, OpenAI has garnered significant investments from tech titans like Microsoft and Oracle, the latter of which has a $300 billion deal with OpenAI.
OpenAI’s Stargate project in Abilene, Texas, backed by Oracle and Japanese behemoth SoftBank, continues to grow more expansive by the month since its announcement at the White House during President Donald Trump’s first week in office.
Nvidia also holds a stake in AI startup CoreWeave, which caters to OpenAI’s colossal infrastructure requirements.
As these increasingly convoluted financing schemes become more common, Silicon Valley experts warn they may obscure genuine perceptions of AI demand. Some have bluntly dubbed the deals “circular financing” or “vendor financing,” where companies lend to their own customers to sustain sales.
Should AI-related loans falter or necessitate restructuring, the repercussions could further destabilize the market, potentially leading the US into a prolonged period of stagnation similar to Japan’s economic slump following its real estate and stock market crisis.
By Wolf Richter, editor at Wolf Street. Originally published at Wolf Street
Much discussion surrounds the potential repercussions of the AI investment bubble, from its market impact upon bursting to its overall significance for the economy. This discourse resembles sentiments at the peak of the Dotcom Bubble, with some asserting it’s worse than the previous bubble while others contend it’s not comparable because the current situation is fundamentally different. Even if it mirrors the Dotcom Bubble and results in a bust, many argue it’s still worthwhile because AI is likely here to stay, similar to how the Internet fundamentally changed life despite the initial losses for investors.
There are numerous authoritative voices contributing to these discussions, ranging from Jamie Dimon, Jeff Bezos, and Bank of England officials to Goldman Sachs analysts and IMF Managing Director Kristalina Georgieva. The consensus examines the intricate relationships and deals within big tech and startups, featuring key players such as Nvidia, OpenAI, AMD, alongside Amazon, Microsoft, Alphabet, Meta, Tesla, and SoftBank.
OpenAI’s estimated worth of $500 billion is staggering, particularly given its substantial cash burn. With substantial arrangements amounting to $1 trillion in partnership with a handful of tech firms—highlighted by Nvidia’s $500 billion stake, Oracle’s $300 billion, and AMD’s $270 billion—the stock prices of these companies have surged sharply on such announcements.
Yet OpenAI does not have $1 trillion, instead relying on generating fresh investments from these same firms that provide its essential technology, attempting to create attractive deals that spike stock values, maintaining a cycle of inflated financial enthusiasm.
The proposition of constructing data centers utilizing Nvidia GPUs requiring 10 gigawatts of power is astonishing. To put this into perspective, the largest nuclear power plant in the U.S., Plant Vogtle in Georgia, has a total capacity of about 4.5 gigawatts, and all U.S. nuclear facilities combined generate 97 gigawatts.
But It’s Real Money Too. A Lot of Real Money.
Big Tech is channeling its vast financial resources into establishing a massive technological infrastructure reliant on data centers capable of consuming copious amounts of electricity to support AI functionality.
These “hyperscalers” are augmenting their financial flow with borrowing, issuing substantial bonds to fortify their projects, while private credit has entered this frenzy, lending considerable amounts to startup “neocloud” companies dedicated to developing data centers and renting computational power; these loans utilize AI GPUs as collateral. The future market value of a used GPU, outpaced by newer models three years from now when lenders might seek to reclaim loans, remains uncertain, yet that’s the collateral being leveraged.
Data center construction is advancing. The investment in equipment needed for these centers surpasses building costs, leading to soaring revenues for hardware and service providers. This surge is heightening stock values within this sector.
The primary bottleneck remains power supply, prompting significant investment in new generation capacities; however, establishing power plants and transmission networks takes considerable time.
Is It Really Different This Time?
The industrial scale of the AI investment bubble is reminiscent of the Dotcom era, when massive telecom infrastructure was necessary to support the Internet’s growth. Building the required fiber optics and servers demanded substantial investments—although the advances were slow, the hype surrounding these stocks ultimately led to a significant market correction, with the Nasdaq index plummeting by 78% in a short span. Many investors faced significant losses, and countless companies vanished or were sold for minimal value when the bubble burst. A year post-crash, this deterioration triggered a recession in the U.S. and a mini-depression in Silicon Valley where much of this unfolded.
Despite this, the Internet flourished in the long run, with Amazon emerging as one of the few success stories amid the collapse. Yet, it’s essential to recognize its unique path compared to the many firms that faltered.
Amid the current hype surrounding AI financing, the interplay of convoluted deals, substantial real investment underscored by leverage, and rising stock prices heightens market anxiety. Discussions abound regarding the inevitability of a selloff—a process that could carry deep ramifications for the U.S. economy—or whether the current trend can continue unabated.
Regardless of the prevailing narrative, it underscores a potent risk. Market conditions are precarious, with the potential for a turn at any moment. If a selloff persists long enough, the investment bubble could burst, reducing the intricate financial structure to mere illusions, yet AI will undoubtedly remain a cornerstone of modern technology.
