The familiar AI gold rush has come to an end. Technology will continue to advance, but the era of unlimited spending and unquestioned valuations is over. The biggest threat to the prosperity of AI comes from financing that sustains the entire ecosystem. Trend lines are plummeting. Global AI financing amounted to 239.1 billion dollars in the first quarter of 2026, 153.1 billion dollars in the second quarter, and 103.3 billion dollars in the third quarter. Has this reached rock bottom? Companies at the core of the AI industry may hold the key to finding the answer.
A significant revaluation has already begun.
Since NVIDIA set a record on October 6th, its market value has evaporated by nearly $2 trillion, representing a decline of over 30% from the peak of what was once the world's most valuable company. In less than six weeks, that glory was gone. On November 12th, Jensen Huang resigned after serving for 33 years. The chief financial officer took on the role of interim CEO, and Congress has called for Jensen Huang to testify. Now, this company, located in the AI economic center, faces both succession risks and national security threats. The recent financial reports may have alleviated some of the impact, but there are still questions that must be answered.
Having excellent technology does not mean that any valuation is reasonable. The people behind the technology must be able to manage the company and its mission responsibly.
The $570 billion problem
Bloomberg's chart on circular AI financing shows that NVIDIA is at the center of a global network that includes Microsoft, Google, Amazon, OpenAI, Anthropic, Oracle, and CoreWeave. Billions of dollars flow between these companies in the form of equity investments, chip purchases, cloud contracts, and guarantees. But how much of this demand is truly natural, and how much is financed by the companies that benefit from it?
Chip manufacturers invest in AI companies, which in turn purchase their chips; cloud service providers invest in AI developers, who promise to reinvest billions of dollars back into cloud services; lenders use these commitments as collateral to provide financing for data centers. NVIDIA has provided a guarantee of up to $105 billion for the leasing obligations of a data center park being built in Ohio for OpenAI, and OpenAI itself is also a customer of NVIDIA.
Capital flows in a cycle, and valuations rise accordingly. This cycle can drive real growth, but it can also make independent demands seem stronger than they actually are. When suppliers finance customers, and both parties' valuations are based on the same income stream, a single default can quickly spread its effects.
It is irresponsible to call the entire system a Ponzi scheme, but it is even worse to claim that it is safe. Morgan Stanley estimates that by 2026, the global issuance of AI related debt will approach $570 billion, more than twice that of last year. As of the end of May, nearly $236 billion has already been issued. Someone will have to pay for this capacity.
The Judgment of the Credit Market
In October, the yield of the CoreWeave 2031-year bond broke through 16%, a level that traditionally indicates difficulties. Reuters reported that the probability of CoreWeave defaulting within five years is close to 63%. Nevertheless, NVIDIA still has an agreement to purchase any unsold production capacity of CoreWeave before 2032.
Lenders now require higher yields and more stringent contractual terms. Credit investors have begun to clearly distinguish between their financial strength and their enthusiasm for AI. As credit tightens, we must begin to ask: in this cycle, who will ultimately bear all the inevitable losses?
In the world of data centers, facts speak louder than speculation.
Yes, NVIDIA still has substantial profits. Microsoft, Google, and Amazon also possess extremely strong financial resources. However, the highly indebted neoclouds and data center developers are facing challenges such as power shortages, construction delays, rising borrowing costs, and local opposition.
Before the demands and profitability have been proven, billions of dollars have already been invested in the AI campus. Data centers will exist in the long term, but speculative projects are facing difficulties. I expect that between 2028 and 2031, there will be a wave of projects that encounter problems and underutilization. The ultimate survivors will have access to cheap electricity and customers with good credit, and they will have signed long-term contracts. Announcing a gigawatt of capacity does not equate to earning one dollar.
AI Politics is changing.
The EU's AI legislation is already reshaping compliance requirements. States in the United States are drafting and passing various legislations, and after the recent mid-term elections, Washington will push harder on issues related to security, energy costs, competition, and national security. The idea of building first and then addressing problems no longer exists. Regulation will determine who can expand and who cannot.
The Re-emergence of the Fiber Optic Construction Boom
The wave of fiber optic construction at the end of the 1990s brought about real technological progress, but it also destroyed a large amount of investor capital. The Internet survived, but many of the companies that financed it did not. AI truly exists, and there is also a great demand for computing power. Technological inevitability does not equate to financial sustainability, but I am still optimistic about the role of AI in the economy and various industries over the long term. In contrast, I am much more cautious about the financial engineering surrounding AI.
Today's situation cannot yet be classified as a debt crisis. However, what we are witnessing is that the financial system that supports AI is facing its first severe stress test. The risk lies in the possibility that financing assumptions could collapse before they can technically realize their economic returns. The winners in the next phase will be those companies that can obtain sustainable returns on their invested capital. The importance of the number of GPU and gigawatt-scale project announcements will diminish.
As this stress test continues, the most closely interconnected participants may discover that their largest customers and their largest creditors are actually in the same circle.
The views expressed in the articles commented on by Fortune.com represent only the author himself/herself and do not necessarily reflect the views and positions of Fortune.











