In recent months, OpenAI has signed agreements worth more than $1.4 trillion in infrastructure—data centers—that will be built in the next 8-10 years. The problem is that to do this they will have to face gigantic credit obligations that will require billions of dollars in 2026, and it is not at all clear how they will be able to face those payments.
bad business. Your current income structure certainly does not support such debt. Sam Altman indicated in X They expect to end the year with more than $20 billion in annualized revenue. Even so, they will continue to be in (very) red numbers, although they also promise that by 2030 they will enter “hundreds of billions of dollars“The accounts do not come out, and that makes it virtually impossible to meet all credit commitments without resorting to extraordinary forms of financing, refinancing or…
Rescue. Last week there was already talk about how both NVIDIA and OpenAI had dropped the possibility that papa state had to rescue them in case of a debacle. Sam Altman himself clarified shortly after that “we don’t have or want government guarantees (…) and taxpayers should not bail out companies that make bad business decisions.”
He does not want a rescue, but he does talk about agreements with the government. Although Altman clarified that he was not seeking government bailouts, he did make it clear that there is a debate about a strategy to face these loans:
“The only area in which we have discussed loan guarantees is in the framework of supporting the construction of semiconductor factories in the United States (…) Of course, this is different from governments guaranteeing the construction of data centers for private purposes.”
It seems impossible for them to get out of this. As analyst Ed Zitron explains in your newsletterOpenAI needs $400 billion over the next 12 months to meet those credit obligations. Not only that: for him OpenAI’s plans to build chips with Broadcom and fill a 1 GW data center or create similar data centers with AMD chips Instinct or with the Vera Rubin from NVIDIA “There is not enough time to build these data centers. And if there was enough time, there would not be enough money. And if there was enough money, there would not be enough (electrical) transformers, electrical grade steel or specialized talent to supply the electricity for these data centers.” That’s all a gigantic house of cards.
Possible strategies. OpenAI increasingly depends on debt issues and strategic investors, but also on those circular financing agreements it has reached with several companies. SoftBank, which already invested in OpenAI, could expand its bet, especially now that it has just sold completely all its participation in NVIDIA. Although the sale has obtained almost $6 billion, the figure is still insufficient even if it is invested in OpenAI. And of course OpenAI could achieve explosive revenue growth, but it is far from clear that it will achieve such growth in the short term.
The other solution: slow down. OpenAI’s excessive ambition makes everything surrounding its agreements and proposals absolutely enormous, and that also affects its credit obligations. Adopting a slightly less risky strategy and setting more feasible deadlines could reduce the financial stress to which the company is subject… but it would also raise doubts about the growth promises that Altman and his people have made for years.
Going public? Another option for OpenAI is to go public now that it has managed to complete the restructuring and has become in a for-profit organization under the umbrella, of course, of the OpenAI Foundation. In recent days there was talk about how this option would allow the company get a billion dollar valuationbut the analysts they doubt that something like this is going to happen in the short term… if it happens at all.
And the bubble keeps growing. Analysts like Scott Galloway they explained recently that the valuations of companies like NVIDIA, Oracle or AMD are conditional on those “handshake” agreements with other companies like OpenAI. For him, these agreements have no substance: there is much ado about nothing. If the market ends up losing confidence, the consequences could be dire and the hypothetical bubble could burst.
Source: Apollo Academy
All eggs in one basket. Stock market concentration does not help. Torsten Sloj, chief economist at Apollo Global Management, has been talking for some time about the dangerous concentration of the S&P 500 index in 2025. A few days ago published a graph in which it showed the returns of various assets in the last five years, and there is a clear conclusion: while “the Magnificent Seven” have grown exceptionally, the rest have barely done so.
Image | Steve Juvetson
In Xataka | There is a race in which Anthropic is winning over OpenAI: that of being profitable
GIPHY App Key not set. Please check settings