Analysis
The circular economy of AI: How big tech is financing itself
Microsoft, Nvidia, and OpenAI are investing in each other’s promises, a trillion-dollar loop that may not survive its own momentum.
The "AI bubble" is probably the first bubble in history whose Wikipedia entry was created before it burst. It began last summer, and, surprisingly, the first to officially link the terms "bubble" and "AI" was OpenAI CEO Sam Altmanת the man behind ChatGPT, which caused a big bang when it was released to the public in late 2022.
"Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes," Altman said in an interview with The Verge, sending everyone scratching their heads and waiting for the rest of his sentence: "Is AI the most important thing to happen in a very long time? My opinion is also yes."
Since the dam broke, and after the words multiplied and joined the chorus warning of the bubble’s inflation, the numbers have also begun piling up. These figures are so large, and so historically unprecedented, that it’s difficult to fully grasp them. And of course, the same Altman continues to feed the system with those impossible numbers. Calcalist presents just a few of them to illustrate the state of AI at the end of 2025.
$5 trillion: Nvidia’s market cap soars to record high
You can’t begin to talk about AI without talking about Nvidia. The company, once known mainly to gamers, recognized the coming revolution and jumped in with full force. Today, its market capitalization is higher than the GDP of every country in the world except the United States and China, according to World Bank data.
The chipmaker, whose processors are almost the only ones capable of running advanced artificial intelligence applications, surpassed the $5 trillion mark last October, just three months after becoming the first company to cross $4 trillion.
Now, the suddenly “boring” $4 trillion club includes only Apple and Microsoft, and few remember that just two years ago, in the summer of 2023, Nvidia passed the $1 trillion mark for the first time. Nvidia’s business performance remains exceptional, and even after two years of extraordinary growth, it continues to surprise Wall Street almost every quarter.
But as with everything, there’s an asterisk. Nvidia currently enjoys near-monopolistic power in the AI chip market, allowing it to charge exceptionally high prices. Together with OpenAI, it is both the root and the main engine of what is now called the “circular economy of AI” - investing hundreds of billions of dollars in companies that, in turn, use that money to buy more of Nvidia’s chips.
$1.15 trillion: OpenAI’s expanding commitments in hardware and cloud
The great circle of artificial intelligence is powered by OpenAI, which, although not yet a public company, still sets the tone for the entire stock market. Its revenues remain tiny compared to the scale of its commitments, projected at $13 billion in 2025, but it has already signed a $300 billion cloud infrastructure agreement with Oracle, $90 billion with AMD, and just last week a $38 billion contract with Amazon Web Services.
How will a private company, which, until recently, was a nonprofit, finance all this? Easily, through enormous investments, most of them from Nvidia, which has pledged $100 billion, and from Microsoft, one of its largest shareholders. This has created a situation in which the value of many tech giants now depends heavily on contracts with OpenAI, contracts whose financing remains uncertain.
This is particularly evident in Oracle’s case, as much of its future revenue depends on Altman’s company. To meet OpenAI’s growing infrastructure demands, Oracle is investing tens of billions of dollars in Nvidia chips. Its dramatic announcement of the OpenAI deal, and even earlier reports of sharply raised forecasts, sent its stock up 30% in a single day. But soon after, questions arose about Oracle’s massive debt, which now reflects a 6x debt-to-equity ratio.
After all, if OpenAI suddenly doesn’t need all of Oracle’s cloud capacity, or can’t finance its vast commitments, how will Oracle pay Nvidia? And perhaps more crucially, what will it do with all of those enormous investments?
$700 billion: Tech giants' investments in AI infrastructure over the next two years
Altman and Nvidia CEO Jensen Huang have thrown the entire tech world into a frenzy of massive investments. These are not financial investments or software acquisitions that can be canceled with a click, but heavy capital expenditures. This item, CAPEX, is already starting to weigh on the earnings reports of large tech companies and cut into their cash flow.
The decline in free cash flow at Amazon, Google, Meta, and Microsoft has been evident since 2024, and according to analysts’ forecasts, between the last quarter of 2024 and the first quarter of 2026 their combined free cash flow is expected to shrink by 43%.
The spending is directed mainly toward building data centers and meeting enormous electricity demands to support them. According to a recent study by Bain & Company, AI companies will need $2 trillion in annual revenue to finance the infrastructure required for AI projects by 2030. At the current pace, based on existing projections and assumptions, they are short by roughly $800 billion in projected revenue. In other words, the numbers don’t quite add up.
500%: The surge in the number of data centers worldwide in the last 20 years
The data center, the unglamorous physical backbone of the cloud, has become the most rapidly growing type of infrastructure in recent decades. To deliver every “cloud” service, somewhere in the middle of nowhere there must be a massive server farm actually providing those services. That’s where Nvidia’s chips go, and where vast amounts of electricity are consumed, another major cost of artificial intelligence that the public rarely considers.
There are currently about 11,000 such facilities worldwide, but this is only the beginning. Mark Zuckerberg’s Meta alone plans to build a data center in Louisiana the size of Manhattan and has taken out a loan of nearly $30 billion to finance it. Around the construction of these facilities, which companies such as Oracle and Amazon are developing at a dizzying pace, a new industry has emerged to accelerate building timelines and reduce energy consumption.
Yet despite such efforts, a recent report by the Rhodium Group predicts that by 2040, server farms will consume 14% of all electricity in the United States, up from just a few percent today. These projections are also fueling energy-sector stocks. Texas-based power company Vistra, for example, has reached a valuation of $65 billion after surging sevenfold in the past three years. Constellation Energy, which is building nuclear power infrastructure for Microsoft, went public on Wall Street just before the generative AI boom and has since jumped 700%, reaching a valuation of more than $100 billion. Another rising star in the field is Oklo, a manufacturer of small modular nuclear reactors designed for data centers, whose stock has climbed 500% in the past year alone.
80% of the gains in the US stock market since the beginning of 2025 are attributed to AI-related companies
All the numbers together, on the one hand, ignite investors’ imagination on Wall Street, but at the same time trigger a mild sense of unease. Although the AI revolution is real and its impact is felt across nearly every sector, it is difficult to ignore the “house of cards” element surrounding it. This is mainly due to the fact that the direction and level of the S&P 500 index are now determined almost entirely by Big Tech stocks.
The US stock market, and particularly the S&P 500, has suffered from abnormal concentration for more than five years, especially since the pandemic. But while the concentration was once spread broadly across technology stocks, it has recently narrowed sharply to one segment: AI.
There is now a clear divide between companies meaningfully engaged in artificial intelligence and those that are not, or worse, those whose business models are threatened by it. The entire technology sector currently represents 35% of the index, an all-time high. In practice, the concentration is even greater, as five companies - Nvidia, Google, Microsoft, Apple, and Amazon - account for 30% of the S&P 500’s total value.
In other words, investing in the S&P 500, an index that includes nearly 500 companies and is considered the most representative of the American economy, no longer provides the diversification that passive investors expect. On the other hand, investors have little reason to complain so far, since when it comes to returns, this concentration has worked in their favor: over the past five years, the S&P 500 has nearly doubled (up 95%) and has climbed another 16% since the beginning of 2025.
AI startups have raised $160 billion since the beginning of the year
The private capital market is not sitting out the party either, though in theory, the risk is more dispersed. In practice, the same institutional investors driving the Wall Street rally are also fueling venture capital and private equity funds. The shift toward AI is clear and one-sided: in the past two years, one out of every two dollars invested by the venture capital industry has gone into AI startups, led, of course, by OpenAI and Anthropic, which have raised tens of billions of dollars in their recent rounds.
OpenAI is now the only private company in history to have reached a valuation of half a trillion dollars, the level at which a massive secondary transaction was recently completed, allowing early employees to sell shares, all without ever turning a profit. In 2025, this concentration reached new heights, with a record 70% of all VC funding going to AI. That helps explain why saying a company “works in AI” today is as broad as saying it “develops software.” The variety is vast, and there is no longer much sense in treating AI investments as a distinct category. AI is, quite literally, eating the world.
Still, beyond the major names behind large language models, the capital surge into AI is also remarkable for its focus on very young companies. Funds are pouring tens or even hundreds of millions of dollars into startups often less than three years old, many with only a few dozen employees. That hasn’t stopped them from leading $100 million-plus “mega rounds” for one-year-old firms.
This phenomenon has made each employee in an AI startup exceptionally valuable, sometimes worth tens of millions of dollars on paper. A striking example comes from Israel: Decart, founded just two years ago during the war, recently reached a valuation of $3.1 billion while employing only 70 people. In other words, each Decart employee is now worth about $44 million.
95% of AI projects within organizations fail, according to MIT research
All these numbers represent the vision of AI, expectations of where the world might be in a few years. But for now, despite the rapid adoption of AI tools and the constant buzz surrounding them, organizational implementation is falling short of expectations.
This matters because for companies to justify tens of billions of dollars in investment, AI needs to become indispensable, deeply integrated into everyday operations across industries. While consumers have embraced AI quickly, with ChatGPT likely the fastest-growing consumer product in history, that’s not enough to support the numbers. A user asking ChatGPT for medical symptoms or vacation tips does so either for free or for a modest $20 monthly subscription.
Indeed, a study by MIT published last summer cast serious doubt on the effectiveness of corporate AI initiatives. The report, The GenAI Divide, which examined the integration of generative AI into daily organizational workflows, found that 95% of AI agents developed within companies, systems designed to automate or replace routine human tasks, simply don’t work.
Globally, companies have invested $40 billion in such AI applications, yet 95% have failed to deliver measurable results or any return on investment. In other words, most enterprise AI adoption projects are stalled, the best-kept secret in the business world today. Everyone loves to talk about the future of AI and brag about creating stunning presentations or videos with a few clicks. But when it comes to complex business use cases, the story is far less glamorous.
