원문 보기
Before the 2008 financial crisis, Wall Street believed the American housing market could not collapse. Michael Bur took the other side of that bet. Now he is standing against the crowd once again, but this time he is not betting against subprime mortgages. He is betting against two of the most aggressive players in the artificial intelligence race, Oracle and Nebus.
Oracle is reporting record revenue. Nebius grew its revenue by 684% in just one year. But behind that extraordinary growth are enormous data centers, billions of dollars in chips, and a financing burden that keeps getting heavier. So Bur's question is surprisingly simple. Is artificial intelligence really a bubble?
Or is the bubble not the technology itself, but the way we are paying for it? On August 7th, 2026, Michael Bur revealed that he had taken short positions in Oracle and Nebus. Shorting a stock is essentially a bet that its price will fall. Bur said he shorted Oracle at roughly $145 per share and Nebius at roughly $212.
This was more than a prediction about two stock prices. Bur believes much of the AI industry is accumulating debt, long-term leases, and purchase commitments tied to infrastructure that may take years to pay for itself. There is an important warning here. Predicting the 2008 crisis does not make Bur automatically right today.
He is better understood as a contrarian investor who looks for risks the market may be too excited to see. So the right question is not will Bur be right again. The right question is this. Can we actually see the risks he is talking about inside these companies financial statements? To most people artificial intelligence looks like a simple chat box on a screen.
But behind that box is an extremely physical and extremely expensive world. AI requires advanced chips, enormous data centers, vast amounts of electricity, cooling systems, and networking equipment. That means the AI race is no longer just a competition to build better software. It has become one of the largest infrastructure races in modern history.
Companies are building for demand they expect years from now. If that demand arrived, these investments could become the digital highways of the future. But if demand slows, if new chips make old hardware less valuable, or if financing becomes more expensive, those same investments could become a crushing weight on corporate balance sheets.
Oracle and Nebius sit directly in the middle of that tension. Start with Oracle. At first glance, the company's fiscal 2026 results looked remarkably strong. Total revenue increased 17% to a record 67.4 $4 billion. Cloud revenue grew 39%. And in the final quarter alone, cloud infrastructure revenue jumped 93%.
Oracle also reported $638 billion in remaining performance obligations. In simple terms, that represents contracted business the company expects to recognize as revenue in the future. Oracle presents that number as evidence of massive future demand. But the other side of the story looks very different.
Oracle generated a record $32 billion in operating cash flow during fiscal 2026. And yet because of its enormous data center investments, free cash flow was negative 23.7 billion. During the same year, Oracle raised $43 billion in debt and $5 billion through equity financing. For fiscal 2027, it expects to raise roughly $40 billion more.
According to a Reuters analysis, Oracle's total debt reached 129.5 billion. Its debt to AITA ratio was roughly 4.3. At Alphabet, Amazon, Microsoft, and Meta, that ratio was below one. Oracle had also committed to roughly $260 billion in long-term data center leases. In July, S&P lowered Oracle's credit rating to tripleB minus, only one level above junk status, and Oracle's capital spending could reach as much as $95 billion in fiscal 2027.
Although the company expects customers to reimburse up to $25 billion of that amount, this is where Bur sees the danger. Oracle's debts and lease obligations are long-term. Some customer contracts are much shorter. If AI demand falls short of expectations, the data centers, debt payments, and lease bills will not simply disappear.
But Oracle has a powerful counterargument. The company has $638 billion in remaining performance obligations. Large AI customers have already prepaid for or directly supplied $75 billion worth of hardware. That reduces the amount of capital Oracle must provide on its own. Oracle is still profitable and growing quickly.
Most analysts tracked by LEG still rate the stock as a buy. So this is not a simple story about a failing company drowning in debt. It is a bet that tomorrow's demand will justify today's enormous commitments. Nebus is not nearly as famous as Oracle, but it may be an even clearer example of how quickly the AI infrastructure race can expand.
Based in Amsterdam, Nebus provides AI computing power to developers and businesses. In the first quarter of 2026, Nebius reported $399 million in total revenue. That was an increase of 684% from the same quarter a year earlier. Revenue from its AI cloud business alone grew 841%. The company also moved from an adjusted loss to $129.5 million in adjusted EID.
Its AI cloud division reported an adjusted Ebidom margin of 45%. Those are extraordinary growth numbers. But during that same 3month period, Nabas spent roughly $2.5 billion on capital investment. Its investment during one quarter was more than six times its revenue. Most of that money went toward GPUs, data center hardware, and expanding capacity.
Nebius also extended the estimated useful life of its server and networking equipment from 4 years to 5. That accounting change spreads the cost of the hardware over a longer period. The equipment may genuinely remain productive for longer, but the change goes directly to one of Bur's biggest concerns.
As new chips arrive at extraordinary speed, are companies assuming that today's hardware will remain economically valuable for too long? NBIS is using several sources to finance its expansion. In July 2026, it secured $775 million in debt financing backed by deployed GPU infrastructure and contracted customer cash flows.
The company's defense is that this debt is not based on empty promises. That distinction matters because the strength of the financing ultimately depends on the quality and durability of the customer contracts behind it. Nebias says it has more than $40 billion in additional contracted revenue from investment grade customers including Microsoft and Meta in its first secured financing agreement.
Customer cash flows and debt together cover more than the full cost of the related infrastructure. So, Nebius presents the same basic contradiction as Oracle. On one side, extraordinary growth, major customers and improving profitability. On the other, capital spending far above current revenue, a continuing need for financing, and technology that can become outdated very quickly.
Bur's AI bubble argument can be reduced to three major risks. The first is financing. Data centers are being built today. While much of the expected revenue may not arrive for years, debt, lease payments, and purchase commitments still have to be paid. Even if demand disappoints. The second risk is technological obsolescence.
As new chips offer more power at lower energy costs, today's GPS may lose economic value faster than companies expect. The third risk is demand and pricing. It is not enough for AI usage to grow. Companies must generate enough sustainable revenue to cover the infrastructure behind it. This is where it becomes important to separate a technology bubble from a failed technology.
Railroads change the world. The internet changed the world. But both revolutions also produced periods of excessive investment. A technology can be real and transformative while the price of every company associated with it can still be irrational. Bur is not necessarily arguing that artificial intelligence does not work.
His argument is that the market may already be pricing in a nearly perfect AI future. There are also serious reasons Bur could be wrong. If AI demand meets or exceeds today's expectations, the infrastructure being built now could become a powerful competitive advantage. New rivals cannot easily recreate data center capacity, energy connections, advanced hardware, and customer relationships.
Bur may also be early. Correctly identifying a bubble and correctly predicting when it will burst are two very different things. Markets can continue rising for much longer than a skeptic expects. In artificial intelligence, the demand is real, the revenue growth is real, and the customers are real.
This entire boom cannot be explained by hype alone. So, is Michael Bur right? Nothing here proves that the entire AI industry is about to collapse, but Oracle and Nebus show where the financial pressure could build. Oracle has record revenue and an enormous backlog of contracted business. It also has negative free cash flow, heavy debt, and long-term lease obligations.
Nebus is growing at an extraordinary rate and winning major customers. But that growth requires infrastructure investment many times larger than its current revenue. That may be the real weak point in the AI boom. The technology can work even if the business model becomes difficult to sustain. Today's spending depends on several things going right.
Demand has to keep rising. Hardware has to retain enough of its value. And financing has to remain available. Bur's argument is simple. If just one of those assumptions fails, the financial pressure could become severe. that Oracle and Nebus are betting that their contracts and growth will justify the spending.
In the end, impressive technology is only part of the story. The financial question is whether that technology can be turned into sustainable cash flow. So, what do you think? Are today's AI investments in another dot style bubble? or are we watching the construction of the world's next great infrastructure revolution?
Share your view in the comments. This video is for educational and documentary purposes only. It is not financial advice.
Comments 0