Holding huge AI infrastructure-related orders is no longer enough to "protect" a company.
Oracle, with $500 billion in orders, has seen its stock price plummet 40% from its September peak. Broadcom currently has approximately $73 billion in backlogged AI product orders, and its stock price reversed course and fell after its latest earnings report.
CoreWave, often referred to as "Nvidia's favorite son," generates over a billion dollars in quarterly revenue, yet it secured orders exceeding $36 billion from OpenAI and Meta within a single week. Over the past month, the company's stock price has cumulatively fallen by 17%.
While outsiders are concerned about whether they have enough financial resources to satisfy customers, they are also concerned about whether the customers themselves are truly "reliable".
When you peel back the onion of AI infrastructure, you end up with just a few giants: Meta, Google's parent company Alphabet, Microsoft, Amazon, Apple, Nvidia, and others, plus star AI startups like OpenAI and Anthropic.
These newly established companies are still very immature, and their infrastructure projects almost entirely rely on external financing, which carries significant risks.
These giants should be the pillars of stability—they are financially sound, have plenty of cash, and are filling the next few years with a crazy infrastructure plan worth hundreds of billions of dollars.
However, the returns brought to them by AI, which is at the top of the spending list, are still small. Whether using their "old capital" to nurture new dreams will drag down the giants depends entirely on whether the dreams are realized in a timely manner.
Success brings joy to all, while failure could mean total defeat.
01
Holding the "future" card, Oracle experienced both great joy and great sorrow in just a few months.
When the good news arrived, Oracle's stock price soared 40% in a single day, and founder and CEO Larry Ellison briefly surpassed Musk to become the world's richest person.
At that time, Ellison shouted, "Artificial intelligence is everything!"
Artificial intelligence is indeed everything, and for Oracle, it was the reason for its current surge in popularity—OpenAI reached a five-year, $300 billion computing power purchase agreement with Oracle, which became the spark that ignited Oracle's stock price.
However, just three months later, Oracle had more orders in hand, but the "magic" disappeared.
Oracle recently released its financial results for the second fiscal quarter of 2026 (corresponding to September to November 2025), with revenue increasing by 14% year-over-year. The company stated that its order backlog has reached $523 billion.
This figure represents an increase of $68 billion compared to the previous fiscal quarter.
Following the release of its earnings report, Oracle's stock price plummeted 11% that day, marking the company's largest single-day drop since January. Looking back from its peak in September, Oracle's stock price has fallen by 40%.
Future orders, amidst the current skepticism surrounding the "AI bubble," have transformed from a hopeful prospect into a heavy burden.
Oracle appears to be struggling – its financial report shows that Oracle has a negative cash flow of $10 billion and quarterly capital expenditures (CapEx) of $12 billion, nearly $3.7 billion higher than analysts' forecasts.
Oracle's CFO revealed that the company's annual expenses have also increased by as much as $15 billion, reaching the level of $50 billion.
The market's biggest fear is: Does Oracle really have the ability to raise so much money to support such a massive AI infrastructure project?
Analysts predict Oracle will need to borrow $100 billion to complete construction. In its second fiscal quarter, the company raised $18 billion in debt, one of the largest debt issuances ever recorded by a technology company.
During the conference call, Oracle vehemently defended itself, explicitly opposing the prediction that it would need to borrow $100 billion, stating that the actual amount raised would be significantly less. The key lies in Oracle's adoption of a "customer bring their own chips" approach.
In other words, instead of Oracle buying chips and then leasing them to customers, customers bring their own chips, which is unprecedented in the cloud services industry.
In addition, Oracle also emphasized that some suppliers are willing to lease rather than sell chips to Oracle, so Oracle can make and receive payments simultaneously.
If what Oracle claims is true, then it can indeed significantly reduce its upfront investment and greatly increase its rate of return.
However, the risk hasn't disappeared for the market; it has shifted: from Oracle to its customers. Customers like Meta and OpenAI are purchasing expensive GPUs and installing them in Oracle's data centers.
Whether Oracle can realize its multi-billion dollar future depends not only on its ability to "deliver" its orders, but also on its customers' ability to "pay." Of Oracle's nearly $500 billion in undelivered orders, about two-thirds come from OpenAI, which is not yet profitable, and another $20 billion is known to come from a new agreement with Meta.
Broadcom is another example of a company that, despite holding a huge number of orders, has received negative feedback from the market.
Broadcom also released its new financial report, achieving better-than-expected core revenue and profit in the fourth quarter of fiscal year 2025, ending November 2, with AI semiconductor-related revenue increasing by 74% year-on-year.
During the conference call, Broadcom CEO Hock Tan stated that the company currently has a backlog of approximately $73 billion in AI product orders, which will be fulfilled within the next six quarters. He emphasized that this is the "minimum," and the backlog is expected to expand further as new orders continue to arrive.
However, Broadcom declined to provide clear guidance on full-year AI revenue for 2026, citing uncertainty surrounding customer deployment pace and potential quarterly fluctuations.
Following the release of the earnings report, Broadcom's stock price initially rose by about 3%, but subsequently fell, with a drop of over 4% in after-hours trading.
Compared to Oracle's dramatic ups and downs, Broadcom only experienced a minor setback, but the underlying market sentiment was similar—people were no longer optimistic about the "future" of massive AI infrastructure development.
Broadcom's customer base is also relatively concentrated, with its AI-related orders mainly coming from OpenAI, Anthropic, Google's parent company Alphabet, and Meta.
02
When you peel back the layers of the AI infrastructure onion, you'll eventually find a few familiar companies—the seven US stock market giants, OpenAI, and Anthropic.
CoreWave, another AI cloud infrastructure startup that has garnered significant attention this year, went public in March. It was the largest tech startup IPO since 2021, and its stock price has since more than doubled, even surpassing the "Big Seven" tech giants.
Its customer base is also extremely concentrated, basically surviving on orders from Microsoft, OpenAI, Nvidia, and Meta.
On Monday (December 9), CoreWave issued another $2 billion in convertible bonds, bringing its total debt to $14 billion as of the end of September. Market concerns have intensified, and its stock price has fallen 17% in the past month.
To reiterate, the market has developed deep doubts about the AI industry as a whole, not only about whether AI infrastructure-related companies can provide services as planned, but also about whether large clients who are making huge transactions can actually deliver on their bills.
The complex cyclical transactions among all relevant parties have formed a tight and opaque web, making everything even more difficult to discern.
If we look at it by the type of customer, startups like OpenAI and Anthropic were the first to raise concerns.
The reason is simple: neither of them has a stable ability to generate revenue, which is far from enough for their expanding infrastructure plans. They need to rely on external financing, and the uncertainty is obvious.
Giants, on the other hand, are more like the weathervane and the safety net in the game.
These tech giants spend hundreds of billions of dollars in capital expenditures annually, a significant portion of which goes towards expanding their data centers. Their combined capital expenditures in 2026 will be more than four times the total expenditures of the U.S. publicly traded energy sector on drilling exploration wells, extracting oil and gas, transporting gasoline to gas stations, and operating large chemical plants. Amazon alone has already exceeded the total capital expenditures of the entire U.S. energy industry.
Compared to fledgling startups, the giants are clearly well-funded, with sound finances and ample cash flow. At least for now, their spending hasn't exceeded their capacity.
For example, Microsoft, Google, and Amazon will spend more than $600 billion from 2023 to this year, with revenue expected to reach $750 billion.
If you look at their recent performance reports, you'll find that their performance is quite strong. "Exceeding expectations" is already standard practice, so there seems to be nothing to worry about. In other words, they can afford to invest heavily in AI infrastructure.
However, upon closer inspection, none of them have fundamentally changed their revenue structure. While AI has begun to generate returns, it often still plays a supporting role in overall revenue, yet it occupies the top spot when it comes to spending.
For example, at the end of July, TheCUBE Research estimated in its quarterly financial report that AI services contributed about 19% to the growth of Azure cloud, exceeding $3 billion, but this contributed less than one-tenth of Microsoft's total revenue.
More than half of Google's revenue still comes from advertising and search, while Amazon's e-commerce and advertising still account for more than 70% of its revenue.
In other words, tech giants are using their mature businesses to nurture the future of AI.
The question is, how long can this nourishment last?
03
The giants have begun to unleash a “borrowing frenzy”.
In September, Meta issued $30 billion in bonds. Alphabet also recently announced plans to issue approximately $17.5 billion in bonds in the US market and approximately $3.5 billion in bonds in the European market.
Data from Bank of America shows that in September and October alone, large technology companies focused on artificial intelligence issued $75 billion in U.S. investment-grade bonds, more than double the industry's average annual issuance of $32 billion between 2015 and 2024.
These companies' revenue growth should currently be able to support their spending, but to keep pace with the development of artificial intelligence, they will eventually need more debt.
A Wall Street Journal analysis pointed out incisively that AI is weakening giants.
As of the end of the third quarter of this year, Microsoft's cash and short-term investments accounted for approximately 16% of its total assets, down from approximately 43% in 2020. Alphabet and Amazon have also seen significant reductions in their cash reserves.
Alphabet and Amazon are expected to have lower free cash flow this year than last year. While Microsoft's free cash flow appears to have increased year-over-year in the last four quarters, its disclosed capital expenditures do not include long-term lease payments for data centers and computing equipment. If these expenditures were included, its free cash flow would also decline.
This trend seems destined to continue.
Analysts estimate that Microsoft will spend approximately $159 billion next year, including leasing expenses; Amazon is expected to spend approximately $145 billion; and Alphabet is expected to invest $112 billion. If these predictions come true, these companies will have invested a total of $1 trillion over four years, most of which will be in the field of artificial intelligence.
Taken together, these changes—decreasing cash balances, reduced cash flow, and increased debt—are fundamentally altering the business models of technology companies.
The technology industry is becoming more and more like industries such as semiconductor manufacturing, where hundreds of billions of dollars are invested in building cutting-edge factories that take years to build but even longer to reap the rewards.
Deploying hundreds of billions of dollars across hundreds of massive data centers, AI infrastructure already presents significant challenges from an execution perspective alone.
Data centers consume enormous amounts of power—GPUs require massive amounts of electricity for computation—and the current power grid cannot handle the surge in demand. Secondly, cooling is also a problem. GPUs operate at very high temperatures and require large amounts of fresh water to keep the equipment running. Some members of the community have begun to oppose the construction of data centers, fearing they will affect water supplies.
This year, Nvidia and OpenAI jointly announced a new agreement worth up to $100 billion, with OpenAI planning to deploy 10 gigawatts of Nvidia systems. However, Nvidia's CFO recently admitted that the plan is still in the letter of intent stage and has not yet been formally signed.
This casts a shadow over the "credibility" of the bustling AI infrastructure deals, and also suggests future uncertainties.
The reasons for the delay in signing the agreement have not yet been disclosed, but the "risk factors" section in Nvidia's filings with the SEC can be used as a reference.
In the document, Nvidia warned that if customers reduce demand, delay financing, or change direction, the company may face risks such as "excess inventory," "penalties for order cancellations," or "records for inventory write-downs and impairments."
Furthermore, the availability of "data center capacity, power, and capital" is crucial for the deployment of AI systems. The document states that the construction of power infrastructure is a "several-year process" and will face "regulatory, technical, and construction challenges."
Even if AI infrastructure progresses smoothly in the end, it is not the end of "success".
AI infrastructure is ultimately meant to serve AI needs. If infrastructure has been built but market demand has not been met, then underutilization of the infrastructure will result in huge losses.
Of course, not everyone is frowning and expressing concern. Supporters believe it's a gamble worth taking because the demand for AI will grow exponentially, not linearly.
Analyst Azeem Azhar calculated that direct revenue from artificial intelligence services has increased nearly ninefold in the past two years.
In other words, if this growth rate continues, it's only a matter of time before AI companies start generating record profits.
"I think people who are fixated on the specific financing methods for these investments are outdated. Everyone assumes that this technology will develop at a linear pace. But artificial intelligence is an exponentially growing technology. It's a completely different model," Hazard said.
The question is whether and when AI will start to generate explosive "profits".
Ultimately, whether AI infrastructure will drag down tech giants is a matter of the AI market demand chasing after AI infrastructure. If it catches up, AI infrastructure will be incredibly worthwhile; if it doesn't, massive data centers will eventually become ghost towns. That would be the best proof that tech giants have made the wrong bet on AI, and it would have disastrous consequences.
