① Last Friday, US AI technology stocks plummeted, with Nvidia among them. Oracle bone script Hebotong As tech stocks suffered setbacks, investor concerns about an AI bubble and the prospects for computing infrastructure resurfaced; ② Broadcom's stock price plummeted 11.43% after releasing strong earnings, fueling market concerns about declining gross margins and anxieties about an AI bubble.
Last Friday, US AI technology stocks experienced a bloodbath: tech stocks led by Nvidia , Oracle , and Broadcom fell sharply, reigniting investors' concerns about an AI bubble and the prospects of computing infrastructure.
In fact, during the recent earnings season, Nvidia , Oracle , and Broadcom all delivered strong financial results, and company executives repeatedly emphasized the promising prospects for company growth during earnings calls.
However, unlike the initial frenzy of investors going "all in" on AI, this time, investors are not focusing on the dizzyingly high growth rates, but rather on expenditure costs, gross margins, and risk factors.
This also indicates that the trading logic surrounding AI is gradually shifting from "grand narratives" to "real-world returns." Some investors worry that the bursting of the AI bubble seems to be imminent.
Broadcom's impressive performance has been met with a harsh blow from the market.
Broadcom serves as a prime example of this shift. Despite releasing a largely better-than-expected earnings report last week, its stock price plummeted 11.43% on Friday. This was the stock's largest single-day drop since its 17.4% plunge in January.
The financial report shows that in artificial intelligence Data Center Broadcom achieved strong revenue growth driven by a surge in sales of custom processors and related network chips. In a conference call with analysts, Broadcom CEO Hock Tan repeatedly emphasized that the current $73 billion order backlog does not represent all revenue delivered over the next 18 months, and that the company's order volume could still accelerate in the future.
A year ago, such financial reports and results would likely have been met with enthusiastic investor enthusiasm. But now, the market's reaction has been a massive sell-off.
The market's reaction has baffled even some Wall Street analysts. Bernstein analyst Stacey Rasgon bluntly stated that Broadcom's stock price decline after releasing its fourth-quarter earnings report was perplexing.
"Broadcom's share price fell, and the anxiety surrounding AI- related stocks seems to persist," Rasgon said in the report. "Frankly, we're not sure what more the market could ask for, given that the company's performance in the AI field has not only exceeded expectations but is also accelerating."
Some analysts believe that market concerns may stem from Broadcom 's declining gross margin. Due to increased costs from previous production line expansion, Broadcom's profit margins are expected to be severely impacted in the short term.
Broadcom CFO Kirsten Spears acknowledged in an earnings call that the "gross margin will be lower" for some of Broadcom's AI chip systems because the company needs to purchase more components to produce server racks.
"This stock (Broadcom) has risen 75-80% year-to-date. A pullback now is normal," said Mizuho Securities. "We will buy on pullbacks," analyst Vijay Rakesh said on Friday.
However, some analysts believe that Broadcom's stock price decline may be more due to the market's overall anxiety about an AI bubble.
Oracle's "AI ambitions" are being questioned.
Just one day before Broadcom's plunge, US tech giant Oracle also experienced a nearly 11% drop, which is enough to reflect the widespread anxiety in the AI market.
In its latest earnings report released last week, Oracle's remaining performance obligations (RPO, i.e., unconfirmed contract revenue) in the second quarter increased by 438% year-on-year to $523 billion, far exceeding the consensus expectation of $502 billion from FactSet analysts, and further increasing by 15% on top of the huge RPO in the previous quarter.
This means that in order to meet strong market demand, the company may need to raise further debt financing to invest in its project construction.
However, the market is increasingly skeptical about whether the business model of investing heavily in building AI data centers equipped with expensive GPUs and then renting them out is truly attractive. If a rough and simple calculation is performed, the answer seems to be no.
According to rough estimates from analysts cited by foreign media, even with optimistic forecasts, the return on investment for Oracle's many large projects will be far below 10% in the future, and the return will be even lower if the gross profit margin has not yet reached the target range.
Considering the cash flow risks Oracle would have to take in this process, such a rate of return does not seem satisfactory.
Furthermore, in a presentation in October, Oracle predicted that its adjusted gross margin for its AI infrastructure business would be between 30% and 40% in the future. The company also cited an AI deal as an example, stating that it would generate $60 billion in revenue and $21 billion in gross profit over six years.
However, judging from the financial reports alone, Oracle seems to have a long way to go to achieve this goal. The reports show that Oracle's AI infrastructure business had a gross margin of only 14% in the quarter ending in August. After deducting operating expenses, the operating profit would be even lower.
Another notable announcement during Oracle's earnings call was its updated capital expenditure forecast. Oracle raised its full-year forecast by $15 billion and claimed that these expenditures are expected to generate $4 billion in incremental revenue in fiscal year 2027.
Putting these two figures together seems highly illogical: Oracle is investing an additional $15 billion this year, aiming for $4 billion in revenue next year. At current profit margin levels, it remains to be seen when this investment will recoup its costs.
Amid market anxieties, a short article about Oracle has further fueled panic : according to sources, Oracle has postponed the completion date of some of its data centers prepared for OpenAI from 2027 to 2028.
The aforementioned reports exacerbated the sell-off of AI concept stocks, affecting not only the technology sector but also power and energy-related stocks in the US market.
Despite Oracle's subsequent rebuttal, its stock price still fell more than 4% on Friday. An Oracle spokesperson clarified, "The site selection and delivery timelines were determined jointly and in close consultation with OpenAI after the agreement was signed. Currently, there are no delays in meeting the contractual commitments for any of the sites, and all milestones are progressing as planned."
However, the market's main concern is clearly not just Oracle's project construction progress, but also the high cost pressure brought about by Oracle's current large-scale expansion.
Is the "largest AI infrastructure project in history" also in danger?
Nvidia faces a similar predicament to Oracle. In September of this year, Nvidia and OpenAI jointly announced a remarkable plan—OpenAI will use Nvidia's systems to build and deploy at least 10 gigawatts (GW) of AI data centers, using millions of Nvidia's graphics processing units (GPUs) to train and deploy OpenAI's next-generation AI models.
Nvidia CEO Jensen Huang claimed this would be "the largest AI infrastructure project in history," and analysts estimated the deal could bring Nvidia up to $500 billion in revenue. This news initially spurred investors to significantly increase their holdings, driving up Nvidia's stock price.
However, two months later, Nvidia's executive vice president and chief financial officer, Colette Kress, admitted to investors that the much-hyped collaboration was still in the letter of intent stage and no final agreement had been reached.
When asked about the actual amount of money locked in the 10 gigawatt commitment, Kress said, "We have not yet reached a final agreement."
This news was undoubtedly a bucket of cold water poured on investors.
It remains unclear why the deal has not yet been completed, but Nvidia's latest quarterly report provides clues. The document explicitly states: "There is no guarantee that any investment will be completed on the terms expected, and it is even possible that it will not be completed at all." This involves not only the collaboration with OpenAI, but also Nvidia's planned $10 billion investment in Anthropic and its investment in Intel. A plan to invest $5 billion.
In a lengthy “Risk Factors” section, Nvidia details the fragile architecture that underpins such massive transactions.
The company emphasized that the reliability of this situation depends solely on the global capacity to build and operate the data centers required for its systems. Nvidia must place orders for GPUs and high-bandwidth memory more than a year in advance. Network equipment and other components are typically sold under non-cancellable, prepaid contracts. Nvidia warns that if customers scale back, delay financing, or change course, it may face "excess inventory," "cancellation penalties," or "inventory provisions or impairments."
The biggest influencing factor appears to lie in the physical world: according to Nvidia, the supply of “data center capacity, energy, and funding” is critical for customers deploying the AI systems they are promised. Power infrastructure construction is described as a “years-long process” and faces “regulatory, technological, and construction challenges.”
Nvidia warned that if customers cannot obtain sufficient power or financing, it could "delay customer deployment plans or reduce the adoption of AI applications."
Nvidia has also acknowledged that its own pace of innovation makes planning more difficult. The company has shifted to releasing a new architecture annually while still supporting previous product lines. The company notes that the faster pace of architecture updates "may increase the difficulty of predicting demand" and could potentially lead to "reduced demand for current product lines."
These statements echo warnings from “AI pessimists” like Wall Street’s “big short” Michael Burry, who has claimed that Nvidia and other chipmakers have excessively extended the lifespan of their chips, and that the eventual devaluation of these chips will disrupt the investment cycle.
When will the huge investment yield returns?
In fact, the decline in tech stocks such as Oracle, Broadcom and Nvidia last Friday was just a microcosm of the market's anxiety about the prospects of large-scale infrastructure investment in the AI industry.
Since the advent of ChatGPT, Nasdaq The S&P 100 has more than doubled, which means that the US stock market's tolerance for massive investments in AI infrastructure is nearing its limit. As major tech giants frantically purchase AI hardware infrastructure such as AI GPUs and AI ASICs, investors are urgently asking: when will we see returns?
Morgan Stanley Analysts estimate that by 2028, large tech companies will have invested approximately $3 trillion in artificial intelligence infrastructure, while their own cash flow will only cover half of that amount.
To avoid burning through too much cash, giants like Meta and Oracle are using private equity and debt to fund their data center construction.
Goldman Sachs An analyst assessment found that hyperscale data center companies (those with massive cloud computing power) Technology companies with computing power have taken on $121 billion in debt in the past year, more than 300% higher than the typical debt burden in the industry.
Silicon Valley has taken on such a huge amount of new debt because it believes that the enormous new revenue generated by artificial intelligence in the future will be enough to pay off this debt.
But is that really the case? At least for now, investors are increasingly skeptical.
(Article source: CLS)
