Who can still afford the AI accounts of the "Seven Sisters" in the US stock market?

Author: Chen Le ｜ Editor: Ye Jinyan ｜ Produced by: Deep Web · Tencent News Xiaoman Studio

After the U.S. midterm elections, Wall Street's optimism about AI plummeted within two days.

On November 5, the market value of the "Seven Sisters" of U.S. stocks evaporated by approximately $1.04 trillion in a single day, but the declines varied significantly: Apple fell slightly by 1.53%, while Tesla dropped sharply by 10.77%; as the biggest winner of the AI wave, Nvidia plummeted by 9.26%, losing over $440 billion in market value in one day, with declines 2 to 4 times that of its chip buyers (Microsoft, Google, Amazon).

The trigger for this crash was the U.S. midterm elections. After the Democratic Party won control of both houses of Congress on November 3, U.S. stocks initially rose the next day, but by the 5th, investors began to worry that the new Congress might tighten AI regulations and slow down data center construction, leading to a single-day drop of 5.52% in the Nasdaq. On the 6th, tech stocks rebounded, with the Nasdaq rising by 1.09%, Nvidia up by 2.78%, and AMD soaring about 11%, but Apple and Microsoft continued to decline. The Nasdaq still accumulated a drop of 3.4% for the entire week.

In fact, except for Tesla, the fundamentals of the giants have not deteriorated: Nvidia's revenue doubled year-on-year, Apple set a new high for the same period, and the cloud businesses of Microsoft, Google, and Amazon, as well as Meta's advertising business, continued to grow.

Economist Song Qinghui analyzed that policy uncertainty and market anxiety about AI investment returns have resonated. The valuations of tech giants have already heavily discounted AI's high growth, but the massive investments have yet to translate into equivalent profits. In the third quarter of this year, Microsoft, Google, Amazon, and Meta continued to increase investments in data centers and other infrastructure, with the growth rate of capital expenditure far exceeding operating cash flow.

Every hefty sum spent by the giants contributes to Nvidia's revenue. Once buyers "run out of money," how long can Nvidia's high growth be sustained?

Figure: Capital expenditure on infrastructure by tech giants and Nvidia's revenue trends

If data centers stop working, the "shovel sellers" will hurt the most

The answer to this question primarily depends on whether data centers can be built as planned. After the elections, in addition to concerns about whether AI investments can be recouped, the market has another layer of worry: will the regulatory policies of the new Congress and local governments slow down data center construction? Compared to cloud giants, Nvidia, which relies on downstream procurement, is more sensitive to this change.

Figure: Status of restrictive proposals on data center regulation in the U.S. and distribution of computing power planning

Nvidia's net profit in the most recent quarter approached $60 billion, and it estimates that revenue will increase by another 12% quarter-on-quarter next season. However, over 90% of its revenue is highly dependent on data centers. In its customer structure, cloud giants and AI startups, as well as government and enterprise clients, each account for half, with the latter growing faster. This model determines that Nvidia's performance is deeply tied to the pace of downstream infrastructure: if data centers are built quickly, shipments will increase; if progress is hindered, orders can only be delayed.

This election has introduced variables at this stage. On voting day, out of 41 local proposals related to data centers in 10 U.S. states, 40 had voting results leaning towards restricting data center development, with 27 restrictive or oppositional proposals being passed. The newly elected Democratic governor of Texas, Inohosa, even promised during the campaign to fully suspend data center construction until specific legislation is enacted.

According to semiconductor research firm SemiAnalysis, most of the data centers scheduled for delivery in 2027 have already begun construction, and the real impact of this election is on the planning for 2028, with Texas alone accounting for nearly 30% of the new computing power capacity in the U.S. in 2028. The policy changes brought about by the election may primarily affect Nvidia's growth expectations for 2028 and beyond.

If projects are delayed, the impact on both ends of the supply chain will not be the same: buyers can postpone some capital expenditures, while Nvidia may face the risk of delayed chip order deliveries. The market on November 4 has already shown this divergence: according to foreign media reports, on the day investors bet on the giants cutting AI spending, Microsoft and Amazon rose against the trend by about 6%, while Nvidia fell alone.

On the 5th, the sell-off spread throughout the entire AI supply chain. Amid Qualcomm's weak performance and market concerns about policies slowing down AI infrastructure, the Philadelphia Semiconductor Index plummeted by 12% in a single day, a decline far exceeding that of cloud computing companies.

But this does not mean that the demand for computing power has peaked. Song Qinghui reminds that programmatic trading, leveraged funds, and options hedging can significantly amplify short-term volatility, and a single-day crash does not represent a proportional loss of a company's intrinsic value. On the 6th, Nvidia rebounded by 2.78%, but it has not yet recovered the previous day's 9.26% drop. The upcoming quarterly financial report on November 17 will become an important window for the market to assess its growth prospects.

In contrast, Apple and Tesla are at opposite ends of this logical chain. Song Qinghui points out that the growth logic of the two companies is entirely different from that of AI infrastructure companies: Apple's profit base relies on phones and services, with relatively little dependence on data center construction, making it more resilient to declines; Tesla experienced the largest drop, also affected by Musk's political investments. Musk previously invested over $400 million to support the Republican Party, and after the Democratic victory, its business landscape may face more regulatory scrutiny. Within two days after the election, Tesla and SpaceX's combined market value evaporated by over $400 billion.

Who will pay the next hefty bill?

Whether data centers can be realized ultimately depends on funding. This massive capital is primarily borne by Nvidia's core buyers, Microsoft, Google, Amazon, and Meta. The latest financial reports show that the four companies are still increasing AI investments, but they can no longer rely solely on self-generated cash flow to support it.

Figure: Comparison of operating cash flow and capital expenditure gaps for Microsoft, Google, Amazon, and Meta

A year ago, the combined operating cash flow of these four companies exceeded fixed asset expenditures by about $61.7 billion; in the third quarter of this year, this surplus turned into a gap of about $26.4 billion.

This is not because AI is unprofitable; Google Cloud's operating profit in the third quarter was three times that of the same period last year, AWS's operating profit increased from $11.4 billion to $17.8 billion, and Microsoft's Azure revenue grew by 48%.

The gap arises from timing: the money for building data centers must be paid upfront, while revenue can only be recouped years later. For example, Google has $561 billion in cloud orders, over 55% of which will not be recognized as revenue for two years, but its signed future procurement commitments amount to $924 billion, 1.65 times the orders.

And the upfront payments are increasingly reliant on borrowing. Google has increased its long-term debt from about $20 billion to $125 billion within a year and has launched a stock issuance plan exceeding $80 billion; Amazon's free cash flow has turned from a net inflow of $14.8 billion over the past 12 months to a net outflow of $42.4 billion, mainly relying on bond issuance to make up the difference. Meta's free cash flow was negative for the first time in over a decade, and the trading level of its long-term bonds is now comparable to many BBB-rated companies.

Invesco believes this reflects investors' concerns about excessive spending and the uncertainty of AI monetization. Among the four, only Microsoft still has about $18.6 billion in free cash flow and has not issued large-scale bonds in recent years. According to Bank of America statistics, the five major cloud providers are expected to issue about $121 billion in dollar bonds by 2025, more than four times the average level of the previous five years.

As the scale of external financing expands, financing costs have also begun to become an important factor affecting the pace of data center expansion. Song Qinghui stated that persistent inflation and high interest rates will push up financing costs, and projects that rely on external financing and have long return cycles should appropriately slow down expansion. Currently, the yield on 10-year U.S. Treasury bonds remains above 4.8%. Approval is another hurdle. SemiAnalysis pointed out that while the giants can afford the costs brought by new regulations, this does not mean they can deliver computing power on time, and projects stalled by suspension orders are also difficult to quickly transfer to other states.

The most pressured are still the fastest-growing customers of Nvidia. Pan Helin, a member of the Information and Communication Economic Expert Committee of the Ministry of Industry and Information Technology, pointed out that unlisted large model companies like OpenAI and Anthropic are highly reliant on external financing. If interest rates remain high and regulations tighten, their subsequent financing difficulties will significantly increase.

This is precisely why Nvidia has fallen harder than its buyers. Nigel Green, CEO of deVere Group, believes that investors will next scrutinize each company to see who can truly convert AI investments into revenue.

On November 6, although Nvidia rebounded, it still did not recover most of the previous day's losses. Compared to Apple's relatively independent profit model from AI infrastructure, Nvidia's high growth relies more on continuous investments from downstream. How quickly buyers can build data centers will directly impact the next round of growth for this "shovel seller."