What is capital reassessing after two sharp declines in the US stock market

Economic Observer Follow 2026-11-22 09:24


Ah, is it another US stock market crash

On November 20th, the major stock indexes of the US stock market plummeted again. The investors' exclamation is like the echo of 15 days ago.

On November 5th, the Nasdaq Composite Index plummeted by 5.52%, but the market rebounded strongly a few days later. Will the rebound plot be replayed next week?

This time, the market seems more complex. As of November 20th, Brent crude oil has risen by nearly 50% this year, while spot gold has fallen by over 18%. Assets that were once sought after are now facing a price rearrangement.

A record breaking IPO (initial public offering) seen as a "barometer of sentiment in the AI (artificial intelligence) sector" did not "save" AI stocks.

On November 19th, Anthropic, an American artificial intelligence research company that has only been established for five years, landed on NASDAQ and received a huge amount of new capital. The next day, Nvidia's stock price plummeted by about 14%, the Philadelphia Semiconductor Index plummeted by 10.7%, and global technology stocks suffered another heavy blow. What's even more intriguing is that three days ago, Nvidia just submitted a financial report showing that its revenue doubled year-on-year.

The good news is not absent. However, looking back at the beginning of the year, I feel lost - the market is no longer willing to pay for these messages at the previous prices.

At that time, global funds were still searching for a home on both ends: one end was investing in AI to purchase growth; On the other end, embrace gold and US bonds to hedge risks. Nowadays, AI and gold are under pressure, and some funds have returned to US Treasury bonds.

Funds have not disappeared, but it seems to be recalculating the price of time: how much future does technology have? How much does it cost to reach that future, and how long do we have to wait?

Why did it fall again after rebounding

Looking back at November 5th, the Nasdaq Composite Index plummeted 5.52%, the S&P 500 plummeted 4.36%, and the Philadelphia Semiconductor Index fell over 12%. After the midterm elections in the United States, concerns have arisen regarding the approval of data center construction, electricity costs, and regulatory prospects; The weak smartphone chip business of Qualcomm has also increased the selling pressure on the semiconductor sector.

However, the market did not continue to decline. On November 6th, the US stock market initially stabilized and experienced a strong rebound on November 9th: the S&P 500 rose 2.35%, the Nasdaq Composite Index rose 2.74%, and the Philadelphia Semiconductor Index rose 6.9%. The decline in oil prices has eased some inflation concerns, and buying at low prices has driven upward expansion. However, Nvidia still fell on that day, and the rebound did not solely rely on AI leaders.

15 days later, the second shock arrived. On November 20th, the Nasdaq Composite Index plummeted 5.28% and the S&P 500 plummeted 3.70%; The Philadelphia Semiconductor Index has plummeted 17.1% in a week. Asian technology stocks were also under pressure on the same day.

The magnitude of the two sharp drops is similar, but the triggers are not exactly the same.

This time, more pressure falls on the company itself. The management changes brought about by Huang Renxun's departure have not been fully digested by the market, and the media has further disclosed the investigation by the US Securities and Exchange Commission into some accounting treatment issues of Nvidia. At the same time, the highly anticipated Anthropic approached its IPO price on the second day, failing to boost confidence in the AI sector.

From policy constraints on data center construction to corporate governance, compliance, and IPO valuation risks, different news has successively impacted the market. Investors' doubts about AI profit realization and capital return have become more prominent.

The change in volatility records this process. On November 5th, the US stock market volatility index (VIX) closed at 28.18; On November 9th, it dropped to 19.92; On November 20th, it rose to 35.89. The MOVE index, which measures the expected volatility of US Treasury bonds, also rose from around 86 on November 9th to around 106 on November 20th. The uncertainty in both the stock and bond markets is increasing.

The performance of gold is particularly abnormal. At the beginning of this year, the international gold price was close to $5600 per ounce, but on November 20th it fell to around $3520, with a pullback of over 30% from its high point. The US dollar, real interest rates, and investors' demand to reduce leverage may all create pressure.

But safe haven funds have not disappeared. On that day, US Treasury bonds rose, and the 10-year US Treasury yield fell by about 10 basis points to around 4.95%. Gold and stocks are falling synchronously, while US bonds are receiving buying interest, and funds are choosing safe haven tools again. November 20th coincides with the expiration of monthly options, and trading mechanisms such as market makers hedging and leveraged funds reducing positions may amplify the decline. The actual impact still needs to be tested by position and transaction data.

Market downturns are sometimes just a matter of a few trading days, and the issue of capital returns cannot be answered within a few days. The most indicative of this change is a listing that should have boosted market confidence.

Two bills

On November 19th, Anthropic closed at $97.17 on its first day, hitting a low of $83.05 the next day, only 5 cents away from breaking, and ultimately closing at $88.91 (-8.50%).

Indeed, the fact that the stock price has almost fallen back to the issue price does not necessarily mean a financing failure. Anthropic's announcement shows that the total financing from this IPO and synchronized private placement is approximately $119.5 billion.

What's even more worth looking at is another bill. In the final prospectus, Anthropic disclosed a total of approximately $900.9 billion in committed expenses for purchases, leases, and leases that have not yet been fulfilled, mainly related to computing power and infrastructure, with some arrangements extended until 2045. Approximately 451.6 billion US dollars are expected to occur before the end of 2030; Based on the end of September 2026, the expected payment obligations due in the next 12 months are approximately $70.7 billion.

119.5 billion US dollars is the total amount of Anthropic's financing transaction this time; 900.9 billion US dollars are contract commitments distributed over the future years, not debts already borrowed; Approximately 70.7 billion US dollars is a short-term payment arrangement calculated at a specific point in time. The three cannot be simply subtracted, let alone declared that Anthropic has a funding gap based on this. The prospectus also disclosed that Anthropic had approximately $146.6 billion in liquidity resources prior to its listing.

The problem is not whether the company can pay its bills today, but whether its future revenue growth can continue to cover the costs of computing power procurement, model development, and financing.

Anthropic's third quarter revenue was approximately $17 billion, with computing and infrastructure expenses of approximately $9.9 billion. The income growth is rapid, but it is still not enough to prove that the investment return for the next decade has been locked in. Even if the usage of the model continues to increase, if the price of inference services decreases and training costs increase, the revenue growth may not necessarily translate into profits synchronously. The efficiency of computing power utilization, model service prices, actual customer payments, and contract fulfillment pace may all change this account.

This listing also presents another side of the internal capital relationships within the AI industry. Google, Nvidia AMD、 Broadcom, Microsoft, Micron, TSMC and other related entities participated in Anthropic's synchronous private equity financing subscription of approximately $26 billion.

Investors may ask: How much of the company's growth comes from real payments from end customers, and how much is built on mutual investment, long-term procurement, and credit support from industry partners? To answer this question, it is necessary to verify the contract and the actual payment of the end customer.

In contrast to Anthropic, Nvidia stands out. On November 17th, Nvidia announced quarterly revenue of $115.8 billion, a year-on-year increase of 103%, with data center revenue reaching $107.6 billion. This achievement did not prevent the stock price from plummeting three days later. Huang Renxun's recent departure has brought new corporate governance variables, but it cannot conceal more common market issues.

One company has already earned huge profits, while the other has just obtained huge financing. However, investors are starting to ask: how much capital is needed to maintain the next round of growth?

Analysts say that raising money does not necessarily mean making money, and the current strong income does not automatically guarantee tomorrow's valuation.

Entering the perspective of the central bank

At present, this question is entering the vision of the central bank from corporate financial reports.

Both the Federal Reserve and the European Central Bank have recently focused on the debt, long-term procurement contracts, leasing, and guarantee relationships behind the expansion of the AI industry. The Federal Reserve's research mentions potential off balance sheet obligations exceeding $3 trillion, while the European Central Bank's estimates for different sample companies exceed $4 trillion.

Of course, the two statistical criteria are different and should not be added together, let alone equal the debt losses that have already occurred. However, they point to a problem - cloud service providers support data center construction with long-term procurement contracts, and project companies raise funds from banks, bond investors, and private credit institutions based on this; When the project is delayed and the computing power income is lower than expected, payment and debt pressure may be transmitted along the contractual relationship.

For large platforms, long-term contracts can lock in computing power; For project companies borrowing to build computer rooms, the same contract may be the basis for financing. When the project is delayed, income may be moved back, but interest still needs to be paid.

What's even more tricky is that the three times are not consistent. In terms of technical time, chips and models iterate rapidly, and advanced equipment may require frequent updates. In terms of construction time, power access, computer rooms, cooling facilities, and approval often take several years. In terms of capital time, equipment payments, interest, and debt maturity dates will not be automatically postponed due to project delays.

Federal Reserve Governor Lisa Cook recently pointed out that American companies have announced over $2 trillion in specific data center construction plans, but only a portion of them have actually been implemented. Federal Reserve Vice Chairman Philip Jefferson warned that AI investment may first drive up equipment, energy, and funding demand, but the widespread productivity benefits will need to be adopted and organized by businesses before they can be realized later.

This is reminiscent of economist Carlotta Perez's analysis in "Technological Revolution and Financial Capital": Financial capital often discovers, amplifies, and trades the future value of new technologies faster than the production system. It can support infrastructure construction and may also allow valuation and financing scale to run ahead of actual returns. Subsequently, capital and industry need to be re coordinated.

However, historical theory is not the "basis" for the market to peak. It is worth noting that Perez is not describing the disappearance of technology, but rather the initial optimistic pricing of financial capital for the future, which needs to gradually accept the reality test of production, demand, and institutions. Perez also pointed out that AI is better understood as a further deepening of the information and communication technology revolution. The two sharp declines in the US stock market in November are not enough to prove that AI capital expenditures have irreversibly peaked.

The credit market has already adjusted ahead of time. According to data from Bank of America, from October 8th to 29th, the credit spread of AAA rated commercial real estate mortgage-backed securities (CMBS) in data centers increased from 175 basis points to 195 basis points, while the credit spread of BBB rated CMBS increased from 400 basis points to 435 basis points. Capital has not completely withdrawn from AI infrastructure, but has begun to demand higher risk compensation for different credit ratings.

Brian Levitt, Chief Global Market Strategist at Jingshun, proposed another possibility: a slowdown in data center construction may not necessarily mean the end of the AI cycle, but may also alleviate some funding and interest rate pressures, extending the investment cycle.

The boundary between the two scenarios is cash flow. If the real demand and operating income continue to grow, the investment can still continue after valuation adjustment; If the tightening of financing leads to project delays, reduced equipment orders, and increased debt pressure, the adjustment of the financial market may in turn slow down industrial expansion. At this point, what really needs to be monitored are credit spreads, loan extensions, capital expenditure guidance, and project production progress, rather than just focusing on the Nasdaq Composite Index or Nvidia stock price.

Technology is still moving forward, but capital is starting to recalculate the cost of reaching the future. Financial capital can purchase the future in advance, but cannot cash everything in advance for the future.


The chief reporter of the Economic Observer has long focused on macroeconomic, financial and monetary markets, insurance asset management, wealth management, and other fields. More than ten years of experience in financial media industry.