There was one good piece of news and two bad pieces of news in the risky asset market last week.
The good news is that the prices of data center bonds, which plummeted due to default risks, have stabilized, and credit default swaps ( CDS ) have seen a slight decline. The bad news is that stocks related to the AI concept have fallen, and semiconductor stocks have suffered heavy losses; at the same time, crude oil prices have risen sharply due to tensions in the Middle East, with Brent breaking through the 116-dollar mark, causing gold and silver to lose their luster. U.S. Treasury bonds have become a safe haven for funds, and the U.S. dollar exchange rate remains stable at high levels.
Over the past two months, the U.S. bond market has already voted with its feet against the overexpansion of the AI industry, and now it's the stock market's turn. The AI story has grown bigger and bigger, turning into a race for capital expenditure. Data centers, GPU, electricity, network equipment, cloud computing—anything related to AI is attracting investment. Brookings estimates that by 2032, the cumulative investment in U.S. AI infrastructure could exceed $10 trillion, which is equivalent to 3.6% of GDP each year. This scale is comparable to the total investment in several waves of technological development, including railways, highways, and early electrification projects.
In the past, the market only saw growth and looked forward to the future; now, however, people begin to calculate the costs and focus on what's right in front of them. The stock market can reap the benefits of future growth and is more willing to take risks; whereas the returns in the bond market are the fixed interest rates agreed upon at the time of borrowing, and the risks have to be borne by investors, which is why the bond market is the first to face these risks. In the past two months, the yields on bonds related to NeoCloud and data centers have significantly increased, the price of CDS has risen sharply, and some bonds have entered a stressful range. The market is concerned that the infrastructure related to AI may experience a collapse first in the credit market.
The author is not so pessimistic for now. Over the past two years, the financing window has been wide open, and many NeoCloud (new clouds of AI) have already raised enough funds for about a year ahead. The current increase in corporate bond yields affects the cost of future financings, not the cash in one's pocket today. Short-term liquidity may not immediately become an issue, but the real trouble will arise when old debts mature and need to be refinanced; at that time, the cost for companies to issue new bonds could be quite high. By then, funds will no longer be influenced by narratives such as strong demand from AI, but will instead consider future cash flows, the quality of contracts, project progress, and leverage ratios, which are indicators of repayment ability.
The situation is different for Hyperscalers (a super-large-scale cloud service provider). Alphabet, Amazon, Meta, etc., have exhausted their free cash flows and will also need to issue bonds in the future. This year, the capital expenditures of the five major tech companies have approached $800 billion (and next year are expected to exceed $1.5 trillion significantly), which is higher than their combined free cash flows. However, these companies have strong profitability and competitive core businesses, so even if financing costs rise, it remains controllable for now. As a result, the bond market has begun to distinguish between companies that have the ability to repay their debts and those that are still just making promises (i.e., "telling stories").
The reaction of the stock market has been slower than that of the bond market. Over the past period, the valuation of AI stocks was actually based on several very optimistic assumptions: there would be a long-term shortage of computing power, the greater the capital expenditure, the better; AI revenues would eventually catch up with investments; and all participants would ultimately receive a share of the profits. In the short term, this narrative still holds true. GPU The spot prices are still high, large technology companies continue to increase their capital expenditures, and AI hardware companies even have orders for the next two to three years. The demand is real, the orders are real, and so are the profits.
However, according to Gartner, by 2030, AI data centers must generate $8.5 trillion in revenue in order for the investments made by these companies to yield an acceptable return. What is even more concerning is that today, the companies in the AI industry that are reaping huge profits are not necessarily those that are investing heavily. The profit prospects of companies that are investing heavily in data centers are uncertain, and there are difficult-to-answer contradictions such as the contradiction between meager profits and exorbitant valuations, the contradiction between short-term investments and long-term profitability, and the contradiction between open-source and closed-source approaches.
Bond investors have been worried for a long time about whether they will be able to repay their debts next year, while stock investors are only now starting to worry about whether they will be able to make money in the future. This is also why the author believes that this market adjustment may not be too mild. However, an adjustment does not mean the end of the story. AI remains a great revolution and will still be the most important theme in the capital markets for the coming years. The profit models of large technology companies are still in place, AI hardware orders are still there, and the market's fear sentiment has not disappeared. After stock prices fall, I believe funds will return.
However, in the next round of AI market trends, there are three types of companies that pose a higher risk. The first type consists of companies that require substantial long-term investment, but whose future profitability is highly uncertain. Capital expenditures must be paid today, yet it will take many years to know whether they will generate profits. Once interest rates rise, their valuations are the first to be affected. The second type are those that plan to go public (IPO), but ultimately fail to do so and are also unable to raise enough funds in the private equity market. Over the past two years, they have had ample capital, and problems could have been covered by financing; however, once capital markets tighten, flaws in their business models will become exposed. The third type includes companies that already have high leverage and continue to borrow new debt to repay old debt.
The author still has a more positive view of the companies on both ends. On one end are the super-large technology companies with strong profitability and healthy balance sheets. They have the capability to continue spending heavily, and they also have mature businesses that can afford to invest in AI. On the other end are hardware companies whose orders have already been secured for the next two to three years. No matter which AI model ultimately wins, there will still be a demand for GPU, network equipment, power supplies, and data center equipment.
Every round of technological revolution has seen over-investment. This was true for railways, electrification, and the internet; AI is no exception. Many infrastructure projects were built during the most frenzied periods of capital activity, but although the technological revolutions ultimately succeeded, many early participants were lost along the way. The author believes that this adjustment may not necessarily be a bad thing. It forces capital to re-evaluate profitability and returns, encourages companies to reconsider their leverage, and allows the market to discern who is truly making money and who is just riding on the wave by telling stories.
The topic shifts to France. French government bonds have fallen out of their "comfort zone," and the risk of a debt crisis is beginning to emerge. The yield on French 10-year government bonds once rose to 5.08%, hitting a new high since 2002. The market usually considers France and Germany as core countries within the eurozone, but currently, the pricing of French bonds resembles that of a peripheral country with high debt, high deficits, and low political enforcement capabilities. Today, the cost of issuing French 10-year government bonds is actually higher than that of Italy, Spain, Portugal, Ireland, and Greece, which were once dubbed the "PIGS" countries.
Rising oil prices, inflationary pressures, and increasing yields on U.S. long-term bonds have certainly pushed up the financing costs for all governments. But on top of that, France is also facing an additional premium. What investors are concerned about is not whether Paris has the money to repay its debts today, but whether it will still have the political capacity to control its debt tomorrow.
In 2025, France's fiscal deficit was 5.1% of GDP, which represents a slight improvement from the previous year, but it is still significantly higher than the EU's 3% fiscal red line. What is more embarrassing is that the debt burden has not decreased despite the improvement in the deficit. France's public debt rose from 110% in 2023 to 113% in 2024, and further increased to 116% in 2025, with a debt balance of about 3.5 trillion euros. The European Commission predicts that this year France's debt GDP rate will reach 118%, and it will exceed 120% next year.
A temporary increase in deficit is not terrible; what's truly frightening is when debt continues to rise over the long term instead of decreasing. The French government's tax burden is already at the highest level in the world, making it difficult to solve the problem by raising taxes, as well as to cut pensions, welfare, local finances, and public sector spending. At the same time, refinancing to repay old debts with new ones incurs higher borrowing costs, and interest expenditures are continuously rising. France's government interest expenditure as a percentage of GDP will increase from 2.2% last year to 2.6% this year.
This creates a vicious cycle: high deficits → increased debt issuance → rising interest rates → increased interest payments → further deterioration of fiscal conditions. If economic growth were fast enough, this cycle could be broken. However, the European Commission expects France's economic growth to be only 0.8% in 2026 and just 1.1% in 2027. As nominal financing costs continue to approach 5%, and real economic growth is so weak, the debt dynamics naturally become even less favorable.
What truly unsettles the market is not fiscal arithmetic, but political chaos. France has also had high deficits in the past and has experienced street protests. In the past, when faced with social pressure, governments often used fiscal spending to gain compromises, but today this path is becoming increasingly difficult to follow. Issues such as schools, pensions, and living costs have sparked a new round of protests and riots, yet France has almost no room for fiscal maneuvering. If the government spends money to quell public grievances, the bond market will not agree; if it insists on cutting spending, the electorate will not accept that either. The bond market demands austerity, but political reality calls for expansion.
French politics is highly fragmented, and the government lacks a stable majority. The 2027 budget was supposed to be an opportunity to restore fiscal credibility, but it has instead become a stress test for the market to assess the government's ability to execute its policies. IMF points out that France's short-term financing situation is still under control, but a credible medium-term fiscal consolidation plan must be put in place. Of course, the French debt market is large, liquid, with an average debt maturity period that is longer, and the investor base is also healthier than that of Greece at the time; what the market fears is not a default by the country tomorrow, but rather that the debt begins to escape political control.
The most dangerous development in France at present is that risks are shifting from fiscal figures to politics, and then from politics back to the bond market. The higher the yield, the more the government needs to implement austerity; the more austerity there is, the greater the social resistance; the less stable society becomes, the harder it is for the government to push through reforms; and the harder the reforms are, the higher the risk premium that investors demand.
The government is at a loss in the face of high fiscal deficits, and the street riots that have occurred across France for several weeks in a row are essentially an expression of anger towards the economic reality. The government has managed to stop the violent protests through curfews, which may have a temporary effect, but the root of the problem has not disappeared. In my opinion, France has not yet reached the point of a debt crisis, but crises usually do not begin on the day when a country is unable to repay its debts. When a government loses its political capacity to make decisions and rectify problems, it is only a matter of time before a crisis occurs.
The focus of this week's market is on the interest rate decisions of three central banks: The Federal Reserve is likely to maintain the federal funds rate in the range of 3.75% to 4% unchanged, with attention on how Federal Reserve Chairman Powell will discuss a potential interest rate hike in December; the European Central Bank is also not expected to raise rates, which may imply a hike in December; the Bank of Japan is also unlikely to raise rates, which could similarly suggest a hike in December. In addition, attention should be paid to U.S. GDP, PCE inflation, and the Eurozone GDP.
Author's Note:The author is the President and Chief Economist of淡水泉 (Hong Kong). This article represents solely the author's personal views and does not represent the official stance or predictions of the institution to which they belong, nor is it intended as investment advice or solicitation.
The euro shared currency prevented capital flight and the devaluation of the franc. However, these protections came at a cost, with pressures manifesting in the gradual widening of interest rate spreads, the erosion of budgets by interest payments, and changes in prime ministers, rather than a sudden monetary crisis.
Most large tech stocks fell.
What are the potential risks?
The real watershed lies in governance, responsibility, and human judgment.
The five-year credit default swap spread widened by 135 basis points to a record high of 1,117 basis points, with market pricing implying a default probability of about 63% within five years.
