Selloff in long-dated government bonds continues! 10-year U.S. Treasury yield breaks above 5.2% again, "AI agent explosion VS rising financing costs" narrative confrontation escalates

Title context: Selloff in long-dated government bonds continues! 10-year U.S. Treasury yield breaks above 5.2% again, "AI agent explosion VS rising financing costs" narrative confrontation escalates Text: The investment wave in AI computing infrastructure is facing an increasingly sharp set of contradictions: the strong computing power expectations brought by the large-scale expansion of AI agent applications have reinforced the necessity for enterprises to lock in computing resources on a large scale in advance, while energy inflation and the monetary tightening expectations led by the Federal Reserve are simultaneously raising the financing costs of these investments. In early Asian trading on September 28, after Trump rejected Iran's latest proposal to reopen the Strait of Hormuz, Brent crude rose 1.4% to $105.80 per barrel, and U.S. Treasuries once again suffered heavy selling: the two-year yield, sensitive to the policy rate, rose 5 basis points to 4.90%, while the 10-year U.S. Treasury yield rose 4 basis points, climbing above 5.20% once again after last Thursday. In the early Asian session, Japanese and Australian government bonds also continued to come under significant selling pressure, following U.S. Treasuries. For computing infrastructure-related companies such as Oracle, Amazon, as well as Google and CoreWeave, which are borrowing to expand AI data centers, this means that new financing will require paying higher interest and other financing costs; for equity investors, it means that future earnings must continue to be tested by a higher discount rate and higher financing costs. Whether AI computing demand can continue to grow, and how much cash return this growth can ultimately leave for shareholders, are becoming two questions that need to be answered separately. The direct catalyst for this round of global bond market adjustment is the market's reassessment of the timeline for energy supply recovery and how much more the Federal Reserve needs to raise rates. In addition, for the yield curve of 10-year and longer-dated U.S. Treasuries, the more critical structural force comes from "rising fiscal deficits + AI bond issuance" competing for the global pool of duration bond funds. To understand the 10-year U.S. Treasury yield, this "anchor of global asset pricing," it is necessary to distinguish between the expected path of future short-term rates and the term premium required for holding long-term bonds. On September 16, the Federal Reserve raised the policy rate by 25 basis points to 3.75%-4.00%. Expectations of further rate hikes under the inflation backdrop will be transmitted to long-term bonds through the future rate path; in addition to the AI bond issuance frenzy and the U.S. Treasury's fiscal expansion jointly competing for the global pool of funds, fiscal financing, inflation uncertainty, and the supply-demand structure of long-term bonds are all factors affecting the additional compensation investors demand. In Japan, financial markets are also losing the buffer provided by the past low-interest-rate environment. Recent quotes show that Japan's 10-year government bond yield reached 3.115% at one point on September 25, the highest since 1996; 20-year and 30-year yields rose to 3.900% and 4.130%, respectively, on September 24. On September 18, the Bank of Japan decided to raise the policy rate from about 1.00% to about 1.25%, and explicitly pointed out that high oil prices, yen depreciation, and AI-related demand are pushing up corporate costs, with some pressure beginning to pass through to consumer prices, while wage pass-through and inflation expectations are also strengthening. Oil prices rise after Trump rejects Iran proposal, bond selloff returns On Monday, U.S. Treasuries came under selling pressure again. Earlier, U.S. President Donald Trump rejected Iran's latest proposal to reopen the Strait of Hormuz, pushing international oil prices higher, with Brent crude rising more than 2% in early trading to approach $107, intensifying market concerns about global inflation. The rate-sensitive two-year U.S. Treasury yield rose 5 basis points to 4.90%, while the 10-year U.S. Treasury yield rose 4 basis points to 5.20%. Japanese and Australian sovereign bond prices also fell. Last week, against the backdrop of hawkish comments from Federal Reserve officials, U.S. Treasury yields across maturities surged to multi-year highs, and Monday's Treasury selloff came after that. As shown in the chart above, the 10-year U.S. Treasury yield is hovering near its highest level since 2007. In early Asian trading hours, Brent crude rose 1.4% to $105.80 per barrel. Earlier, Iran said it would not relax the conditions for reopening the Strait of Hormuz, once again increasing pressure on the Federal Reserve to curb inflation through rate hikes. According to Axios, Trump said that although he rejected Tehran's latest proposal, negotiations are expected to resume this week. Damien McColough, head of fixed-income research at Westpac, said: "The Federal Reserve's continued hawkish signals and oil prices remaining above $100 are key factors driving the bond market weaker." Meanwhile, U.S. Treasury Secretary Scott Bessent urged Federal Reserve policymakers to remain "open-minded" on interest rates, arguing that productivity gains from artificial intelligence and deregulation will help control U.S. inflation. The "yield backlash" of the AI boom: the stronger AI computing demand is, the more expansion requires clear accounting of funding AI financing expansion has extended from investment-grade tech giants to record-breaking high-yield bond deals. On September 24, SoftBank set the terms for a new $10 billion and EUR 1 billion bond issue, totaling about $11.1 billion, which media widely described as the largest high-yield corporate bond transaction on record globally; according to SoftBank's announcement, the scheduled issuance date is September 29. Among them, the 7.5-year dollar bond carries a coupon of 9.75%, and the proceeds will be used for purposes including paying the final $10 billion installment of its additional investment in OpenAI. This transaction shows that AI financing channels remain open, but the price required to obtain huge amounts of capital is already quite high. Large technology companies are also expanding financing scale and extending funding maturities. Google parent Alphabet completed a $25 billion dollar bond issuance on August 10, with the longest tranche maturing in 2066; Amazon raised GBP 4.25 billion, about $5.76 billion, on September 9 through its first sterling bond transaction. More noteworthy is the supply outlook: Goldman Sachs forecast data show that total debt issuance by hyperscale cloud providers is expected to reach a record $420 billion in 2027, 60% higher than the estimated 2026 level. Financing pressure is coming simultaneously from benchmark rates and credit spreads. Market data on September 22 showed that spreads for AI-related issuers were about 115 basis points, higher than the 78 basis points for the broad investment-grade market. Institutions interviewed emphasized that the additional compensation involves not only credit judgment, but also continuously increasing bond supply, uncertainty in issuance pace, and portfolio concentration limits; these spreads cannot be directly equated with default probability. For new dollar fixed-rate financing, the cost is usually composed of the benchmark yield for the corresponding maturity, typically closely linked to the 10-year U.S. Treasury yield known as the "anchor of global asset pricing," plus credit and liquidity compensation. When both components rise at the same time, even if companies can still finance smoothly, their project returns may be compressed. To some extent, this set of contradictions can be summarized as the "long-dated U.S. Treasury yield curve backlash" of the AI boom: in order to realize future productivity dividends, companies first expand demand for chips, power, construction, and financing; if short-term demand grows faster than supply expansion, it may increase cost and inflation pressures, thereby tightening the financing environment. Bessent emphasizes that AI and deregulation can raise productivity and help control inflation, a logic focused on long-term improvement in supply capacity; model analysis by researchers at the St. Louis Fed suggests that even before productivity gains are realized, optimistic expectations for future growth may first stimulate current demand and inflation. The two judgments involve different time horizons: long-term cost-reduction potential cannot automatically offset the long-term financing cost pressure and resource constraints during the construction period. From the financial perspective of AI infrastructure projects, what truly needs attention is how much room can still be left between return on investment and cost of capital, and whether cash flow can keep up with the debt repayment schedule.