Samsung and TSMC's blowout earnings fail to ignite an AI rally! U.S. Treasury yields hit a 24-year high, and the AI bull market faces a "computing power cash return stress test"

Title context: Samsung and TSMC's blowout earnings fail to ignite an AI rally! U.S. Treasury yields hit a 24-year high, and the AI bull market faces a "computing power cash return stress test" Text: Even as Samsung and TSMC delivered strong report cards, the money flows are revealing a pricing shift in the AI investment frenzy: investors are beginning to demand that data center operators actively prove that continued expansion of AI computing power investment can generate cash returns sufficient to cover higher funding costs. Samsung's preliminary third-quarter operating profit officially announced on Thursday reached 107.4 trillion won, up 782.5% year-over-year, equivalent to 8.825 times the year-earlier level, with revenue up 126.59% year-over-year; TSMC's Q3 revenue grew about 51%, jointly confirming that AI infrastructure procurement remains strong. However, against the backdrop of 10-year and longer-dated U.S. Treasury yields climbing to a 24-year high, persistently elevated energy costs, and continuously expanding financing needs among AI technology companies such as cloud computing vendors, record-breaking results are still not enough to dispel market doubts about the sustainability of the AI computing power boom. On October 8, Samsung closed down 2.42% at 262,000 won, SK Hynix fell 2.44%, and the KOSPI dropped 2.62% to 6,625.93 points, falling more than 2% for a second consecutive trading day. That day, foreign investors and institutions net sold approximately 2.42 trillion won and 2.09 trillion won respectively, while individual investors net bought approximately 3.74 trillion won. This indicates that the market is undergoing a divergence in which institutional funds are reducing positions while retail investors are absorbing the selling; options expiration, semiconductor ETF rebalancing, and position reduction ahead of the Korean holiday also amplified pressure from interest rates and oil prices. The core reason for Samsung's stock weakness lies in the gap between record profits and the earnings sustainability investors demand. An analyst expectation survey used by Bloomberg showed that its profit was slightly below the 108.7 trillion won indicated by Bloomberg estimates; LSEG expectations cited by Reuters were about 106.1 trillion won, and profit was slightly above that level. Therefore, an explanation closer to market performance is that funds are reassessing how long memory price increases can last, how much incremental contribution HBM share gains can make, and whether future memory chip capacity expansion can still maintain high returns if cloud AI application revenue generation is weak. The appreciation of the won also significantly weakened earnings performance after converting dollar revenue into won. Demand in the AI computing power supply chain itself still has clear technological support. A single request from an intelligent agent may trigger multi-step reasoning, retrieval, and tool calls, increasing demand for computation, memory capacity, and data movement; AMD's announced MI455X single GPU is configured with 432GB HBM4, and a Helios rack composed of 72 GPUs is equipped with about 31TB HBM4, reflecting the physical demand for high-bandwidth memory from computing power expansion. Counterpoint raised its forecast for third-quarter DRAM price increases quarter-over-quarter to 10%-20%, and Citi expects 12-layer HBM4 prices may rise 100%-150% next year, both pointing to memory manufacturers still having strong earnings elasticity. Undoubtedly, major chip design and manufacturing companies including Samsung, SK Hynix, as well as Nvidia, AMD, TSMC, and Intel, are realizing profits brought by the shortage of core AI computing hardware, and whether these profits can persist ultimately depends on whether downstream cloud computing customers and data center operators in the broad sense can convert expensive computing power into sustainable AI revenue-generation data. Samsung and TSMC failed to ignite investor enthusiasm, and record-breaking results seem to have lost their sensational effect Samsung Electronics' quarterly operating profit reached nearly nine times the year-earlier level, and TSMC's third-quarter revenue surged 51%, reflecting the generous profit returns brought by continuous AI computing infrastructure spending. However, investors reacted tepidly to the record-breaking results announced one after another that day, because the market is increasingly worried about whether this investment boom, increasingly supported by debt, can continue. On Thursday, Samsung's stock fell 2.4% in Seoul. In the Tokyo market, shares of two of the company's suppliers, including Tokyo Electron, Advantest, and Ibiden, also declined. Investors are questioning how long AI spending can last, especially amid rising global borrowing costs and pressure on companies' massive capital expenditure plans. "Can we see the kind of incredibly strong and sustained returns currently promised by those data center operators, including cloud computing giants?" said Jun Bei Liu, co-founder and chief portfolio manager of Ten Cap Investment Management. Samsung, the world's largest memory chip manufacturer, reported preliminary operating profit for the quarter ended in September of a record 107.4 trillion won, about $80.1 billion, while revenue more than doubled. Still, the profit was slightly below the average analyst forecast. " Samsung delivered record profit and still missed expectations, which is enough to show how demanding the expectation threshold for the AI investment trade has become," said Josh Gilbert, chief analyst for Asia-Pacific at eToro. Technology companies are racing to lock in memory supply needed to run AI services. High-bandwidth memory (HBM) memory chip components-key chips used alongside AI accelerators from Nvidia and others-are attracting massive investment, corporate capital expenditure, and accelerated shifts and allocation of memory capacity, while NAND-related flash memory and low-power chips are also in strong demand. Record-breaking results and elevated expectations highlight how significantly the economic benefits of the AI boom are tilting toward Samsung, SK Hynix, and Micron. As shown in the chart above, the AI investment boom helped Samsung set another record profit. Note: Third-quarter 2026 data are preliminary operating data announced by Samsung management. Counterpoint Research raised its forecast for third-quarter DRAM price increases quarter-over-quarter from the previous 5%-10% to 10%-20%, citing customers placing orders early. " Samsung's current valuation is quite low. Next year, they still have considerable potential to unlock in HBM, mainly because the related benefits have not yet been fully reflected," said Neil Shah, vice president of research at Counterpoint. Citi analysts Peter Lee and Jayden Oh expressed a similar optimistic judgment, expecting 12-layer HBM4 prices may rise 100%-150% next year, significantly boosting Samsung's average selling price and profit margin. However, due to the stronger won, the institution has lowered its full-year profit forecast for Samsung. This boom has transformed Samsung's semiconductor business. Just a few years ago, the business was still loss-making due to a severe demand downturn. Now, AI infrastructure construction is consuming large amounts of advanced DRAM, and Samsung is also narrowing the gap with SK Hynix in HBM. The strong performance of this cycle is also reflected in South Korea's trade data: semiconductor exports in September reached more than three times the year-earlier level. Wall Street analysts on average expect Samsung's chip division to post 110 trillion won in operating profit, while the electronics business division will record a loss. The Suwon-based company will release full financial statements on October 29. Still, caution in the market remains. Even after Samsung announced record results in July, its stock price is still about 25% below its June high. For an industry prone to alternating booms and busts, investors remain somewhat skeptical of profit surges. Samsung and SK Hynix are currently expanding capacity while also increasingly supporting AI-sector companies to drive demand for their own products. As shown in the chart above, Samsung's stock price, after its recent rebound trajectory, is still more than 20% below its all-time high. Highlighting current market conditions, AMD CEO Lisa Su emphasized long-term strategic partnerships with Samsung and SK Hynix during her visit to Seoul this week. Su confirmed that AMD has begun volume shipments of its next-generation Instinct MI455X and Helios systems using HBM4. "Memory is absolutely critical to the core computing process of overall AI data centers; supply has been very tight," Su told reporters. She added in the interview that AMD's HBM cooperation with Samsung and SK Hynix will continue across multiple product generations. "We are encouraging partners to build as quickly as possible." From energy shocks to the market's urgent pursuit of cash returns on computing power investment The impact of the U.S.-Iran situation on asset prices is continuing to transmit through shipping, fuel, and financing costs. Iran recently reiterated that it will not restore normal navigation through the Strait of Hormuz until relevant conditions are met; from September 28 to October 5, at least 12 attacks, attempted attacks, or harassment incidents targeting oil and gas transport vessels occurred around the strait, the highest weekly level since the war began. Persistent shipping risks make it difficult for the market to be confident that energy supply can recover steadily. As of 18:41 Beijing time on October 8, Brent crude oil futures were at $105.20 per barrel, up 4.99% on the day, and WTI was at $92.75 per barrel, up 5.06% on the day. Based on the closing prices on February 27, the last trading day before the war broke out on February 28, of $72.48 and $67.02, the two have risen about 45.14% and 38.39% respectively. That day's gains were also affected by production shutdowns caused by a hurricane in the U.S. Gulf of Mexico, leaving the energy market under the dual pressure of geopolitical and weather disruptions. The pressure of energy inflation is even more prominent in refined oil products. The International Energy Agency pointed out that nearly 3 million barrels per day of refining capacity in the Middle East has been shut down due to attacks and blocked exports, U.S. diesel prices have nearly doubled from pre-war levels, and Europe and Asia are close to that increase. Rising diesel prices directly raise transportation, engineering construction, and industrial production costs, while disruptions to natural gas supply increase electricity cost pressure in some regions. The IEA has decided to accelerate the implementation of previously promised reserve releases and prioritize diesel supply, but releasing inventories mainly alleviates short-term gaps, and the recovery of shipping and refining facilities still determines whether supply can continue to improve. Long-term bond pricing includes the average expectation of short-term interest rates over many future years, as well as the term premium investors require to hold long-term bonds. The market reducing bets on a rate hike at the next meeting does not mean the long-term financing environment is easing in tandem. Energy shocks make it harder for inflation to fall, fiscal deficits and expanded bond supply increase financing pressure, and AI companies borrowing debt compete with governments for long-term funds. The UK also faces a test of fiscal credibility before the budget, while Japan is simultaneously affected by doubts about fiscal expansion and expectations of central bank normalization. In the global government bond market, the U.S. 10-year Treasury yield, known as the "anchor of global asset pricing," briefly surged to about 5.36% during trading on October 7, setting another highest level since 2002. The U.S. 30-year Treasury yield briefly rose to about 5.73% during trading, remaining continuously in the highest range since 2002, about a 24-year high, while the UK 30-year Treasury yield hit its highest level since January 1998, hovering near 6.05%. This round of U.S. Treasury yield increases also has a characteristic directly related to AI valuations: real interest rates and term premiums are rising, causing yields on AI technology corporate bonds that compete with U.S. Treasuries for the same long-duration funding pool, as well as these companies' financing costs, to continue climbing. A Reuters analyst team, citing a New York Fed model, pointed out that the 10-year U.S. Treasury term premium has risen to its highest since 2014, and the recent rise in nominal yields has mainly been accompanied by higher real yields, while long-term inflation expectations have remained relatively stable; U.S. Treasury Department data show that on October 7 the real yield on 10-year Treasury Inflation-Protected Securities reached 2.92%. All of this means that funding providers' demands for real returns in the credit market are rising, and companies' forward profits need to clear a higher valuation threshold. The continued rise in long-term yields is becoming a powerful catalyst for pricking part of the AI asset bubble, and its transmission process has a clear sequence. First, a higher discount rate lowers the present value of future cash flows, and technology stock valuations come under pressure first; subsequently, as new financing costs determined jointly by Treasury benchmark rates and credit spreads rise, data centers need higher utilization, more stable customer payments, and faster cash recovery to cover construction, chip procurement, operations and maintenance, and debt expenses; if these conditions cannot be met, project delays and downward revisions to capital expenditure will then gradually transmit to orders for GPUs, HBM, advanced packaging, and equipment. The profit Samsung realized today comes from purchases that have already occurred; the stock market tends to assess in advance how quickly the next round of purchases can continue. The truly economically meaningful "AI kill line" is when the expected return on new computing power projects remains below their cost of capital, further damaging financing and expansion capacity. This also explains why Panmure Liberum senior strategist Joachim Klement linked high financing costs with the sustainability of AI investment and proposed a pessimistic bear-case scenario in which the AI bubble may burst in 2027-2028. For popular AI technology stocks, the next stage of divergence will focus more on: who can convert Token growth into revenue growth, who can improve profit per unit of computing power through software and hardware efficiency gains, and who can rely on sustained cash returns to support expansion. The record-breaking profits of AI computing power leaders still matter, but the sustainability of capital returns is determining how much valuation the market is willing to pay for those profits.