Korea's September Exports Top $120 Billion for First Time; Semiconductor Exports Surge 262.8%; U.S. Treasury Yields and Micron Earnings Move Markets

Korea's September Exports Top $120 Billion for First Time; Semiconductor Exports Surge 262.8% YoY

According to preliminary data released by South Korea's Ministry of Trade, Industry and Energy on Oct. 1, Korea's September exports jumped 83.5% year over year to $120.94 billion, the first time monthly exports have exceeded the $120 billion mark. The growth rate accelerated from 68.7% in August and far exceeded the median forecast of 65.4% from 13 economists surveyed by The Wall Street Journal. The monthly trade surplus reached $49.85 billion, setting a record. ^[1]

Semiconductors were the core engine of the export surge: September semiconductor exports soared 262.8% YoY to $60.3 billion, surpassing the $46.7 billion record set in August and accounting for about 49.9% of total exports for the month. Computer exports, highly linked to semiconductors, jumped 435.3% YoY to $7 billion, driven by expanding AI infrastructure demand; wireless communication device exports rose 23.4% to $2.14 billion, benefiting from strong sales of high-end smartphones and iPhone components. ^[1]

Cumulative Exports in First Nine Months Already Exceed All of Last Year

From January to September this year, Korea's cumulative exports reached $814.5 billion, already surpassing last year's full-year record of $709.3 billion, with a YoY increase of 56.8%. On imports, September imports rose 26.0% YoY to $71.09 billion, producing a monthly trade surplus of $49.85 billion; the cumulative trade surplus for the first nine months of this year has reached $252.5 billion. ^[1]

September coincided with Korea's Chuseok holiday, with fewer actual working days than in the same period in previous years, but average daily exports still rose 105% YoY to $5.63 billion, the first time in history exceeding $5 billion per day. By category, 11 of 20 major export categories recorded growth in September: ship exports rose 29.9% YoY to $3.76 billion on increased deliveries of liquefied natural gas (LNG) carriers; cosmetics exports reached $1.51 billion, a record high, maintaining steady growth in major markets including China, the U.S., ASEAN and the EU. ^[1]

Inflation Data Beats Expectations, Yet U.S. Treasury Yields Still Rise

The U.S. Treasury selloff continued, and even as inflation data unexpectedly softened, the 10-year Treasury yield climbed to a multi-decade high of 5.28%. ^[2]

The 10-year U.S. Treasury yield surged more than 50 basis points in September alone, breaking above 5.3% and touching its highest level since 2002, far above its 2007 peak; the 10-year real yield rose 57 basis points in the month. According to reports, this selloff has evolved from being fundamentally driven into a vicious cycle dominated by technical forced selling—rising yields triggered some funds to cut positions under duress, further pushing down bond prices and driving up borrowing costs, which in turn sparked a new round of selling. ^[6]

On inflation, data released Wednesday showed the Fed's preferred inflation gauge, the personal consumption expenditures (PCE) price index, held at 3.4% YoY in August, below the market expectation of 3.7%, but the result had little effect in lifting the bond market. ^[6]

Wall Street offered two interpretations of the selloff's causes. Ed Yardeni, president and chief investment strategist at Yardeni Research, believes the unwinding of the yen carry trade may be the main cause; Shoki Omori, chief Japan fixed income strategist at Deutsche Bank, believes the rise in Japanese government bond (JGB) yields itself is the key, and noted that the current "simultaneous decline in bonds and the yen" is more consistent with inflation and interest-rate repricing. ^[2]

Trump Announces Voluntary 'AI Agreement,' Six Companies Sign

On Sept. 30, after lunch with more than 20 AI executives in the East Room of the White House, Trump announced to reporters a voluntary "AI agreement," signed by six companies: Anthropic, OpenAI, Google, Meta, xAI and Nvidia. According to the Financial Times, the two-page document was hastily drafted in the East Room and even contained a typo. ^[3]

According to the text of the agreement, signatories must fulfill three core obligations: establish a "multi-layered control and audit" mechanism, work with independent external auditors, and set up an independent committee at the board level. The agreement is not legally binding, and Trump himself called it a "moral constraint." ^[3]

Micron: Record Quarterly Results, Tighter Supply-Demand in 2027

Micron Technology closed out fiscal 2026 with a quarterly earnings report that broke multiple records. Results released after the U.S. market close on Sept. 30 Eastern Time showed that in the fourth quarter of fiscal 2026 ended Sept. 3, Micron's quarterly revenue reached a record $54.2 billion, up 31% QoQ and up 379% YoY, while quarterly gross margin climbed to 87%; total revenue for all of fiscal 2026 reached $133.2 billion, 3.5 times the record level of the previous fiscal year, with data center revenue quadrupling. ^[4]

On the earnings call, Micron Chairman and CEO Sanjay Mehrotra said the company has signed 26 long-term agreements locking in about $150 billion in long-term orders, and that memory market supply and demand in 2027 and 2028 will be even tighter than in 2026, with "no end in sight to supply and demand returning to balance." CFO Mark Murphy said first-quarter fiscal 2027 revenue reached $61.5 billion and earnings per share were $38.15, and noted that this quarter will be the low point for gross margin in fiscal 2027. Micron also announced capital expenditures of $25 billion in the first half of fiscal 2027 and pledged to return 100% of excess cash to shareholders in the future. ^[4]

Google's Gemini 4 Release Draws Internal Doubts

On Wednesday, Google officially released Gemini 4 Argon externally, positioning it as the flagship AI model from the company's long-term R&D. However, according to Bloomberg, although the model scored leading marks on multiple benchmarks, some Google insiders with direct access to the model were reserved about its actual performance, especially in coding ability. Google explicitly denied the doubts, saying claims that Gemini 4 performs poorly in coding and other areas were "inaccurate"; after the news emerged, Alphabet's after-hours gains narrowed significantly. ^[5]

The report cited people familiar with the matter as saying the model is lacking in front-end design, while some employees believe Anthropic's Fable and OpenAI's Astra have surpassed Gemini in the pace of capability iteration. Industry insiders attribute the phenomenon of "high benchmark scores, mediocre real-world experience" to the industry's chronic "over-optimization" (Benchmaxxing). In addition, the report cited people familiar with the matter as saying Gemini 4 is massive, and large models usually mean higher inference costs. ^{[5] [1]}