Editor's note: The Global Morning Briefing summarizes global news reported by The Seoul Economic Daily.
Iran War Rattles Prices as Soldiers Say Americans Are Not Told the Full Story
The United States has spent more than 50 trillion won on the war with Iran in the roughly five months since fighting began, and rebuilding its weapons stockpiles will take more than five years, according to a congressional study. With no end to the war in sight despite the enormous outlays, a deployed U.S. soldier came forward with photographs of bases destroyed in airstrikes, saying the reality is not being properly reported.
According to Reuters and other outlets on the 15th, the Congressional Budget Office estimated additional Defense Department spending at about $38 billion (roughly 52 trillion won) between the outbreak of the war on Feb. 28 this year and Aug. 1. Of that, replacing expended munitions accounted for $21.7 billion (about 29.68 trillion won), or 57% of the total, followed by $10.4 billion (about 14.2 trillion won) tied to increased aircraft flight hours.
The CBO projected that an additional $2 billion (about 2.7 trillion won) a month would be needed if fighting stays at low intensity, and more than $3 billion (about 4.1 trillion won) a month if it escalates. Stockpiles of key weapons such as interceptor missiles are estimated to have been drawn down by half to as much as two-thirds since June last year, and the CBO concluded that restoring them to previous levels will require at least five years. The surge in war costs is also feeding into prices, with personal consumption expenditures inflation in the first quarter of next year expected to run 0.5 percentage points higher than previously forecast.
Meanwhile, CBS in the United States published photographs, based on accounts from active-duty personnel deployed to the Iran war, of an airborne early warning aircraft struck by a direct hit at Prince Sultan Air Base in Saudi Arabia and of destroyed barracks at Camp Buehring in Kuwait.
One active-duty service member told CBS that Americans have not been told about such damage, describing the situation as being punched in the face with one's eyes closed and saying troops are left only to watch bases being destroyed.
Separately, Yemen's pro-Iran Houthi rebels are widening the scope of their attacks on Saudi Arabia, and the Saudi military said it shot down a Houthi drone launched toward Mecca. Facing a security crisis, Saudi Crown Prince Mohammed bin Salman agreed with Egyptian President Abdel Fattah el-Sisi on a joint response to safeguard the Red Sea.
A Seven-Year Math Gap: What PISA Says About U.S. and Chinese AI Talent
As the United States and China compete fiercely for supremacy in artificial intelligence, critics say declining basic academic skills among American students, compounded by excessive reliance on AI, are leaving them behind China in achievement.
On the 14th, The Wall Street Journal warned in an editorial titled "Chinese Students Are Way Ahead of American Kids" that the U.S. education system is losing its competitive edge, citing results from the Programme for International Student Assessment (PISA) 2025, run by the Organisation for Economic Co-operation and Development.
In PISA 2025, students from China's Beijing, Shanghai, Jiangsu and Zhejiang regions scored 612 in mathematics, 597 in science and 527 in reading, taking first place in math and science. The United States, by contrast, scored 463, 502 and 490, trailing China by 149 points in mathematics. Although those figures represent four regions with high economic and educational levels rather than a national average for China, foreign media assess that Chinese students are ahead of American students overall.
The Journal said the ultimate gap in math ability between the United States and China amounts to more than seven years, and that American schools are not producing enough students with advanced engineering and mathematics skills.
China is presenting that talent pool as a strength of its AI industry. According to Sina Finance on the 15th, Guo Ping, chairman of Huawei's supervisory board and a former chairman of the company, said China is weaker than the United States in computing power but is not behind in data utilization, industrial deployment or talent.
On U.S. campuses, warnings continue that AI is weakening students' thinking skills. In a recent report, a Massachusetts Institute of Technology committee on AI use said generative AI is significantly changing university culture and that a phenomenon of "cognitive surrender" is emerging, in which students turn to AI for answers instead of thinking and researching on their own when they encounter difficult problems.
MIT also raised concerns about deepening isolation as students rely on AI chatbots rather than working through problems with professors or classmates, and about an "illusion of learning" in which students get instant answers and believe they have studied even though they have not fully understood the material. An earlier MIT study also found that people who used chatbots as writing aids had difficulty recalling what they had written immediately after finishing their essays.
Calls to Slow AI Development Collide With OpenAI's Funding Talks
Even as the AI industry warns of the technology's risks and argues for slowing development, reports that OpenAI is weighing a massive new funding round have stirred controversy, coinciding with growing criticism that calls to slow AI amount to large players protecting their own turf.
On the 15th, the Financial Times reported, citing sources, that OpenAI is considering additional fundraising at a target valuation of $1.2 trillion (about 1,630 trillion won).
OpenAI raised $122 billion in March this year at a valuation of $852 billion, and if the new round is completed, its valuation would top $1 trillion for the first time. The talks come as plans for an initial public offering have been pushed to next year. OpenAI Chief Executive Sam Altman said on the 12th that a listing is unlikely before 2027 and that pursuing one would be unwise amid mounting concern over the existential risks of AI.
Against that backdrop, an "AI standards body" said to be under discussion by OpenAI, Anthropic and Google is drawing antitrust scrutiny. Anthropic Chief Executive Dario Amodei proposed on his blog on the 12th that certain safety-related discussions need an antitrust exemption in the interest of AI safety.
Federal Trade Commission Chairman Andrew Ferguson, however, said he was highly skeptical, characterizing the idea as an attempt to build a moat that blocks competitors from entering the market. David Sacks, chair of the White House science and technology advisory council, also said companies can slow down on their own without government permission and said the coordination looks like a cartel.
As criticism mounted, Anthropic co-founder Tom Brown, who handles government affairs, is reported to have held a video conference on the 15th with U.S. Commerce Secretary Howard Lutnick and Emil Michael, chief technology officer of the Defense Department, in what is presumed to be an attempt to ease tensions with the administration.
Talk is plentiful, but actual regulation is expected to take time. OpenAI has said it will support passage of AI regulatory bills moving through the Senate and House, but the House will not reconvene until the November midterm elections, making passage this year unlikely.
AI Shakes Treasury Demand in U.S. 'Reverse Crowding Out'
Big Tech companies raising money for AI infrastructure are absorbing investment in the U.S. ultra-long bond market at yields above those on Treasuries, pushing up U.S. government bond yields, according to one analysis.
On the 16th, Japan's Nihon Keizai Shimbun said competition for funds between the government and AI companies is intensifying in the U.S. ultra-long bond market.
When government bond issuance lifts market interest rates and dampens private investment, the phenomenon is called crowding out. Recently the reverse has been occurring, the paper said, with corporate fundraising encroaching on the government's channel for financing its budget, a "reverse crowding out" effect.
According to the London Stock Exchange Group, issuance of U.S. Treasuries with maturities beyond 20 years totaled $229.6 billion (about 314.39 trillion won) from January through August this year, down 26% from a year earlier, while issuance of dollar-denominated investment-grade corporate bonds rose 78% to $263.7 billion. Corporate bond issuance by technology companies in particular jumped from $16 billion (about 21.9 trillion won) in 2024 to $61.1 billion (about 83.66 trillion won) in the January-August period this year, and the sector's market share climbed from 6.1% to 24.1%, overtaking energy and finance to become the largest.
A prime example is Google parent Alphabet, which issued its first yen-denominated corporate bond in May this year and followed with bonds denominated in euros, Swiss francs and Canadian dollars. In February, it drew attention by issuing a 100-year bond denominated in British pounds.
Securities industry officials say Alphabet is securing funds without concern for a wider spread over Treasury yields, driven by a belief that falling behind in the AI investment race would leave it permanently behind, Nikkei reported.
The problem is that demand once headed for the Treasury market is shifting to corporate bonds. Masahiro Koide, an executive officer at Mizuho Securities who oversees corporate bond underwriting, said that if AI and technology companies keep issuing bonds in the ultra-long segment to the point of pushing aside demand for U.S. Treasuries, it could spill over into higher long-term U.S. interest rates. The 30-year Treasury yield has been trading in the 5.3% range this month, the highest since 2007.
Micron Unveils 512GB Server DRAM, Widening AI Memory Race to DDR5
Micron Technology has unveiled a 512-gigabyte DDR5 DRAM module aimed at AI and data center servers. It doubles capacity while cutting power consumption by more than 60% compared with existing next-generation server products, achieved by applying to a DRAM module the vertical stacking technology used in high-bandwidth memory for AI. Competition in server memory is expanding beyond HBM into ultra-high-capacity server DRAM.
Micron said on the 15th that it had demonstrated the 512GB DDR5 DRAM module on a next-generation server platform. Maximum data transfer speed is 9,200 MT/s, or 9.2 billion transfers per second, and a server with 24 memory slots fully populated can use up to 12 terabytes of DRAM. AMD and Intel are validating the product on their next-generation server platforms, and Micron is targeting mass production in the second half of 2027.
The key features are high capacity and power efficiency. Existing next-generation DDR5 server modules topped out at 256GB, and doubling that allows fewer data exchanges with storage in AI inference and large-scale data analysis, improving processing efficiency.
Power consumption also improved. According to Micron, the 512GB module operates at 16 watts, more than 60% below the 44.2 watts required to assemble the same capacity from four 128GB modules, and in some data analysis workloads performance was up to 40% higher than with a 256GB DDR5 configuration.
The new product uses stacking technology that piles multiple DRAM chips vertically and connects them with through-silicon vias. The same technology is used in HBM, but unlike HBM, which transfers data at high speed next to a graphics processing unit, this product is DDR5 serving as a server's main memory.
Samsung Electronics developed the industry's first 512GB DDR5 module using eight-layer TSV technology in 2021, but the maximum capacity in its officially released DDR5 server lineup is currently 256GB. SK hynix is also targeting the AI and data center market with 256GB-class DDR5 modules, and competition among the three memory makers is intensifying in ultra-high-capacity server DRAM as well as HBM.
