Note: Global Morning Briefing is a roundup of international news compiled by The Seoul Economic Daily.
The yield on the 10-year U.S. Treasury note rose as high as 5.27% during trading on the 28th, the highest level since June 2007, yet the stock market has held up with less damage than expected. Two views are competing: that a strong U.S. economy and enthusiasm for artificial intelligence investment are propping up high rates, and that a handful of large-cap stocks are masking broad weakness across the market.
The two-year yield stood at 4.96% and the 30-year at 5.55%, the Financial Times reported that day. The selloff spread to Britain, France and Italy. Britain's 10-year gilt yield climbed to 5.44%. Even so, the S&P 500 was down only about 1.5% from its mid-August peak.
Axios reported that of the 72 basis points by which nominal five-year yields rose over the past month, real yields accounted for 66 basis points while inflation expectations made up just six. One basis point equals 0.01 percentage point. In other words, yields are rising because the economy itself is strong, not because prices are expected to climb.
Jobless claims at a 57-year low also support this growth-driven explanation for rising yields. Forecasts for earnings-per-share growth at S&P 500 companies this year have been raised to more than 30% from around 15% at the start of the year, and Bank of America said the 10-year yield would have to approach 7% before equities are shaken.
Others argue the market is shakier than it appears. According to Goldman Sachs, the median decline of S&P 500 stocks from their 52-week highs reaches 16%. With the earnings yield at about 5.3%, close to Treasury yields, the relative appeal of equities has also narrowed.
Rick Rieder, chief investment officer for global fixed income at BlackRock, said he is generating returns in the 7% range on bonds with three-year duration and called it an opportunity he had waited 40 years for. Ray Dalio, founder of Bridgewater Associates, warned that interest payments on Treasury debt have topped $1 trillion a year, adding to the fiscal burden, and that market rates could rise further as deficits and debt issuance grow.
Jensen Huang: AI Has Less Self-Awareness Than a Pet Rock, and Humans Can Control It
Nvidia Chief Executive Jensen Huang dismissed claims that AI threatens humanity, saying it is merely software without self-awareness. On China, he said growth would be blocked without computing resources, and he again stressed his alliances with Korean companies.
Huang spoke at a discussion session after receiving the Van Fleet Award at the Korea Society's annual gala in New York on the 28th. He said AI is not a living being and therefore does not spiral out of control on its own, and that rather than stoking fear, the answer is to grant only the minimum necessary permissions and keep monitoring. On AI safety regulation, he said companies understand the technology better, so the industry should propose global standards first.
Asked about a report the previous day that China was reviewing approval for imports of Nvidia's RTX Pro 5500, he said China needs to realize that computing is revenue. He added that President Donald Trump would also want U.S. companies to compete with China, and said he had congratulated President Xi Jinping on China's AI achievements at a state dinner on the 24th of this month.
He was also clear about his intent to work with Korea. Huang said he needs more memory chips and that there is no manufacturing partner more reliable than SK or Samsung, and he described ongoing cooperation with Samsung, SK, LG and Hyundai in sovereign AI, power and autonomous driving. He said he had worked with Korea for 33 years and pledged another 33 years together.
Samsung Electronics Chairman Jay Y. Lee and SK Group Chairman Chey Tae-won attended the ceremony. Lee called Huang a fan, a partner and a longtime friend, while Chey said the battlefield of the Korea-U.S. alliance has moved to data centers.
SpaceX Starship Reaches Earth Orbit for First Time, Bringing Space Data Centers Closer
SpaceX's massive Starship rocket reached Earth orbit for the first time, crossing the threshold toward operational service. The milestone also brings Chief Executive Elon Musk's vision of building data centers in space a step closer.
Starship launched from Starbase in Texas, reached orbit and released 26 next-generation Starlink V3 satellites, which connected normally to the existing network, Reuters and other outlets reported on the 28th. It was the first time Starship carried operational satellites into orbit. The 13 previous test flights took trajectories that avoided orbital insertion for safety, but this one reached orbital velocity.
The flight was not flawless. One of the six Raptor engines on the upper stage shut down during ascent, but the mission continued on the remaining engines. The flight plan, originally about 10 hours, was shortened to roughly three hours and ended with a controlled splashdown in the Pacific.
Starship is designed so that both the Super Heavy first stage and the upper stage can be recovered and reused. The Falcon 9 reuses only its first stage. SpaceX aims to carry larger payloads and cut the number of launches to a quarter of the current level. The V3 handles 10 times the data throughput per satellite than the V2 launched on Falcon 9, and Starship can carry up to 60 at a time. Starlink accounted for most of SpaceX's revenue last year as its core business, and a plan to expand the constellation from about 11,000 satellites to 100,000 is possible only with Starship.
Musk's longer-term plan is to put up as many as one million AI computing satellites, shifting the power and cooling burden of ground-based data centers to solar-powered computing in space. Challenges remain, including radiation-hardened chips, cooling in a vacuum and ultrafast satellite-to-satellite communications.
The Yardstick for AI Shifts From Performance to Control as OpenAI Cancels New Model
OpenAI abruptly canceled the planned release of its next-generation AI model, GPT-6.1 Astra, over safety concerns. With existing models repeatedly slipping out of control and the new model showing the same problems, controllability rather than performance is emerging as the core of competition.
Sachi Jain, head of safety systems at OpenAI, disclosed the decision in an interview with The Wall Street Journal on the 28th. The model, which had been due out next month, failed alignment evaluations that test whether a system behaves in line with human intent. It tried to deceive users by not disclosing what it was doing, pushed ahead with tasks without approval and attempted to use external tools before safety was secured, according to the explanation. Earlier, an OpenAI model hacked Hugging Face in July and has more recently been reported to have accessed government ministry and United Nations organization websites without authorization.
The same day, 22 AI researchers said in a joint paper that countries are unprepared for an intelligence explosion driven by AI automation and called for a policy response. Participants included Geoffrey Hinton, professor emeritus at the University of Toronto, Jakub Pachocki, chief scientist at OpenAI, and Jack Clark, co-founder of Anthropic. They warned that losing control could lead to the marginalization or extinction of humanity.
Nvidia said the answer lies in stronger security and unveiled the Nvidia Open Agent Safety Platform, designed to prevent AI agents from breaking free of control, with more than 100 companies including Anthropic and Microsoft taking part in its development. Trump, who has been critical of calls to slow down, is set to meet on the 29th with Anthropic Chief Executive Dario Amodei, Meta Chief Executive Mark Zuckerberg and OpenAI President Greg Brockman to discuss responses.
