Editor's note: Global Morning Briefing summarizes global news as reported by The Seoul Economic Daily.
"America's Real AI Bottleneck Is Power, Not HBM. We'll Find Alternatives in Asia"
GMI Cloud, a key neocloud partner of Nvidia, has identified electricity as the biggest obstacle to the spread of artificial intelligence and said it will increase investment in Asia. The company believes conflicts over securing power are intensifying in the United States, while Asia is moving faster to build sovereign AI.
Alex Yeh, chief executive officer of GMI Cloud, told The Seoul Economic Daily at the company's Mountain View headquarters in the United States on the 5th of this month that electricity is a more fundamental bottleneck than high-bandwidth memory (HBM). HBM can be obtained on the spot market by paying a premium, the CEO explained, but power requires transformers and transmission grid facilities to back it up, and building one's own power plant is easily blocked by opposition from local communities and political circles. Yeh said data center disputes in the United States lead to supply chain delays, while the burden is relatively lighter in Asia. The CEO also cited the ability to raise funds through Japan at half the cost of competitors as a strength.
GMI Cloud is one of Nvidia's seven major neocloud partners and receives priority supply of the latest graphics processing units (GPUs). It operates data centers in the United States, Thailand and Malaysia, and has decided to build a $500 million data center in Taiwan. In Kagoshima, Japan, it plans to invest $12 billion to establish a 1-gigawatt AI factory. The latest Nvidia GPUs to be installed at the two sites total about 9,000 units. Taiwanese media point to Yeh being the son of Yeh Chia-wen, chairman of Hongyi International and a co-founder of Realtek, as the reason behind the company's rapid local partnerships.
Yeh argued that the sum of GPUs, data and power should be defined as a new form of gross domestic product (GDP). He also assessed South Korea as an important market pursuing sovereign AI, citing the Honam semiconductor cluster. The $668 million investment GMI Cloud secured late last month included participation from Korean companies such as KB Investment, Kyobo Life Insurance and KT, alongside Nvidia.
Meta Accelerates AI Monetization With Paid Muse Wonder Release
Meta unveiled Muse Wonder, a closed and paid artificial intelligence model, on the 12th, moving in earnest into monetization. The move marks a step beyond its previous strategy of emphasizing free and open AI, under a plan to separate paid services based on model performance.
Meta released Muse Wonder, its top-tier series, alongside the premium Muse Wonderful. Wonder is a high-performance model to be installed in Muse, the company's existing AI assistant app, and will be provided first to paid subscribers. The price is $4 per million input tokens and $20 per million output tokens. Meta said the model also features "Active Memory," which remembers conversation history and preferences, and "Active Focus," which handles complex and ambiguous tasks through to completion.
Performance was measured by an external evaluation firm. On Artificial Analysis' intelligence index (AAII), Wonder scored 52 for a tie at seventh place, while Wonderful scored 56 for a tie at second. Models ranking above Wonder include Anthropic's Claude Fable 5.1, which scored 53, as well as the latest models from OpenAI and Google, with Claude Opus 5.5 leading at 58. Following the announcement, Meta shares closed the regular session at $742.96, up 3.38% from the previous close.
Chief Executive Officer Mark Zuckerberg said in an interview with the Financial Times of Britain that Meta's AI capabilities have caught up with rivals and that the company will sharply accelerate the pace of model development. He drew a line, however, at concerns that AI will escape human control. Zuckerberg said he is far closer to the belief that it is a technology that can be controlled through safeguards and built responsibly.
