Power, Not HBM, Is AI's Real Bottleneck, Nvidia Partner Says

GMI Cloud's strategy suggests a calculated move to regions where the company perceives fewer obstacles in developing the foundational power infrastructure required for large-scale AI operations. The CEO noted that while HBM can be acquired on the spot market, power requires extensive facilities and can face significant local opposition, which is reportedly less of a burden in Asia.

The planned $12 billion investment for a 1-gigawatt AI factory in Kagoshima, Japan, alongside a $500 million data center in Taiwan, indicates a tangible shift of AI compute capacity towards the region. This expansion is supported by GMI Cloud's ability to raise funds in Japan at half the cost of competitors and its existing data center operations in Thailand and Malaysia.

Alex Yeh also proposed defining GPUs, data, and power as a new form of gross domestic product (GDP), highlighting the critical role these elements play in the AI economy. The company's recent $668 million investment included participation from Korean firms, further illustrating Asia's involvement in the broader AI infrastructure development and its pursuit of sovereign AI initiatives.