Meta-Fueled AI Overinvestment Fears Diverge From Reality as Memory Shortage Persists
Summary
- Eugene Investment & Securities said concerns over AI overinvestment sparked by Meta are largely disconnected from reality, and that spending on AI data centers is continuing.
- It said the key issue for idle equipment is whether there is a market where it can be monetized, which suggests ROE from AI computing rental revenue has increased.
- It emphasized that as AI data-center investment continues, the memory shortage and memory bottleneck will be difficult to resolve quickly.
Forecast Trend Report by Period
Eugene Investment & Securities report
Concerns that Meta Platforms Inc.'s move into AI cloud infrastructure could stoke excessive spending are largely disconnected from reality, according to Eugene Investment & Securities. What matters is not the existence of idle equipment, but whether there is a market where it can be monetized. The brokerage also said continued investment in AI data centers means the memory shortage is unlikely to ease quickly.
Lee Seung-woo, head of research at Eugene Investment & Securities, wrote in a July 13 report that debate over Meta's sale of surplus computing capacity has once again fueled the AI-bubble narrative. But it also shows that returns on equity from AI computing rentals have risen to the point that even older graphics processing units displaced from training the latest models can still be sold externally.
"What matters is not the existence of idle equipment, but whether there is a market where that equipment can be monetized," Lee wrote.
Lee said concerns in some corners of the market that excess computing resources will slow data-center investment are out of step with reality. Meta recently announced plans to begin construction of a 1-gigawatt data center in Canada that can be expanded to 1.8 gigawatts. It has also said it will double AI data-center infrastructure to 14 gigawatts next year from 7 gigawatts this year.
U.S. semiconductor shares have risen, while South Korea's top two memory makers — Samsung Electronics Co. and SK Hynix Inc. — have fallen sharply, Lee wrote. The debate over Meta's surplus computing capacity, combined with distorted market flows in South Korea's stock market, has created an extremely confusing environment, he added.
But both top-down and bottom-up indicators point to the same conclusion: investment in AI data centers is continuing, cash flow remains solid and memory bottlenecks will not be resolved easily.
Jungsam Ko, Hankyung.com reporter, jsk@hankyung.com
Korea Economic Daily
hankyung@bloomingbit.ioThe Korea Economic Daily Global is a digital media where latest news on Korean companies, industries, and financial markets.