AI-linked stocks fell sharply overnight in the US, with the Nasdaq Composite closing down 1.33% and the Philadelphia Semiconductor Index losing nearly 5%. Names tied to memory, optical communications and AI computing power broadly came under pressure, and the sell-off quickly spilled into other markets.
The article, written by Gelong, says the downturn was driven mainly by three factors: a macro shock tied to oil and inflation expectations, widening doubts about the commercialization path for large AI models, and growing uncertainty in the semiconductor supply chain because of tensions between South Korea and the US over chip investment.
US weakness spread quickly into Asia and China
After the US session, selling pressure moved into Asia-Pacific trading. In South Korea, Samsung Electronics and SK Hynix each fell more than 6%. Leveraged ETFs offering 2x long exposure to Samsung and 2x long exposure to SK Hynix both plunged more than 14%, with leveraged positioning amplifying the move.
China A-shares also felt the impact. Core AI-linked segments including computing power, optical modules and memory generally dropped more than 5%, while several leading names in sub-sectors fell more than 8%.
Oil, inflation and Treasury yields added pressure to growth valuations
The first bearish factor cited in the piece is a rapid shift in the macro backdrop. Escalating tension between the US and Iran pushed up oil prices and inflation expectations, which in turn weighed on valuations across growth sectors.
According to the article, the US announced a suspension of talks with Iran, intensifying regional strains and increasing concerns around the Strait of Hormuz. International oil prices rose again and moved above a three-week high. At the same time, the United Arab Emirates announced a suspension of all trade and financial dealings with Iran, adding to the regional tension.
Higher oil prices revived worries about a rebound in inflation and pushed investors to reassess how much room the Federal Reserve has on policy. Long-dated Treasury yields climbed sharply. The 30-year Treasury yield briefly touched its highest level since 2007, while the 10-year yield also rose markedly.
That environment hits AI particularly hard because the sector combines high valuations with heavy capital spending. The article says AI computing infrastructure depends heavily on debt financing, and AI-related bond supply this year has already exceeded earlier expectations for the full year. It also points to bonds issued by Blackstone for Microsoft data centers, where the offered yield has approached junk-bond territory, as a direct sign that financing costs for AI infrastructure are rising.
Goldman Sachs, as cited in the piece, said massive AI capital expenditures layered on top of sovereign deficits have driven large sums into the bond market, creating the possibility that the Federal Reserve may still have to keep policy relatively tight even if economic data weaken.
As financing costs keep rising, investors are rechecking the logic of expanding compute capacity at any price, and AI hardware names were among the first to be sold.
OpenAI slowdown raised fresh questions about commercialization
The second pressure point came from a more visible split in AI model commercialization. The article says OpenAI's performance slowdown disrupted the market's earlier linear optimism about the application side of AI.
Based on newly disclosed second-quarter figures, OpenAI posted quarterly revenue growth of only 18% from the prior quarter. Operating losses continued to widen, and a series of executive departures raised questions in the market about the stability of internal management.
The article notes that rival Anthropic recorded explosive revenue growth and a small profit. Even so, it cautions that differences in revenue definitions between the two companies mean those results should not be read as proof that the broader industry has already entered a profitable phase.
OpenAI's slower growth reminded the market that commercialization for large AI models is not a smooth process. Enterprise customer conversion and cost control remain major challenges.
In the article's framing, investors had previously accepted the AI capital-spending story with few conditions. Now they are asking how much real revenue and profit all that computing investment can actually generate. That shift means the AI supply chain is moving out of a burn-cash-and-scale phase and into a period where commercialization ability is being tested.
Korea-US chip investment friction added uncertainty to HBM
The third factor was the ongoing tug-of-war between South Korea and the US over semiconductor investment. The article says that conflict is adding uncertainty to the global memory supply chain and directly hitting the HBM segment, which is central to AI computing infrastructure.
South Korea has publicly denied reports that the US asked companies to prioritize building memory-chip plants in the US. At the same time, South Korea has already mapped out more than $580 billion in funding for domestic chip and data-center clusters.
If Samsung and SK Hynix are forced to build memory production lines in the US on a large scale, the article argues, that would consume significant corporate capital and weaken South Korea's domestic semiconductor ecosystem. It adds that US pressure is not limited to tariffs, and delays in investment execution could also spill over into security cooperation between the two countries.
The article presents that tension in practical terms: South Korean memory makers earn substantial profits by supplying HBM to the US AI market, while the US wants that capital brought back onshore for factory construction, with trade penalties hanging in the background if companies refuse.
Markets are worried that if negotiations drag on, either outcome could disturb the global memory supply structure.
- If South Korea compromises, corporate capital could be diverted and profits could be eroded by the high cost of building in the US.
- If South Korea takes a harder line, companies could face trade-barrier risks.
That dilemma, the article says, triggered selling in Samsung and SK Hynix, while leveraged ETFs magnified the drop and fear spread outward through the memory supply chain.
Focus is shifting from spending scale to earnings and financing
The article also argues that a sharp short-term decline does not mean the AI thesis has ended. Long-term demand for AI computing power still exists, but investors are no longer willing to pay extreme premiums for open-ended optimism far into the future.
Attention is now likely to move away from capital-expenditure size alone and toward actual corporate profitability, changes in financing costs and the eventual direction of global supply-chain negotiations.
For China's A-share market, the piece says the external shock currently looks more like a sentiment disturbance. What comes next will depend on orders and earnings delivery within the domestic supply chain, which will help determine whether the sell-off was mainly emotional or reflected genuine deterioration in fundamentals.