Semiconductor-linked trades came under broad pressure from the U.S. session into Asian hours after several negative catalysts hit in quick succession. Nvidia closed down about 5%, the Philadelphia Semiconductor Index fell 2.23%, and the Nasdaq finished slightly lower.
When Asia opened, related AI hardware assets in South Korea, Japan and Hong Kong extended the move. South Korea’s KOSPI fell more than 10% intraday and triggered circuit breakers twice. Heavyweights SK Hynix and Samsung Electronics dropped sharply, while leveraged Hong Kong products tied to the two names saw deeper losses, with the South Korea 2x Long Hynix ETF and 2x Long Samsung Electronics ETF both down more than 20%.
Four lines of pressure converged on the sector
The sell-off was driven by four themes being traded at the same time: concern over China’s progress in semiconductor localization, renewed unease about “circular financing” tied to Nvidia-related AI infrastructure deals, profit-taking after a strong rally in semiconductor shares, and fresh skepticism toward the heavy spending plans of U.S. and Korean tech leaders after Kimi K3 was open-sourced.
China DUV production report hit equipment names
According to The Information, a company backed by Chinese state capital has begun mass production of self-developed immersion DUV lithography equipment, with plans to produce about five units in 2026 and roughly 20 in 2027. That remains far below the 131 immersion DUV systems ASML delivered last year, but the start of domestic DUV production was still seen by the market as an important step for China’s chip supply-chain autonomy.
ASML fell 5.80% on Monday after the report. Weakness spread through the U.S. storage and equipment chain, with SanDisk down 11.02% and Western Digital off 4.21%.
Some institutions said the reaction may have been too aggressive. JPMorgan said China’s domestic immersion DUV equipment is still in the small-scale production stage, and its performance, reliability and large-scale manufacturing capacity remain to be proven, leaving it far from genuinely replacing ASML’s systems. Samsung Securities said Chinese AI chips and server DRAM are unlikely to enter the U.S. data-center ecosystem in the short term, limiting the immediate impact on the current AI semiconductor cycle. Citrini analyst Jukan also said the report offered limited new substance and that the sell-off in equipment names such as ASML looked excessive.
Storage-chain concerns returned after CXMT listing surge
Another source of pressure came from China’s memory industry. ChangXin Memory Technologies surged 466% on its market debut, becoming the largest A-share listed company by market capitalization in China. The market read that as a sign that self-sufficiency in China’s memory supply chain could accelerate.
For global storage leaders including SK Hynix, Samsung Electronics, SanDisk and Western Digital, a better-capitalized Chinese competitor raised the prospect of a longer-term shift in supply dynamics. Older bearish narratives around the memory chain were traded again during the session.
Nvidia CDS jump pushed credit worries into equities
Credit concerns around Nvidia’s $750 billion AI infrastructure cooperation plan also moved from the bond market into stocks. Bloomberg reported that Nvidia was said to be discussing guarantees of up to $250 billion to help OpenAI lease computing capacity from U.S. data-center projects. The company also disclosed cooperation linked to SK Group, the parent of SK Hynix, worth more than $500 billion.
Data from ICE Data Services showed Nvidia’s five-year credit default swaps rose by about 14 basis points at one stage to roughly 82 basis points per year, the biggest intraday jump since the contract began trading actively. A CDS can be understood as insurance against debt default. When CDS pricing rises, the market is usually signaling higher perceived debt risk or greater concern over repayment pressure.
For Nvidia, the credit market is reassessing the balance-sheet strain that could come from large guarantees, customer financing and AI infrastructure cooperation. Hideyuki Ishiguro, chief strategist at Nomura Asset Management, said investors viewed the rise in Nvidia’s credit risk as a negative signal.
Kimi K3 reopened the debate over AI capex efficiency
Pressure also built at the AI model layer. Moonshot AI’s Kimi K3 opened its model weights on July 27, a move widely seen in the market as another important step for Chinese open-source AI models. Public information shows Kimi K3 is a 2.8 trillion-parameter model that supports long context, multimodal functions and agent capabilities, while aiming for near-frontier performance at lower usage cost.
After DeepSeek, another high-performance open-source model from China has moved onto the radar of global investors. That prompted a fresh look at the valuation basis for U.S. closed-source model providers, cloud companies and the AI hardware chain. For U.S. equities, the pressure from Kimi is not about one model alone. It comes from the renewed strength of the “high-performance, lower-cost AI” narrative.
If better model performance no longer depends entirely on ever-larger GPU clusters and ever-higher capital spending, investors are likely to ask whether the marginal returns on continuing data-center expansion by Microsoft, Meta, Amazon and OpenAI could decline, and whether the high valuations enjoyed by Nvidia, Broadcom, AMD and the storage and equipment supply chains can still hold.
In that sense, Kimi’s open-sourcing deepened concern over AI capital efficiency and forced a repricing of the idea that adding more compute is always the answer. More importantly, the market’s focus within the AI trade has shifted. Previously, tech stocks were driven mainly by fast growth in demand for computing power. Now low-cost open-source models such as Kimi, Nvidia’s customer financing arrangements and the scale of data-center capital spending are showing up together, pushing investors to ask whether AI investment can generate enough cash flow and how much customer demand depends on supplier-backed financing.
Fed and BOJ meetings added a macro layer
Macro pressure is building at the same time. Both the Federal Reserve and the Bank of Japan are due to announce rate decisions this week, and concern is rising over the possibility that both central banks could lean hawkish.
The Federal Reserve meets on July 28-29. The current federal funds target range is 3.50%-3.75%. Most traders still expect no change, but the market has priced in roughly a one-third chance of a surprise rate increase. Goldman Sachs called the outcome of this meeting “exceptionally uncertain.” With AI capital spending, oil prices and tariffs still capable of pushing inflation higher, a more hawkish Fed signal could increase valuation pressure on expensive technology and semiconductor shares.
The Bank of Japan meets on July 30-31 and is expected to keep its policy rate unchanged at 1.00%, though the market is more focused on whether it leaves the door open for another hike. Reuters said the BOJ may stay on hold this week while preserving room for further tightening to deal with a weak yen, energy prices and inflation expectations. The BOJ had already raised rates by 25 basis points to 1.00% in June, the highest level since 1995. If the BOJ maintains a hawkish stance, the stability of yen carry trades could also be tested, creating extra deleveraging pressure for global risk assets.