Samsung's record quarter didn't calm chip investors. It sharpened the question they were already asking: how much AI profit is real, and how much has already been priced as if the boom can only go one way?
Here's the number that tells you everything about the mood around AI chips right now: Samsung Electronics guided for Q2 2026 operating profit of 89.4 trillion won, about $58.5 billion, up 1,810.3% from a year earlier, according to Yonhap. The stock still sank 6.9% in Seoul after the July 7 update, Quartz reported. That's not a market glitch. It's a verdict.
Investors aren't punishing Samsung because the quarter was weak. They are punishing the whole trade because expectations have moved faster than the evidence. When a company can post one of the most extreme profit jumps in modern chip history and still lose more than $80 billion in market value in a day, you should stop treating the sell-off as noise.
The pressure returned this week. The Financial Times reported that South Korea's Kospi plunged 8% on July 28, triggering a temporary trading halt, as Samsung and SK Hynix dropped as much as 9% and 10% respectively. Japan's Nikkei fell more than 4%. Memory-chip names have become a direct bet on AI infrastructure spending, and when that bet shakes, Seoul feels it first.
China supplied the spark
The most dramatic new catalyst came from China. Reuters, citing The Information, reported on July 27 that a state-backed Shanghai company had begun manufacturing domestically developed immersion deep ultraviolet lithography machines, the most advanced lithography tools still available to Chinese chipmakers after export restrictions blocked access to ASML's extreme ultraviolet systems.
The details matter. The report said about five machines are expected this year for Semiconductor Manufacturing International Corp, Hua Hong Semiconductor and ChangXin Memory Technologies, with roughly 20 more planned for 2027. It did not identify the manufacturer by name. That distinction is important, because calling Shanghai Yuliangsheng Technology the producer goes beyond what the reported sourcing supports.
China building its own immersion DUV tools is real progress. It is not the end of ASML's advantage. The same Reuters summary said the Chinese system still lags on performance and reliability and needs more testing before mass production. ASML shares fell 4.6% after the report, but five machines is not 131 machines. Scale is the wall.
Also read: TravelAI bought Sonder's name out of bankruptcy and turned a dead hotel brand into an AI travel agent • Cursor ships Moonshot AI's record 2.8-trillion-parameter Kimi K3 on launch day • GPT-5.6 and Grok 4.5 independently toppled a 30-year-old math conjecture and neither result is peer-reviewed yet
Still, markets don't wait for perfect proof. Barron's reported that SanDisk fell 11% on July 27, while ASML dropped 5.8%, Lam Research lost 4.5%, Applied Materials fell 3.6% and KLA declined 3.4%. The VanEck Semiconductor ETF also slid more than 4%, according to market reports. One Chinese machine program was enough to hit both memory stocks and the equipment suppliers that helped define the old moat.
The larger fear is spending
Frankly, the China story looks like the match, not the fire. The fire is the cost of building AI at this scale. PIMCO said in May that consensus estimates for capital spending across the five largest hyperscalers had climbed to nearly $690 billion for 2026 and $870 billion for 2027. That is not a side budget. It is an industrial buildout.
Citi's latest numbers make the strain clearer. Investing.com, via Yahoo Finance, reported that Citi now expects Alphabet, Meta and Amazon to spend a combined $801 billion on capex in 2027, pushing all three into negative free cash flow in 2027 and 2028. Alphabet alone was raised to $308 billion. Meta was put at $205 billion. Amazon was lifted to $288 billion.
You don't need to be bearish on AI to see the problem. The market has already rewarded chipmakers as if the spending path is clean, long and profitable. But if customers have to borrow more, issue more equity, or accept weaker free cash flow to keep buying accelerators and memory, the next question is obvious: who captures the profit, and who just funds the race?
Samsung sits right in the middle of that argument. It benefits from memory demand, especially as AI servers consume more high-bandwidth and conventional memory. But its July earnings reaction showed the limit of good news when investors start worrying about peak margins, future supply and the durability of hyperscaler budgets. A huge number can still disappoint if the stock already assumed something even better.
This is why the sell-off deserves attention. Not because China's first domestic immersion DUV tools suddenly erase Western equipment leadership, and not because Samsung's record profit was secretly bad. The warning is simpler. When a sector needs perfect growth, perfect spending discipline and perfect geopolitical control to justify its price, one credible challenge can pull the whole structure lower.
AI chips are still a great business. They are no longer a free pass.
