The week began with two stories running in opposite directions on the same trading floor. Bank earnings arrived close to expectations, and memory chip shares fell hard enough to drag the technology indexes.
A widely tracked semiconductor index dropped 1.3%, while the Nasdaq 100 slipped about 0.4%. Strength elsewhere in the market absorbed part of the hit: the Dow Jones Industrial Average moved higher and the S&P 500 finished close to flat, leaving the session mixed rather than panicked.
The pressure on chips came from Asia first. ChangXin Memory Technologies, China’s largest maker of dynamic random access memory, said over the weekend that it had made progress on a new architecture designed to pack more data into a smaller area. Its president, Cao Kanyu, described plans for a 4F2 DRAM design, according to reports in Korea and China.
Investors read the claim as a threat to the pricing power of the three companies that dominate high-end memory. Samsung Electronics and SK Hynix each fell more than 7% in Seoul, and both have lost ground in every session since the Korean market opened on Monday, down roughly 15% and 13% respectively. Micron Technology closed down about 4% in New York the previous night.
CXMT is still widely seen as trailing the established players, which is what makes the reaction notable. A claim of progress on density, from a company with state backing and rising capacity, is enough to move valuations when the market has already begun asking how long the artificial-intelligence build-out can keep memory prices high.
That question was sharpened by Cathie Wood. Speaking to investors in Hong Kong, the founder of ARK Investment Management said the memory sector is prone to “vicious” boom-and-bust cycles, and that the arrival of more Chinese suppliers could make those cycles worse. ARK avoids the traditional memory makers for that reason, she said, while adding that she remains optimistic about artificial intelligence and does not accept that the sector has formed a bubble.
The broader backdrop did little to settle nerves. The International Monetary Fund warned this week that an abrupt slowdown in artificial-intelligence investment could trigger a correction in asset prices, a risk that lands directly on the companies that have spent two years adding capacity.
Against that, the bank results were steady enough to keep the wider market upright. JPMorgan Chase, Citigroup, Wells Fargo and Goldman Sachs all reported, opening a season investors are treating as a check on whether credit conditions and deal activity are holding up outside the technology trade.
The two halves of the session reflect a market that has become selective rather than fearful. Money is not leaving equities; it is leaving the part of the market where valuations assume the current rate of spending continues without interruption. Semiconductors sit at the front of that chain, which makes them the first place a doubt gets expressed.
Four concerns are being applied at once. Competition is rising, with Chinese capacity adding supply to a market that rewards scarcity. Interest rates have stayed higher for longer, raising the discount applied to distant profits. Valuations in the sector price in years of growth. And the cost of building the data centers that consume the chips keeps climbing.
The last point is the one memory investors find hardest to model. If the companies funding data centers slow their orders, the effect reaches DRAM and HBM pricing before it reaches earnings, because memory is priced in a spot market that turns faster than the contracts governing most of the industry’s revenue.
Wood’s argument about cycles is a long-standing one. Memory has produced enormous profits in upcycles and punishing losses when supply arrives late and demand cools, and the pattern has outlived every generation of management that promised to break it. What is new is the scale of the current expansion and the number of governments willing to fund additional capacity.
For the trading week, the practical result is a market that will take its direction from two sources. Results from the banks, and then from industrial and technology companies, will say whether the underlying economy is holding. The memory makers will say whether the product at the base of the artificial-intelligence trade is still scarce, or whether the answer is arriving from China rather than South Korea.
Some analysts argue the chip selloff is a valuation story rather than a demand story. Orders for accelerators remain booked well into next year, and the hyperscalers building data centers have not revised their spending plans. On that reading, the fall in memory shares reflects what investors are willing to pay for the growth, not whether the growth is happening.
Others see the CXMT claim as the first crack in a supply story that has carried the sector for two years. If Chinese output reaches the high end of the market in volume, the premium that memory makers charge today becomes harder to defend, and the earnings that justify current share prices arrive later than the models assume. Both readings can survive the same week of trading, which is why the argument is unlikely to be settled by the numbers reported this quarter.
