Investor caution resurfaces
Some market voices are warning that semiconductor enthusiasm may be overstretched — they point to flatness in certain AI names as a sign that expensive pricing or data‑centre slowdowns could trigger demand pullbacks. (x.com)
The warning sign is not that artificial intelligence chip sales have collapsed. It is that some of the stocks tied most closely to the boom have stopped acting like every quarter can keep getting bigger forever, even after strong results. (cnbc.com) Nvidia reported fiscal fourth-quarter data center revenue of $62.3 billion on February 25, 2026, up 75% from a year earlier, and guided to about $78 billion in first-quarter revenue. Its shares still gave back an initial pop and then fell more than 5% the next day. (cnbc.com 1) (cnbc.com 2) That is what “priced for perfection” looks like in markets. When a company beats forecasts by billions of dollars and the stock still stalls, investors are saying the old expectations were already too high. (cnbc.com) The money behind this trade is not small. CNBC reported on February 6 that Amazon, Microsoft, Meta, and Alphabet were on track for combined capital spending close to $700 billion in 2026, much of it tied to artificial intelligence infrastructure. (cnbc.com) That spending mostly flows into data centers, which are warehouse-sized buildings packed with chips, memory, networking gear, and power equipment. The bet is simple: build enough computing capacity now, and customers will show up later. (ft.com 1) (ft.com 2) The caution is about timing, not just technology. If cloud companies slow construction by even a few quarters, chip orders can wobble fast because the supply chain was built for a surge, not a pause. (cnbc.com 1) (cnbc.com 2) There is also a second worry: more efficient software can reduce the number of expensive chips needed for a given task. On March 26, memory-chip stocks fell after Google disclosed research that investors thought could lower future demand for high-bandwidth memory, the premium memory used beside artificial intelligence processors. (cnbc.com) But the same market is still producing very strong numbers underneath the nerves. Taiwan Semiconductor Manufacturing Company said on April 10 that first-quarter revenue rose 35% from a year earlier to 1.13 trillion New Taiwan dollars, and March revenue alone rose 45.2%. (cnbc.com) Broadcom said on March 4 that its artificial intelligence revenue jumped 106% year over year to $8.4 billion, driven by custom chips and networking gear. Marvell said on March 6 that its data-center revenue for fiscal 2026 topped $6 billion, up 46%. (cnbc.com) (cnbc.com) Memory is still tight enough that Micron said in March it was benefiting from a supply crunch tied to Nvidia systems, and CNBC reported in January that the three main memory suppliers were struggling to keep up. That is not what a busted market looks like. (cnbc.com) (cnbc.com) So the argument on Wall Street right now is narrower than “boom” versus “bust.” The factories are busy, the orders are real, and the revenue is still climbing, but investors are no longer willing to pay any price for that growth if data-center buildouts slip, debt piles up, or software starts squeezing more work out of fewer chips. (cnbc.com) (cnbc.com)
