The Picture Today: The Leader Is Flat, the Side Channels Are Hot
If NVDA is the only lens, the AI trade looks like it is cooling. In the May 26, 2026 snapshot from the local market database, NVDA was down 0.7% on the day, down 3.8% over five sessions, and still about 9.3% below its 52-week high. It is not the strongest line in the tape.
Take the chain apart and the picture changes. MRVL is up 23.1% over five sessions, AMD 18.5%, ANET 12.7%, and GLW 11.3%. Hong Kong’s packaging and foundry names ran up 15-25% over the same window. And in the A-share watchlist, memory-interface and equipment exposure climbed too, with the leading memory-interface name up about 30% over five sessions.
So the better description is not that AI has faded. It is that leadership has changed. The GPU leader has stopped leading at the margin, and the market is looking for the next layer of scarcity: networking, ASICs, optical links, advanced packaging, foundry capacity, and memory.
This Is Not Broad Strength. It Is a Narrower Rotation.
The diffusion is not even. In the US AI side channels, MRVL and AMD are strong, while NVDA and CSCO are negative over five sessions. The A-share watchlist is split too: the leading memory-interface name is at a 52-week high, but other memory and component names are flat to down over five sessions. Hong Kong semis are also uneven: packaging and foundry exposure is strong, while parts of the chip-design pocket are down double digits over the same window.
The market is not buying every semiconductor line. It is selecting a smaller group of names that look closer to AI capex, supply bottlenecks, or narrative elasticity. At the index level, AI still looks strong. At the stock level, it is a concentrated reallocation.
The Other Side
The strongest counterargument is simple: this may be a short catch-up move, not a new trend.
NVDA has already had its move. Money looking for lagging side channels inside a hot theme is a normal late-stage behavior. It can keep spreading, but it can also push lower-certainty names too far in the short run. MRVL, AMD, ASMPT, and Hua Hong are high-elasticity names; what rises quickly can also give back quickly.
The second counterargument is that price diffusion is not the same as fundamental diffusion. MRVL, AMD, ASMPT, and related names are easier for the market to imagine as AI beneficiaries. But each company still has to translate the story into revenue, margin, orders, and customer qualification. A five-session move does not prove the earnings conversion has arrived.
The third counterargument is that macro has not fully joined. QQQ is up 3.4% over five sessions, but TLT is still down 1.9% over one month. BTC is down 6.2% over one month, and ETH is down 12.3%. This is not all risk assets warming at once. It is still one growth-equity line doing most of the work.
Closing
The thing to write down today is not simply that AI is up again. It is that AI leadership has changed.
NVDA is no longer the strongest short-term price expression. The side channels are. That can be healthy diffusion, or it can be high-beta chasing late in a crowded line. Both readings are possible. The difference will come from whether these side-channel names can translate price momentum into orders, revenue, and margins, rather than leaving the move inside the idea of the next bottleneck.
What to Watch
The following is an observation framework, not a trading signal.
- Whether NVDA and the side channels resync. If NVDA stays flat while MRVL / AMD / ASMPT keep accelerating, the line looks more like catch-up in the side channels. If NVDA joins again, the chain’s confirmation improves.
- Whether Hong Kong semis keep leading A-share segments. If the packaging and foundry names stay strong, money is leaning toward that exposure. If they fade, this may have been a short catch-up move.
- Whether A-share memory dispersion narrows. The memory-interface leader is strong while the rest of the memory pocket lags. That suggests the market is still selecting perceived leaders, not buying the entire memory chain.
- Whether macro assets confirm. If TLT, BTC, and ETH stay weak, AI equity strength remains a localized line rather than a broad cross-asset diffusion.
Data: local market database (Yahoo Finance daily), snapshot May 26, 2026, covering US / Hong Kong / A-share watchlists and major cross-asset instruments (SPY / QQQ / TLT / GLD / VIX / BTC / ETH). The local database covers the watchlist only, not a full-market scan.