Huang Renxun said one sentence: Nvidia’s chip sales next year will be double this year. After saying it, the stock rose 2% in pre-market trading.

It sounds like he’s talking about the industry, but in reality he’s talking about expectations. The moment expectations are voiced, they’re already part of pricing—like when a host in the countryside is having guests over. The host first announces the number of tables, and once that’s done, the guests feel too embarrassed to come empty-handed.

So what matters isn’t the words “double.” It’s the customer base: who exactly is paying up first for future capacity. If it’s a few cloud providers that have already made money, then demand is spilling over. If it’s new players propped up by financing, then it’s just inventory being moved to a new place.

What gets squeezed is pricing power. If sales double, capacity must double as well; once capacity is rolled out, discounts become inevitable, and gross margins must slide. Expansion cycles are often between 12 and 18 months. By the time capacity lands, demand may not necessarily still be sitting there.

In closing, two idle notes: look to see whether next year’s actual shipments reach double this year’s, and whether the gross margin rate falls while revenue grows. The former is demand; the latter is competition.

If you want to track these metrics across the compute power value chain together with upstream orders, scan the QR code below 👇 to join the group. The tracking sheet in the group is updated every week.

#半导体 #算力 #AI