Tech stocks are selling off sharply Friday, and the culprit isn’t just disappointing earnings or a macro shock — it’s a new Chinese artificial intelligence model that analysts are calling “another DeepSeek moment.” Z.ai, formerly known as Zhipu AI and listed in Hong Kong, launched GLM5.2, a model that Jefferies strategist Christopher Wood says is “almost equal to Anthropic at just a fraction of the price.” Specifically, GLM5.2 costs just one quarter of Anthropic’s Claude per token, putting major pressure on the valuations of both OpenAI and Anthropic ahead of their anticipated IPOs. The Nasdaq fell sharply on Thursday as the news spread, and the selling continued into Friday morning.
The timing is brutal for U.S. AI investors. OpenAI is already reconsidering the timing of its IPO following the lukewarm performance of SpaceX shares after its debut. Now, with a Chinese competitor delivering frontier-level capabilities at 25 cents on the dollar, enterprises have a compelling financial reason to switch providers. Morgan Stanley traders noted GLM5.2 has “very impressive coding capabilities” — the exact use case where businesses have been burning the most AI budget. Deutsche Bank’s Jim Reid had already noted last week that China’s DeepSeek V4-Pro handles “roughly 90% of everyday tasks at about 1.5% of the cost” of Anthropic’s leading Claude Fable 5 model. The era of paying a premium for U.S. frontier AI is facing its stiffest test yet. Jefferies also noted that GLM5.2 likely offers privacy protections comparable to frontier U.S. models — removing another barrier to enterprise adoption.
For retail investors, this selloff is both a warning and a potential reset opportunity. The immediate pressure falls on companies whose valuations are tied to the assumption that U.S. AI model providers maintain premium pricing power — including the broader AI software layer that’s been priced for hypergrowth. However, the deeper infrastructure of AI — data centers, power chips, cooling systems, and networking — remains essential regardless of which model wins the cost war. The “picks and shovels” layer of AI infrastructure is more insulated from this competitive dynamic than the model providers themselves. Investors should use this volatility to audit their AI exposure: the companies selling computing power and infrastructure may weather this storm far better than those selling AI intelligence at a margin.
