U.S. technology stocks are collectively under pressure. Is the market replaying the DeepSeek-style shock script?

The tech sector of the US stock market continued to decline this week, with the apparent catalyst being the possibility of OpenAI delaying its IPO. However, analysts point out that the low-cost AI model from KNOWLEDGE ATLAS may be a deeper threat - it not only affects OpenAI and Anthropic's corporate clients, but also shakes the market's optimistic expectations for AI infrastructure spending. Earlier, Wall Street news reported that OpenAI is considering delaying its initial public offering plans, one reason being the lackluster performance of SpaceX's stock price after its listing, compounded by recent volatility in tech stocks. This news became the direct trigger for this round of selling. However, according to Jefferies strategist Christopher Wood citing industry insiders in a client report, the GLM5.2 model launched by KNOWLEDGE ATLAS has "performance almost equivalent to Anthropic, but the cost per token is only one-fourth of the latter's." He characterized the past week as "another DeepSeek moment." In Friday's US stock market trading, Micron Technology's stock price fell by over 7%, while AMD and Intel both dropped by over 4%, and Oracle fell to a new low after a cumulative decline of 19% over the past five trading days. The low-cost models from KNOWLEDGE ATLAS are disrupting the frontier AI pricing system KNOWLEDGE ATLAS's GLM5.2 model is considered the core variable behind the pressure on the tech sector this time. According to Jefferies analysts in a report on Thursday, this model not only closely matches Anthropic's flagship product in performance, but also boasts a level of privacy protection comparable to top models. Morgan Stanley traders stated on Thursday that the new model from KNOWLEDGE ATLAS shows "extremely impressive coding capabilities" and pointed out that this model poses a direct competitive threat to the enterprise market. They stated: "The trend of enterprises and hyperscale cloud service providers turning to lower-cost models is more like a recalibration of AI payment willingness rather than a decline in AI demand." Deutsche Bank analyst Jim Reid also mentioned similar pressure in a client report on June 18, stating that DeepSeek's V4-Pro model performs similar to Anthropic's flagship product Claude Fable 5 in about 90% of daily tasks but costs only about 1.5% of the latter. Enterprise AI spending patterns are accelerating restructuring AI pricing pressure has triggered a chain reaction on the enterprise side. In recent months, many companies have started cutting AI spending or exceeding their original budgets, making the token pricing of high-end AI models increasingly a pain point for businesses. Jefferies analysts further pointed out that if low-cost models have reliable privacy protection, enterprises are motivated to migrate AI workloads back from cloud service providers to their own servers, fundamentally changing the investment logic of AI infrastructure. "The demand structure is clearly shifting towards low-cost models," Morgan Stanley traders wrote in their report. This trend poses valuation pressure on individual stocks related to Nvidia's supply chain, cloud computing platforms, and data center construction. Cloudy prospects for OpenAI and Anthropic IPOs The rumors of OpenAI delaying its listing, together with intensifying competition, have made market expectations for the valuation of these two leading AI companies more cautious. Analysts are concerned that if enterprise clients continue to migrate to low-cost models, the revenue growth prospects for OpenAI and Anthropic before their IPOs will be eroded, and a potential price war between the two is more likely to compress their valuation space before listing. The continued emergence of lower-cost, performance-equivalent open-source AI models is seen as a more profound threat to the proposed IPOs of these two companies and the entire tech sector. The previous expectation of "astronomical growth" in AI infrastructure spending is now facing pressure for reassessment. This article is a reprint from "Wall Street News," written by Yang Chen, GMTEight editor: Zhang Jinliang.