The Same AI Warning Crashed Chip Stocks But Pushed Cybersecurity Stocks To Record Highs
Triggered by Anthropic CEO Dario Amodeii's lengthy article last weekend titled "We Must Pace the Frontier," in which he called on the entire industry to consciously slow the acceleration of frontier model capabilities and warned that AI agents with stronger capabilities but equally inaccurate could cause losses worth hundreds of billions of dollars within 6 to 12 months.
Amodei's two concerns in the article—recurrent self-improvement and OpenAI's attacks on the open-source model platform Hugging Face—point to the same thing: the growth rate of AI capabilities is outpacing the pace of human supervision and protection.
For chip stocks, this directly challenges the pricing premise of "training scale only increases, never decreases." If frontier labs really slow down iteration, capital expenditure expectations for semiconductors, cloud infrastructure, and data centers focused on model training will have to be revised downward. Bernstein analyst Madison Rezaei pointed out that if AI training truly slows down, demand for data centers in remote areas may decline, since many facilities are originally designed specifically for model training.
But almost simultaneously, the market found another chain of reasoning: the recognition that AI capabilities are "dangerous enough" means that protecting them is a necessity. Evercore ISI analyst Kirk Materne made this logic clear in Monday's client report: regardless of how quickly or slowly AI agents are being promoted to the enterprise side, "these agents still need to be protected, governed, and monitored," so "even if AI training slows, the structural needs of cybersecurity and some infrastructure companies remain valid." He added a further assessment: a broader scrutiny of AI security will prompt companies to pay more attention to identity management, data governance, observability, and security.
The market immediately invested real money to support this logic. It's worth noting that the biggest gains weren't in general software stocks, but in the identity and runtime security segment: Okta (identity deposit and retrieval management) rose over 11%, and SailPoint, both in the identity security track, were named beneficiaries by Materne. This has shifted from "sentiment rally" to "structural beneficiaries."
By The Same Logic, A Day Later In The A-Share Market
This line was realized simultaneously in the A-share market. On September 14, all three major A-share indices closed in the red: the Shanghai Composite Index fell 0.07%, the Shenzhen Component Index fell 0.64%, and the ChiNext Index dropped 1.10%. However, the cybersecurity sector bucked the trend and became the biggest highlight of the day: the Wind Cybersecurity Index surged 4.31%, with sector turnover exceeding 22 billion yuan, main funds net inflowing 363 million yuan, Yongxin Zhicheng hitting the 20% daily limit, Renzihang, NSFOCUS, and AsiaInfo Security rising over 10%, and Tianrongxin hitting the one-word limit-up.
On Tuesday morning (September 15), sentiment continued to spread: Zhongxin Syke, Qiming Information, and Tianrongxin all hit the 10% daily limit, Yongxin Zhicheng rose over 9%, Haohan Deep rose over 8%, Nanling Technology rose over 6%, Digital Authentication rose over 5%, and Sangfor, Qi An Xin, and Dianke Cybersecurity followed suit. Analysts described this round of market movement as a "major reshuffle"—funds exited from the overcrowded AI hardware and semiconductor sectors and shifted toward cybersecurity and enterprise software.
