According to CoinWorld, Scott Wu, co-founder and CEO of Cognition, pointed out on the Founders podcast that some companies have gone "overboard" in evaluating programmer performance by tracking the amount of AI tokens employees consume. He argued that actual completed work orders and output should be the primary metrics. This phenomenon, known as tokenmaxxing in Silicon Valley, involves employees reflexively and excessively using tools like Claude, Codex, or Cursor to gain an advantage on internal AI usage dashboards. Jacob Lauritzen, CTO of legal AI startup Legora, also criticized the practice of displaying token usage in performance reviews, leading to employees artificially inflating their data. Andrew Feldman, CEO of Cerebras Systems, stated at a Bloomberg conference that providing employees with unlimited tokens is foolish and advised companies to focus on cost-effectiveness when using AI tools, opting for low-cost open-source models rather than abusing top-tier closed-source models.

According to Bijie.com, Matrix Partners China (02477.hk) has officially launched its artificial intelligence and computing power service platform, "NovaxAI." This platform marks the formal implementation of the group's AI strategy upgrade plan, focusing on cloud GPU computing resource leasing and scheduling services, computing optimization services, model gateways, and AI intelligent operation and maintenance. By integrating global computing resources and providing unified access and AI-driven operation and maintenance capabilities, it helps enterprises and developers train models, deploy inference tasks, and schedule cross-regional computing resources to achieve reliable and scalable application operation.

According to Bijie.com, South Korean President Lee Jae-myung announced that he views AI as a "total war," planning to invest $518 billion to address the global AI competition. He warned that "speed is the only way to survive." Samsung and SK will each build two new chip factories as part of a total 800 trillion won (approximately $518 billion) AI drive. South Korea also plans to double its DRAM capacity by 2035, build 18.4 GW of AI data centers, and spend 30 trillion won (approximately $19.4 billion) on next-generation memory. Following the announcement of these plans, the Korea Composite Stock Price Index (KOSPI) erased a 3.4% drop.

According to Bijie.com, Jia Yangqing has left Nvidia after only one year, citing unsatisfactory performance of the DGX Lepton project he was in charge of. Semiannalysis reports that Nvidia CEO Jensen Huang was dissatisfied with the project's operation, and the two sides had disagreements regarding product execution and open-source commitments. Nvidia acquired LeptonAI in April 2025 and rebranded it as DGX Cloud Lepton in June of the same year, positioning it as a unified AI computing power marketplace. Although Nvidia promised to open-source LeptonAI's core software platform by 2026, this promise has not yet been fulfilled. Jia Yangqing has been confirmed to have been invited to serve as an advisor to GPU cloud service provider Hyperbolic, providing technical support.