Chain Reaction! After Kimi K3 Launch, Anthropic's Year-End IPO Valuation Expectations Drop 11%; Moonshot AI Urgently Suspends New C-End Subscriptions

On Thursday last week (16th), Moonshot AI unveiled its next-generation large language model, Kimi K3, triggering significant shifts in the global AI competitive landscape and market expectations. Data from prediction platform Polymarket the following day showed that after Kimi K3 launched with 2.8 trillion parameters, a 1 million token context length, and an open-source strategy, the market's probability of Anthropic reaching a $1.5 trillion valuation by year-end dropped to 67%—an 11 percentage point decline from previous estimates.

Analysts noted that the emergence of high-performance open-source models is prompting the market to reassess the competitive advantages and valuation potential of major AI companies, intensifying industry competition.

User demand quickly pushed computing capacity to its limits. On Sunday (19th), Moonshot AI announced that user request volumes had exceeded expectations since the K3 launch, bringing existing compute clusters close to their processing limits within the past 48 hours. To ensure service quality for existing subscribers, the company has immediately suspended new consumer (C-end) subscriptions and is prioritizing compute resources for current members. Services will gradually resume once capacity expansion is complete. The team is accelerating infrastructure scaling efforts.

Alongside this adjustment, Kimi announced that future new subscriptions will separate Kimi core benefits (Web, App, Work) from Kimi Code benefits, enabling more precise allocation of computing resources.

Currently, Kimi K3 operates on a pay-per-use basis: input costs RMB 2 per million tokens (cache hit) or RMB 20 (cache miss), while output costs RMB 100.

Zhang Yutong, President of Moonshot AI, revealed on Saturday (18th) that the K3 model weights will be open-sourced soon. Additionally, the company recorded the largest single-day increase in annualized revenue (ARR) last Friday, demonstrating strong product-market fit. However, short-term computing bottlenecks remain a critical challenge to resolve.