Skip to main content

Kimi K3 Hits Capacity Wall Just 3 Days After Launch

Kimi K3 Hits Capacity Wall Just 3 Days After Launch

Moonshot AI's latest large language model, Kimi K3, became a victim of its own success. On July 19, just three days after its July 16 debut, the company announced it was temporarily halting new consumer subscriptions. The reason? User demand had overwhelmed the available computing power.

Demand Skyrockets, Capacity Buckles

Kimi K3 is Moonshot's most powerful model yet, boasting 2.8 trillion parameters and a 1 million token context window. It's designed for heavy-duty tasks like long-range programming and end-to-end knowledge work. But the response from users was far beyond what the team anticipated.

"Within 48 hours of launch, user request volume significantly exceeded our estimates and approached the capacity limit of our current computing cluster," the company stated. To protect the experience for existing subscribers, Moonshot decided to stop accepting new paying users until it can expand its infrastructure.

A Rights Split and a Revenue Spike

Alongside the subscription pause, Kimi introduced a change: for future subscribers, the main Kimi rights and Kimi Code rights will be separated. The main rights cover Kimi Web, Kimi App, and Kimi Work. The move is meant to better allocate computing resources, especially as programming features become more resource-intensive.

Interestingly, the demand surge also brought a financial bright spot. On July 18, Kimi President Zhang Yuting revealed that the day after K3's launch, the company's annual recurring revenue (ARR) saw its largest single-day increase ever. The revenue spike underscores the market's hunger for advanced AI capabilities, even as it exposes the infrastructure challenges.

The Real-World Test

Kimi K3 uses a pay-as-you-go pricing model: 2 yuan per million tokens for input (cache hit) or 20 yuan (cache miss), and 100 yuan per million tokens for output. The model weights are also set to be open-sourced soon, according to Zhang.

This episode highlights a growing tension in the AI industry: as models move from PowerPoint presentations to production environments, the gap between user enthusiasm and computing capacity becomes glaring. For Moonshot, the immediate priority is clear: expand computing power as fast as possible. The company says it is accelerating expansion and will gradually resume new subscriptions as additional capacity comes online.

Key Points

  • Kimi K3 launched July 16 with 2.8 trillion parameters and a 1 million token context window.
  • New subscriptions paused July 19 due to demand exceeding computing capacity.
  • Existing subscribers unaffected; all resources prioritized for them.
  • Rights split introduced: main Kimi rights and Kimi Code rights will be separate for future subscribers.
  • Revenue spike: ARR saw its largest single-day increase on launch day.
  • Computing expansion underway; new subscriptions to resume gradually.