Kimi K3 Goes Viral in 48 Hours: GPU Overload Forces Membership Suspension, Daily Revenue Up 6x

Kimi K3, the world's first open-source 3 trillion parameter model from Chinese AI company Moonshot AI, has been so overwhelmed by user demand that the company suspended new consumer subscriptions just 48 hours after its July 18 launch. On the evening of July 19, the team announced an immediate halt to new user sign-ups, reserving all available computing power for existing subscribers.

Kimi K3 demonstrated strong benchmark performance, topping the Frontend Code Arena with 1679 points ahead of Fable5 and GPT-5.6 Sol, and achieving the highest score among all evaluated models on the SWE Marathon benchmark at 42.0 points. On DeepSWE it scored 67.5 and on Program Bench it reached 77.8.

Developers quickly turned K3 into a creative playground, building web apps, games, 3D scenes, and interactive applications. Some users created open-world games they described as “unbelievable” just months ago, while others pitted K3 against mainstream overseas models in head-to-head comparisons, often praising K3's scene completeness and visual detail.

The surge in request volume pushed GPU clusters to their limit. Each model inference requires significant GPU compute, and as concurrent requests grew beyond capacity, response times degraded and queues formed. The team had no choice but to halt new subscriptions to protect the experience of existing users.

Once subscriptions reopen, Moonshot AI plans to split its membership system into two tiers: Kimi's main features (Web, App, and Work) will be separated from Kimi Code, reflecting the vastly different compute requirements of conversational AI versus coding agents. This restructuring allows more precise resource allocation and creates flexibility for future pricing adjustments.

The market impact has been dramatic. Bloomberg reports that daily revenue has grown at least 6x since K3's launch. Moonshot AI's annualized recurring revenue reached $300 million in June, up from $200 million in April — a $100 million increase in just two months. The company is also pursuing a new funding round at a potential valuation exceeding $30 billion.

The Kimi K3 phenomenon reveals a counterintuitive dynamic in the AI industry: cheaper, more capable models don't reduce compute demand — they expand the user base and enable more complex applications, pushing GPU demand even higher. As models enter the trillion-parameter era, the compute bottleneck may be shifting from training to inference, posing a fresh challenge for the entire industry.