Moonshot AI pauses Kimi K3 subscriptions as demand strains compute capacity

Chinese artificial intelligence startup Moonshot AI has temporarily paused new consumer subscriptions after overwhelming demand for its recently launched Kimi K3 model stretched its computing capacity.

The move comes as the company seeks fresh funding and prepares for a potential initial public offering (IPO) in Hong Kong.

Two people familiar with the matter told Reuters that Moonshot is unwinding its existing offshore corporate structure ahead of a planned Hong Kong listing.

The sources said the restructuring forms part of the company's IPO preparations.

The company has also engaged financial advisers, including Goldman Sachs and China International Capital Corp (CICC), to discuss the proposed listing.

The people, who requested anonymity because the discussions are confidential, said the IPO timetable remains subject to change.

Funding efforts continue as valuation climbs

Founded in 2023 by Yang Zhilin, an AI researcher who pursued doctoral studies at Carnegie Mellon University, Moonshot has emerged as one of China's closely watched AI startups.

According to a fundraising teaser seen by Reuters, the company raised more than $2 billion in May from investors including Meituan, China Mobile, and CPE.

The fundraising brought Moonshot's total historical funding to more than $5.5 billion.

The same fundraising document showed the company has since started seeking up to $2 billion in additional capital. It also indicated that Moonshot reached a valuation of $30 billion in June.

Growing demand for increasingly advanced AI models has strengthened investor interest in China's leading AI startups.

However, the trend has also increased the need for significant investment in computing infrastructure.

Several competitors, including DeepSeek, have also sought external funding to expand computing capacity as Chinese AI developers compete with US rivals.

Demand for Kimi K3 exceeds forecasts

Moonshot said on Sunday that Kimi K3 attracted significant user interest immediately after its launch, creating "unprecedented compute challenges."

The company said user requests during the previous 48 hours had risen sharply beyond its expectations and were approaching the limits of its existing computing clusters.

As a result, Moonshot said it would immediately suspend new consumer subscriptions while reserving available computing resources for existing paid subscribers.

The company said current paying users would not be affected by the capacity shortage.

Moonshot also announced plans to divide future memberships into two separate subscription options, including a dedicated coding plan.

The company said the change is intended to better align computing resources with different user requirements.

Capacity constraints follow the flagship model launch

The computing bottleneck comes shortly after Moonshot introduced Kimi K3 on Friday.

The company described the model as a 2.8 trillion-parameter system and said it is the world's largest open-weight AI model.

"Kimi K3 has received far more love than we expected, and our GPUs are feeling it," Moonshot said on X.

The company added that new subscription slots would return in stages as additional computing capacity becomes available.

Kimi K3 is designed for coding and agent-based tasks.

These workloads typically require repeated model calls and greater inference capacity, making them more expensive to operate at scale.

Although open-weight models allow users to download and customise the underlying system, analysts said relatively few users are likely to host a model of Kimi K3's size because of the hardware costs involved.

Moonshot said Kimi K3 delivered competitive performance against leading US models in certain technical tasks.

The company added that independent evaluations have also indicated strong performance.

Chinese AI race intensifies

The Kimi K3 launch comes as several Chinese AI companies introduce increasingly capable models at lower costs.

Alibaba, which is an investor in Moonshot, announced on Sunday that its 2.4 trillion-parameter Qwen3.8-Max-Preview model had launched on its AI platforms ahead of a planned open-weight release.

At the same time, US export controls on advanced Nvidia chips have made access to computing power an increasingly important constraint for AI developers seeking to scale their models.