China's AI Strategy Shocks US by Releasing "Core Blueprint" After Initial Low-Cost Push

Chinese artificial intelligence startup Moonshot AI is making a aggressive push to secure global developers and corporate clients by releasing the model weights of its high-performance model, Kimi K3.

Following its low-cost pricing strategy, the company is widening the scope of openness to allow businesses to modify and operate the model directly, putting direct pressure on the closed AI model market dominated by US companies such as OpenAI and Anthropic.

According to Bloomberg News, Moonshot AI was scheduled to release the model weights for Kimi K3 on July 27.

Once the weights are released, developers can download the model directly to modify it for their specific purposes, or operate it within their own data centers and cloud environments.

Unlike major US AI firms that provide their models solely through APIs, Moonshot AI has positioned the ability for client companies to directly control their data and operating environments as a key competitive advantage.

Industry observers interpret this as a targeted strategy aimed at corporate markets with high security requirements, such as finance, the public sector, and manufacturing.

Kimi K3 is a large-scale AI model equipped with 2.8 trillion parameters and a context window of 1 million tokens.

Since its release this month, it has drawn significant market attention after receiving evaluations that it significantly narrowed the performance gap with leading US models in major evaluations.

The achievement has drawn further interest as it was accomplished under circumstances where Chinese firms face limited access to NVIDIA accelerators due to US advanced AI semiconductor export controls.

In fact, Moonshot AI's daily revenue is known to have surged by at least sixfold following the launch of K3.

Building on this momentum, Moonshot AI is pursuing a new funding round targeting a valuation of 50 billion dollars, and is considering an initial public offering on the Hong Kong stock exchange as early as this year.

Analysts interpret this release of model weights as a strategy that prioritizes expanding the developer ecosystem and distribution network over immediate model usage fees.

The approach aims to secure market share by increasing model utilization and redistribution, and subsequently expand revenue streams through APIs, enterprise services, and cloud hosting.

However, limits remain to the global proliferation of Chinese AI models.

Major US cloud providers such as AWS, Microsoft Azure, and Google Cloud do not yet officially offer Kimi K3.

US government checks also remain a variable.

A White House official claimed that Moonshot AI trained K3 using NVIDIA chips subject to US export controls, and also raised suspicions of so-called "distillation," which utilizes the output values of US corporate models.

The US-China AI competition is now expanding beyond the race to secure advanced semiconductors into a battle over who can attract more developers and companies to seize control of the AI ecosystem's standards.

Reported by Kim Minjeong | Video by Lee Eui-sun | Graphics by Yang Hye-min | Produced by SBS Digital News
※ Please note: This article was translated by AI and may contain errors.