The Day It Was Called "AI Communism"—The Full Release of Kimi K3's Weights Changed the Industry Atmosphere
On July 27th, Moonshot AI released all the weights (the model parameters themselves) of its massive model, "Kimi K3," for free and unlimited download. Anyone can download, run, and modify it on their own server. This incident has had far-reaching consequences, going beyond simply being another "China has released a new model." This article will examine how the industry is being shaken by this weight release.
An "Extraordinary" Scale of 2.8 Trillion Parameters
First, let's grasp the numbers. Kimi K3 is a Mixture-of-Experts (configuration that switches between multiple expert networks) model with 2.8 trillion parameters. Of its 896 expert networks, only 16 per token are actually active. This "sparse activation" design keeps inference costs down relative to the overall size of the model. It also boasts a massive context window of 1 million tokens, allowing it to load entire codebases and lengthy documents at once.
It is the largest open-weight model to date. According to charts presented by Moonshot itself, K3 significantly outperforms competitors such as DeepSeek (1.6 trillion), Xiaomi (1.02 trillion), and Alibaba (397 billion). While it lags slightly behind Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol in benchmarks, it reportedly outperforms their older models in coding and agent-related evaluations.
Stock Market Moves Immediately After the Announcement
Before discussing the technical details, let's look at the market reaction this announcement triggered. On July 17th, the day K3 was announced, the stock prices of Chinese AI companies plummeted. Z.ai fell by as much as 30% in the Hong Kong market, MiniMax by as much as 16%, and Alibaba by 4%. In the US market, Nasdaq fell 1.4% and NVIDIA by 2.2%. What's interesting is the divided opinion within the industry regarding this reaction. Patrick Moorhead, CEO of Moor Insights and Strategy, commented to CNBC that the market reaction was "a chilling overreaction, similar to the DeepSeek shock." The volatility in stock prices itself was attributed to other factors such as corporate earnings, oil prices, and geopolitical risks, suggesting that K3 is more accurately described as "one of the triggers" rather than the "sole cause." In fact, some technology stocks recovered their value the following week.
Demand So High It Had to Suspend New Registrations in Just 3 Days
Even more impressive than the technical performance is the sheer scale of the demand. Just three days after K3's launch, Moonshot stopped accepting new subscriptions. This was because demand completely exceeded the computing resources the company had secured.
Business figures have also shown rapid growth. It has been reported that Moonshot's daily revenue grew at least sixfold after the K3 launch. Monthly annualized revenue (ARR) rose from $200 million in April to $300 million in June, and the company is reportedly considering new fundraising at a valuation of $50 billion, and an initial public offering (IPO) in Hong Kong as early as this year.
Why "Weight Disclosure" Has Such Strategic Significance
This is the technical point I want to emphasize most. Disclosing weights is not simply a "generous free provision." It is also a shrewd strategy of transforming the computing resources held by other companies into a competitive advantage for the company.
Hosting massive models on their own servers requires enormous computing resources and operational costs. However, by disclosing the weights, the costs of procuring and operating the GPUs needed to actually run them become the responsibility of the companies and individuals using them. The ability for anyone to self-host K3 on their own GPU cluster creates a situation where "anyone with an NVIDIA GPU can become a direct competitor to OpenAI and Anthropic." This scenario poses a significant threat to closed-model companies like OpenAI and Anthropic.
The powerful expression "AI communism"
Interesting in this context is a comment posted on X by Dean W. Ball, OpenAI's head of strategy. He describes a world dominated by open weight models as a "dystopian hellscape" that could lead to "full AI communism."
While this is a powerful expression, it reveals how seriously closed-model companies perceive this move towards weight disclosure as a threat. The situation where free, high-performance software can rival paid, proprietary products could shake the very foundations of their business models.
What Engineers Should Consider
From an engineer's perspective, the combination of a 1 million token context window and cost-effectiveness through sparse activation is a design that deserves genuine praise. However, as AI researcher Nathan Lambert points out, the true significance of this incident lies in the possibility that the performance gap between closed and open models, and between US-made and Chinese-made models, has narrowed from the previously discussed 6-9 month gap to around 3-5 months.
Furthermore, behind the scenes of this weight disclosure, another case involving Moonshot (illegal distillation of Anthropic models and unauthorized access to regulated chips), which I previously discussed in this column, is also unfolding. With both technological capabilities and the surrounding geopolitical tensions intertwined, the competition surrounding open weight models is likely to intensify even further.