Moonshot AI’s 2.8 Trillion Parameter Model Just Became the First From China to Top a Major Coding Benchmark
On July 16, the Chinese AI start-up Moonshot AI released its Kimi K3 large language model. Kimi K3 ranks third on Artificial Analysis’ Intelligence Index and became the first Chinese model to top a major coding leaderboard, Arena.ai’s Frontend Code Arena.
K3 is an open-weight AI model, free for anyone to download and modify. It supports the case for investors who’ve questioned the size of the investment allocated to the AI build-out. Combined capital spending by Microsoft (MSFT +0.03%), Amazon (AMZN +0.81%), Alphabet (GOOG -0.88%) (GOOGL -0.96%), and Meta Platforms (META +0.37%) for 2026 was recently estimated at over $725 billion, up from $410 billion last year.
A frontier-level model, made available for free download on the open-source AI platform Hugging Face, also puts pressure on premium-tier pricing from labs such as Anthropic and OpenAI.
Image source: Getty Images.
Incentives for cloud providers and frontier labs may be “misaligned”
Cloud providers such as Microsoft, Google, and Amazon sell compute capacity. Affordable tokens from a variety of model makers increase demand for that compute while reducing reliance on a select few, even if model margins compress.
Microsoft reported that its cloud business grew at the fastest pace in four years. Growth accelerated across all three companies, but only Microsoft expects to be free cash flow positive in fiscal 2027.
Today’s Change
(0.03%) $0.13
Current Price
$499.99
Key Data Points
Market Cap
Day’s Range
$498.73 – $505.18
52wk Range
$349.20 – $553.72
Volume
28.8M
Avg Vol
41.2M
Gross Margin
67.94%
Dividend Yield
0.71%
The frontier labs are more exposed to this risk. OpenAI and Anthropic lack the diversified profit centers that the hyperscalers enjoy, and need to continuously spend on training the next model, while recouping costs through premium pricing for the latest models. Frontier-level, open-weight models from China make this more difficult to achieve over the long run.
Within days of K3’s release, the debate over whether to regulate open-weight models intensified. Nvidia CEO Jensen Huang posted a letter on social media, signed by 25 companies, in support of open weights. Notably, it was Huang’s first posting on the X social media platform.
Then, more than 1,000 employees at leading AI labs, including their lead scientists, asked Washington for tools to “deliberately pace” AI development. As former Microsoft executive Steven Sinofsky noted: “It is their company. They could just stop.”
Revenue and volume are diverging
Anthropic and OpenAI are still growing at historic rates. According to third-party trackers, Anthropic’s revenue run rate has reportedly risen from $10 billion at the start of the year to over $70 billion, while OpenAI appears to be catching up based on recent remarks from its CFO.
On platforms like OpenRouter, which developers use to route queries to different models, token volume from U.S.-based models has fallen from roughly 70% to 30%, while volume from Chinese models has grown to around 60%. The premium models still capture the vast majority of spending, but the cheaper alternatives are taking share.
Enterprise spending on leading U.S. models won’t slow anytime soon. But if open-weight models continue to improve, they’ll become harder to dismiss over time.
The commoditization question will continue to evolve, and the next generation of models from Chinese labs will receive far more attention from users and regulators.
