Kimi K3 Creator Eyes IPO. Moonshot AI Seeks to Raise $3 Billion

Kimi K3 Creator Eyes IPO. Moonshot AI Seeks to Raise $3 Billion

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
4. 9. 2026
2 minutes reading · 7 views
Listen to the article
Audio version of the article
Kimi K3 Creator Eyes IPO. Moonshot AI Seeks to Raise $3 Billion

Beijing-based Moonshot AI, which created the Kimi K3 model, has filed for an initial public offering and aims to raise three billion dollars through it. Before Moonshot could submit the application, it had to dismantle its offshore holding structure and relocate its headquarters to China, according to two sources. This was in response to new rules imposed by Chinese authorities, which forced it to dissolve its original structure involving a foreign holding company.

When trading will actually begin depends on obtaining approval. The date of the listing is not yet known. This year, Chinese technology companies are raising capital at a record pace. According to LSEG data cited by the Wall Street Journal, proceeds from Hong Kong IPOs exceeded 41 billion dollars by mid-August, more than double the amount raised during the same period the previous year. If Moonshot carries out its plan, it will be one of the largest initial public offerings Hong Kong has seen in recent years.

Rapid success in the AI world

Its founder is researcher Yang Zhilin, who studied at Carnegie Mellon University and Beijing’s Tsinghua University. The company gained recognition thanks to its Kimi series of models, which can process unusually long texts.

Further progress came in July this year, when Moonshot released the Kimi K3 model with 2.8 trillion parameters. The company claims it is the world’s largest open-weight model. The system became a global hit, and the company is now seeking to bring it to U.S. cloud platforms. It is negotiating revenue-sharing agreements with Microsoft, Amazon, and Google that would allow these companies to host the model and sell access to it to their own customers.

This has already been reflected in its financial results. Moonshot’s annual recurring revenue rose to 300 million dollars in June, up from 200 million dollars in April.

Suspicions of sanctions evasion

The U.S. administration has turned its attention to Moonshot. Allegations have emerged that the company trained the Kimi K3 model using Nvidia chips whose export to China is prohibited, and that it distilled its capabilities from Anthropic’s models. Treasury Secretary Scott Bessent said he could add the company to a trade blacklist. Moonshot has objected in particular to the claim that it distilled the model’s capabilities.

Nevertheless, the company’s valuation continues to rise. An ongoing funding round values the company at 50 billion dollars. However, its Chinese competitors are valued even higher, with DeepSeek worth around 74 billion dollars and Z.AI valued at 66 billion dollars.

Source: finance.yahoo.com and channelnewsasia.com

Advertisement

Content created with help from UpTier.

SEO and GEO on autopilot. UpTier’s multi-agent systems write and optimize content for search engines and AI answers.

Discover UpTier ↗

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

Ideogram 4.5 brings precise image editing and in-image text translationIdeogram 4.5 brings precise image editing and in-image text translation
Ideogram has released a model focused on successive image edits. It supports reference images, masks and high-resolution crops, with tools to change or translate text directly within an image.
3 min read
5. 10. 2026
OpenAI introduces text watermarks in its API, plans EU rollout for ChatGPT and CodexOpenAI introduces text watermarks in its API, plans EU rollout for ChatGPT and Codex
API customers can enable watermarks for select models. They are set to reach eligible ChatGPT and Codex users in the EU over the coming weeks. Access to the detector will initially be restricted.
3 min read
5. 10. 2026
Arena introduces image-model post-training with prompt-adherence checksArena introduces image-model post-training with prompt-adherence checks
Applied to FLUX.2-dev and Ideogram 4, the method combines human preferences with checks on individual prompt requirements. Arena reports improvements in automated evaluation and on its live leaderboard.
3 min read
5. 10. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok