Moonshot AI · 25

FT: Dark Side of the Moon is restructuring the equity structure and introducing state-owned shareholders to obtain regulatory approval to go public in Hong Kong

Comparatively, according to the Financial Times, the Chinese AI startup Moonshot AI (Moonshot AI) is adjusting its corporate structure and bringing in a number of investors with a state-owned background to obtain approval from the regulatory authorities to go public in Hong Kong. The Kimi K3 model under Dark Side of the Moon recently closed the performance gap with Anthropic's leading model and was welcomed by developers. The company plans to raise capital through an initial public offering (IPO) for the next phase of large-scale model development and business expansion. Prior to listing, Dark Side of the Moon may require dismantling the existing red-chip structure. Since its core business was previously controlled by overseas entities and raised in US dollars through overseas investors, the Chinese regulatory authorities imposed restrictions on the overseas listing of some technology companies with overseas structures, causing many AI companies, including Dark Side of the Moon and StepFun (StepFun), to suspend IPO preparations. According to reports, Dark Side of the Moon changed the entity in China from a limited liability company to a limited liability company last week, which is an important step in advancing its listing. The company is currently coordinating with investment banks and a team of lawyers to resolve the issue of overseas investors' shareholding transfers. An investor revealed that The Dark Side of the Moon has recently been notified to shareholders, that the restructuring of overseas entities is still being promoted, and it has not yet been decided how to transfer the rights and interests of overseas shareholders to domestic entities. Possible solutions include overseas investors setting up shares held by domestic investors, or selling overseas shares first and then re-subscribing for domestic shares, but they all involve a lengthy approval process. In terms of financing, Dark Side of the Moon recently completed two rounds of financing. One round was valued at around US$30 billion, and the valuation is expected to reach US$50 billion after completion. The company previously attracted state-owned institutions such as the National Artificial Intelligence Industry Investment Fund to participate in the investment. The latest shareholder list also includes investors under the National Social Security Fund, Shanghai and Guizhou Local Government Guidance Funds, and People's Daily. On August 3, earlier market news stated that the Dark Side of the Moon plans to submit an IPO application in Hong Kong as early as this month, or raise about 3 billion US dollars. In response, Dark Side of the Moon responded that the news was untrue.

14d ago

Dark Side of the Moon starts a $50 billion pre-IPO round of financing, and the amount was “suddenly tight” after K3 was released

Comparative news, according to an exclusive report by the “Science and Technology Innovation Board Daily”, Moonshot AI (Moonshot AI) Series G (Pre-IPO) financing has officially started, and the current round of valuation has reached 50 billion US dollars. People familiar with the matter revealed that after the release of the K3 model, the amount of financing was “suddenly tight,” and several institutions have registered investment intentions. Participants are required to complete the payment by August 15. To directly enter the shareholders' register, the investor's management scale must be at least 500 million US dollars. The company's previous Series F financing ($35 billion valuation) was fully settled last week, and actually raised $3.5 billion. The company was valued at around $18 billion in April of this year, nearly tripling in three months. Sources said that the core driver of this round of valuation jumps was the release of the K3 model. On July 27, Dark Side of the Moon officially opened the full Kimi K3 model weight, making it the world's first 3 trillion parametric open source model to launch. Currently, the company's API revenue accounts for more than 70% of total revenue, and ARR has grown several times since K3 was released. In response to recent reports that “an IPO application will be submitted to the Hong Kong Stock Exchange as soon as this month,” people familiar with the matter responded that “the news is untrue; Series G financing is still ongoing.” As of press release, Dark Side of the Moon has not officially responded to this news.

17d ago

Foreign media: The Dark Side of the Moon may use Nvidia Blackwell chips to train Kimi models. The effectiveness of US export controls is being questioned again

Comparatively, according to The Information quoting people familiar with the matter, Moonshot AI (Moonshot AI) is seeking to obtain more Nvidia Blackwell AI chips to train the next Kimi K4 model. The report said that Dark Side of the Moon also used Nvidia chips when training Kimi K3 before, although US export controls prohibited the export of advanced AI chips to China. The report pointed out that if the relevant news is true, it means that Chinese AI companies can still obtain Nvidia's latest generation AI chip resources, and the US may once again face doubts about the effectiveness of advanced AI chips. Analysts believe that this will further exacerbate the concerns of the US market that Chinese AI models will quickly catch up with leading models such as OpenAI and Anthropic.

23d ago

Research: Kimi K3 scored 32.2% in the cyber attack test, more than 40 percentage points behind the US frontier model

Comparatively, a joint evaluation by the British Institute for Artificial Intelligence Security and the US Artificial Intelligence Standards and Innovation Center showed that Moonshot AI's Kimi K3 can assist in the execution of exploits and simulate cyber attacks without substantial blocking, yet its overall cybersecurity capabilities still clearly lag behind the US leading model.

29d ago

US Office of Technology Policy Director Accuses Moonshot AI of Distilling Anthropic Fable to Develop K3

Comparing news, Michael Kratsios, director of the US White House Office of Science and Technology Policy, wrote that the available information shows that Moonshot AI has distilled Anthropic's Fable model by building a large-scale internal platform to develop its K3 model, and has switched access methods several times to evade detection. He also pointed out that Moonshot AI has purchased an GB300 server and visited a GB300 device in Thailand, which may be used to train AI models. Kratsios said that the US side supports legal distillation and open innovation, but opposes stealing American technology through large-scale clandestine industrial distillation. Moonshot AI has yet to publicly respond to this.

30d ago

Kimi employee responds to White House allegation: 15 days to distill cutting-edge models? Then you have to file a Guinness

Comparing news, Michael Kratsios, director of the US White House Office of Science and Technology Policy, wrote that Moonshot AI (Kimi) referred to Anthropic's Fable when developing the K3 model and established an internal platform capable of large-scale distillation of American models. In response, Moonshot AI team member Randy Xia directly retweeted the response: Fable was only released on July 1, and K3 went live on July 15. According to this timeline, Moonshot is equivalent to completing model research, distillation, training, and release within 15 days. He joked: We trained a brand new cutting-edge model in just 15 days, which simply set a Guinness World Record.

30d ago
Domestic model Zhimi K3 quickly launched Bai B.AI to experience the world's top AI products in one stop

Domestic model Zhimi K3 quickly launched Bai B.AI to experience the world's top AI products in one stop

Recently, Kimi K3, the latest flagship model of the Dark Side of the Moon, which is regarded by the industry as the “light of the big domestic model,” officially launched White B.AI. Bai B.ai became one of the first platforms to complete Kimi K3 API access, and is now open to developers. After its release, Kimi K3 quickly became the focus of attention in the global AI industry. The test results of Artificial Analysis, a world-renowned independent AI model evaluation platform, showed that the Kimi K3 Comprehensive Intelligence Index ranked among the top three in the world and performed well in multiple agent task evaluations. Elon Musk also left a “Impressive (Impressive)” comment under the relevant review post. The Financial Times also recently published an article focusing on Kimi K3, believing that it has further narrowed the gap between China's big model and the world's cutting-edge level. Through Bai B.AI's unified API, existing developers can quickly call Kimi K3 without registering a separate account or maintaining a new key, and can flexibly switch and use it in combination with mainstream models at home and abroad. The domestic big model Zhiguang was quickly launched. The first batch of Bai B.AI was opened and launched as a representative of the domestic big model that has received much attention recently. After the launch, the Kimi K3 quickly entered the developer testing and application stage. After Bai B.AI completes the first batch of connections, existing users can directly call it through the original account and API system. There is no need to register or reconfigure the interface separately, and it is also easy to manage keys and bills of different models uniformly. Under Bai B.AI's unified API system, Kimi K3 can also be used with other mainstream models at home and abroad. The development team can flexibly switch models according to task difficulty, response speed, and cost requirements, and can also assign different models to different parts of the same business process. Compared to connecting each model one by one, this method is easier to carry out effect testing, horizontal comparison, and subsequent deployment. According to Bai B.AI product documentation, Kimi K3 provides 1,048,576 token context windows, supports text, image, and video input, and focuses on long-term programming, complex knowledge work, and agent tool calls. Large code warehouses, lengthy documentation, continuous research tasks, and workflows requiring multiple rounds of execution and tool collaboration are its main application directions. The first batch of Bai B.AI is open and launched, which also allows these capabilities to enter developers' actual testing and product development processes more quickly. Source: Artificial Analysis In the Artificial Analysis test, Kimi K3 scored 57 points; its GDPVal-aa v2 Elo score reached 1668, ranking first in the AutomationBench-AA automated workflow evaluation with a score of 53%. When nine comprehensive tests were completed, Kimi K3's output token usage was reduced by about 21% compared to the previous generation K2.6. While improving capabilities, it also improved the execution efficiency of complex tasks. These results show that the market's evaluation of large models is becoming more realistic. In addition to the list results, developers are also paying more and more attention to whether the model can read a large amount of material, use external tools, and actually complete a lengthy task. Kimi K3's performance in agent and knowledge work reviews is an important reason why it quickly gained attention after launch. Continuing to expand the domestic model matrix, Bai B.AI helps China's big models go global. In addition to the KIMI series, Bai B.AI also currently supports various large domestic models such as MiniMax, GLM, and DeepSeek, covering different application directions such as reasoning, programming, long context, and agents. The addition of Kimi K3 has further improved the platform's domestic model service matrix, providing global developers with more options that can be directly accessed and compared horizontally. In recent years, the competitiveness of large domestic models is no longer limited to the Chinese scene. The high performance/price ratio, open weight strategy, rapid iteration ability, and continuous investment in long-term context and agent tasks are attracting more overseas developers. An analysis recently released by OpenRouter shows that at the beginning of June 2026, the Chinese model's share of token usage on its platform surpassed that of the US model, and large domestic models such as DeepSeek were the main drivers of this round of growth. The Financial Times also pointed out in a recent report that Chinese models, including DeepSeek and Moonshot AI, are being rapidly adopted by developers, and their degree of openness and cost advantages are driving more global companies to try out the Chinese model. After the release of Kimi K3, it was affected by...

31d agoburnking#B.AI

J.P. Morgan chanted Zhi Spectrum's excessive pullback and maintained an increase rating, boosting the stock price to rebound by nearly 37% in a single day

Comparatively, at a time when the AI chain in the US stock market has fluctuated drastically recently, the market once saw Kimi K3 as a new DeepSeek moment: a low-cost, high-performance, and open-weighted Chinese model, which once again raised investors' concerns about the return on US AI capital expenditure, model pricing power, and computing power requirements. WSJ says Moonshot AI's new model has heightened chip investors' anxiety. J.P. Morgan said in the latest China Artificial Intelligence Industry Report that Kimi K3 has indeed changed the valuation framework for leading Chinese model companies. According to the report, after K3 was released, the market doubted the long-term leadership of companies such as Smart Spectrum and MiniMax, so its valuation ratio was lowered from about 30 times the previous P/ARR forecast for 2030 to 20 times. However, J.P. Morgan believes that the correction of Zhi Spectrum's stock price by more than 50% already reflects this level of pressure excessively. The bank maintained the smart spectrum increase rating and lowered the target price from HK$2,400 to HK$1600; MiniMax maintained a neutral rating and lowered the target price from HK$240 to HK$160. The core judgment of the report is that K3 has shortened the GLM-5.2 leading window, yet it has not excluded intellect spectra from the Chinese frontier model camp. According to J.P. Morgan Chase, GLM-5.2 is still one of the leading production models in China, and Smart Spectrum will re-verify competitiveness through GLM-5.3 and 2T+ flagship models in the next few months. If new models can re-enter the cutting edge array in coding, inference, and intelligence tasks, intellect will still have the ability to sustain commercial growth. More importantly, commercialization of the Chinese model is still in its early stages. J.P. Morgan estimates that Smart Spectrum's latest indicative ARR is about 1 billion US dollars, DeepSeek is about 500 million US dollars, MiniMax and Kimi are about 300 million US dollars each, and China's leading independent model suppliers total about 2.1 billion US dollars; in contrast, Anthropic ARR has reached about 69 billion US dollars. This means that the revenue scale of Chinese model companies is still significantly small, and future growth space has not been reduced by the advent of Kimi K3. Kimi K3 also left another signal to the market: stronger domestic models are trying to prove their capabilities at higher prices. The report mentioned that the price of the K3 API is significantly higher than the previous generation K2.7 Code, reflecting its improved capabilities in high-value tasks such as coding. If customers are willing to pay for stronger reasoning, longer context, and programming skills, the business model of Chinese model companies will shift from low price competition to ability pricing. Therefore, the tone of this report is not pessimistic. Kimi K3 brought about a revaluation and amplified the US AI chain's concerns about cheaper Chinese models; however, according to J.P. Morgan, this shock also proved that the Chinese model layer is approaching the global frontier. For Smart Spectrum, the next round of model launches will be a key window for repricing.

32d ago

Dark Side of the Moon: User demand far exceeds expectations, and resource pressure is being relieved through model performance optimization and computing power supplementation

Comparing news, today, Huang Zhenxin, head of the B-side enterprise business of Moonshot AI (Moonshot AI), revealed to the “Science and Technology Innovation Board Daily” reporter that after the launch of K3, user demand far exceeded previous expectations. There was a situation where resources were tight across the platform and registration of new users was limited. The team had made sufficient preparations and still did not match the needs. The team clearly prioritizes the user experience of old paying users, will not fully liberalize new user registrations for the time being, and will not sacrifice the old user experience in exchange for short-term user growth. Currently, resource pressure is being gradually relieved through model efficiency optimization and computational power supplementation. The specific recovery time will be announced separately according to the resource reserve situation. (Financial Services Association)

32d ago