Moonshot AI dropped its biggest bet yet at the World AI Conference in China this week. Elon Musk lauded the model as "impressive," "while OpenAI's head of strategy, Dean Ball, warned that the open-weight release "pushes the tech world closer to" full AI communism."
Kai-Fu Lee, the veteran entrepreneur and investor, drew a sharp comparison. He said, "OpenAI and Anthropic will be the iPhone, the Chinese models will be the Android," and assessed that Chinese models lag US rivals by about six months. His argument: US closed models will generate the largest profits, while Chinese models will capture a broader user base.
Moonshot did not always look like a winner. The company got a jolt in early 2025 when DeepSeek released its groundbreaking R1 model on the exact same day as Moonshot launched its closed-source model, K1.5. The timing took founder Yang Zhilin by surprise, according to sources familiar with the matter, leading to an urgent meeting shortly after the Lunar New Year break. At that meeting, Yang resolved to release the company's next top model as open-source, mimicking DeepSeek's strategy, and urged his team to maintain a long-term outlook.
Get the market news that matters in a five-minute read with Market Briefs, our free daily newsletter
The first major impact of DeepSeek on Yang was a mass departure of users. The company had invested heavily - tens of millions - in promoting the Kimi chatbot, yet DeepSeek's breakthrough rendered that spending obsolete overnight. Following this, Yang halted all marketing of the Kimi chatbot, according to the same sources, and redirected efforts toward enterprise and developer customers.
The company rolled out K2 in July 2025, which drew international acclaim from both coders and financial backers. Sam Altman, OpenAI's chief, cited Kimi and DeepSeek together as evidence that Chinese open-source AI was endangering America's dominance. Tech firms in Silicon Valley, including Cursor and Perplexity AI, started incorporating K2 into their offerings.
During K2's development, Yang changed direction. He abandoned a previous plan to limit K2 to 200 billion parameters, increasing it to 1 trillion, believing that size would yield superior long-term results. He explained to his team that the model's name was a reference to Mount Godwin-Austen, the second-highest mountain globally, recognizing that Kimi still had a tough climb ahead.
K2 emulated DeepSeek's recipe of high performance, low cost, and open-source design, while adding proprietary advances such as the MuonClip optimizer that enhanced training stability and cut computing needs significantly.
Over the remainder of 2025, Moonshot iterated on K2, releasing the K2.5 in January 2026 and deploying an agent swarm to automate corporate workflows. This capability propelled the company's annualized recurring revenue to over $100 million by March, according to insiders. By April, that same revenue figure climbed to $200 million, and by June it hit $300 million.
Concurrently, the strong launches of competitors like Z.AI and MiniMax encouraged investors to pour money into Moonshot. Within five months, the company's valuation soared to $20 billion, up from $4.3 billion in December, backed by investors such as IDG Capital, Alibaba, Meituan, and several of China's largest state funds. In a year-end message to employees, Yang wrote: "Artificial general intelligence is a marathon, and we are still at the very beginning."
Andy Mok, who serves as a senior research fellow with the Center for China and Globalization, described the launch as a "DeepSeek 2.0 moment," contending that China's political system creates a demand market attracting many competitors.
"Following the release of the Kimi K3, Moonshot's daily sales have surged by at least six times, forcing the company to temporarily freeze new subscriptions to deal with a compute crunch". The company is currently closing an investment round that sets its worth at over $30 billion, with plans to go public on the Hong Kong exchange in about six months.
For investors, the takeaway is straightforward. The race between US and Chinese AI is not over by a long shot. If Kai-Fu Lee is right, both sides can win - just in different ways. American closed models may make the most money, but Chinese open models will get more market share.
Join Market Briefs, our free daily newsletter, for a quick daily rundown of the markets
