MaHui | MaHui Member Moonshot AI Closes $500 Million Series C, Aims to Become a World-Leading AGI Company

On December 31, 2025, Zhilin Yang, founder and CEO of MaHui member Moonshot AI, sent an internal letter announcing that the company had recently completed a $500 million Series C funding round. Source Code Capital participated in the company's early-stage funding.

Moonshot AI (hereinafter referred to as Kimi)'s Series C was led by IDG, with follow-on investments from existing shareholders including Alibaba, Tencent, and Huiwen Wang. The proceeds will be used to more aggressively expand GPU capacity and accelerate the training and development of the K3 model. Following this round, Kimi's post-money valuation jumped to $4.3 billion (approximately RMB 30 billion).

In 2025, Kimi achieved multiple breakthroughs. In particular, the release of K2 and K2 Thinking marked an important step for Kimi on the path to AGI.

In his internal letter, Zhilin Yang stated that the company's most important goal going forward is to surpass frontier companies like Anthropic and become a world-leading AGI company. The 2026 strategy will advance along three fronts.

Below is Zhilin Yang's internal letter:

Through our collective efforts, 2025 has been a year of breakthroughs for Kimi. The release of K2 and K2 Thinking marked an important step on our path to AGI. China's first trillion-parameter foundation model, the first open-source agentic model, achieving SOTA on core benchmarks such as HLE by surpassing OpenAI, a distinctive creative writing style beloved by global users, and the first successful large-scale training using second-order optimizers — the K2 series has taken Kimi from China to the world, generating significant impact in Silicon Valley and the broader global tech community, trending on Twitter worldwide, and earning high praise from renowned technology leaders including Nvidia's founder, the All-In Podcast hosts, Anthropic's co-founder, Perplexity's founder, a16z's founder, Vercel's founder, and Nobel laureates.

They're all using Kimi K2

These technological breakthroughs have also laid the groundwork for product explosion and rapid commercial progress. Starting in May, Kimi began frequently rolling out new agent features, launching Researcher, OK Computer, PPT, Kimi Code, and other new products with increasingly powerful capabilities. Leveraging the SOTA performance of the K2 model, C-end commercialization grew exponentially, with overseas and domestic paying users averaging over 170% month-over-month growth from September to November. Meanwhile, the release of K2 Thinking also significantly drove API revenue growth, with overseas API revenue increasing 4x from September to November.

Thanks to these breakthroughs, the company recently completed a $500 million Series C funding round that was significantly oversubscribed, with current cash holdings exceeding RMB 10 billion. Compared to the secondary market, we believe we can still raise larger amounts from the primary market — in fact, our combined Series B/C funding already exceeds most IPO proceeds and follow-on offerings by listed companies. So we're not in a rush to go public in the near term. Of course, we plan to use going public as a means to accelerate AGI, acting when the time is right with the initiative in our hands.

The Series C proceeds will be used to more aggressively expand GPU capacity and accelerate the training and development of the K3 model. A portion will also fund the 2026 incentive plan and option buyback program. In 2025, based on SOTA results, various incentive measures including salary adjustments, option grants, and cash bonuses were awarded to 324 people. Before the 2026 Spring Festival, we will finalize and distribute the reward plan for K2 Thinking and subsequent model and product releases. The company's average incentives in 2026 are expected to be 200% of 2025 levels, while we also plan to significantly increase the option buyback quota — detailed plans will be announced shortly.

The company's most important goal going forward is to surpass frontier companies like Anthropic and become a world-leading AGI company. The 2026 strategy will advance along the following fronts:

1. Through technical improvements and further scaling, the K3 model will increase effective FLOPs by at least one order of magnitude, catching up to world frontier models in pre-training capabilities.

2. Vertically integrate model training and agent product taste, making K3 a more "distinctive" model that lets users experience entirely new capabilities that other models haven't defined.

3. In product and commercialization, focus on agents rather than absolute user numbers, continuously pushing the boundaries of intelligence, creating greater productivity value, and achieving order-of-magnitude revenue growth.

This is a noisy era, but for me, Kimi's starting point was simple: pure curiosity. Curiosity about where the ceiling of AGI lies, about what our ideal model looks like, about how to build the capabilities we want. The joy this curiosity brings is very real. When the agent features we pulled all-nighters on finally go live, when we wake up and open TensorBoard to see training curves that blow our minds. We share these precious processes of approaching truth, and believe this simple joy can carry us through cycles amid the noise.

In 2026, Kimi will become an LLM that is "distinctive" and "undefined." Whether it's technical uncharted territory that others didn't dare to bet on, or aesthetic convictions that require a bit of obsession, I believe more Kimi-defined innovations can make unique contributions to accelerating human civilization. This distinctiveness is the greatest meaning of our existence.

Let's keep pushing!