Moonshot AI commercialization is accelerating, but major internet platforms remain strong competitors

Overview

This report summarizes the key points from a conference call between Nomura’s China internet team and Moonshot AI experts. Moonshot AI is a Beijing-based generative AI startup founded in early 2023 by Dr. Yang Zhiling and Tsinghua alumni, with core products including the Kimi Assistant, the K series foundation models, and Agent products such as Kimi Code, Kimi Agent, and Kimi Work. The company has backing from Alibaba, Tencent, HongShan, ZhenFund, and IDG Capital, and media reports indicate it completed USD2bn financing in May 2026 at a post-money valuation of USD20bn.

Core views

Experts believe Moonshot AI commercialization is accelerating, with ARR expected to move from USD400-500mn in mid-2026 to above USD1bn by year-end, driven mainly by overseas API usage and AI programming capabilities. Future growth is expected to expand from AI programming to vertical sectors such as healthcare, education, and finance, where scarce high-quality proprietary data will become a key differentiator. Nomura agrees that independent AI labs like Moonshot can have tactical leadership in certain vertical scenarios, but does not fully believe they will eventually beat large internet platforms; Nomura argues that companies such as Alibaba and ByteDance have strong technology, capital, and data ecosystems, and have strategic necessity to continue investing in proprietary SOTA LLMs.

Analysis framework

The report is based on expert call information and qualitatively analyzes Moonshot AI’s revenue growth, product roadmap, model open/closed-source strategy, inference costs, vertical-industry opportunities, and China’s LLM competitive landscape, comparing expert views with Nomura’s own assessment of large internet platform competitiveness.

Methodology notes

• Expert interviewExpert conference call minutes

  Supplement the assessment of operating performance and competitive structure for unlisted AI companies with expert viewpoints from the industry.

  The report’s primary information comes from the Moonshot AI expert call and covers themes such as ARR, revenue structure, overseas demand, product roadmap, compute costs, and industry competition.

• Industry competition analysisLLM competitive landscape analysis

  Compare independent AI laboratories and major internet platforms on model capabilities, data, capital, and commercialization pathways.

  Experts are inclined to be optimistic about Moonshot, DeepSeek, and other pure AI labs; Nomura, however, emphasizes that major internet platforms have capital, data ecosystems, and strategic need for self-developed models.

• Commercialization analysisARR and revenue mix analysis

  Track AI company commercialization progress using indicators such as ARR, API calls, subscriptions, and private deployment.

  Moonshot’s revenue focus is shifting toward API calls, consumer subscriptions are growing slower, and private deployment share is relatively small and declining.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Moonshot AI

  Core research subject of the report; an unlisted generative AI company.

  Strengths

  Fast ARR growth, strong overseas API usage, and recognition for Kimi and K-series models in AI programming and Agent capabilities, with strong fundraising ability.

  Weaknesses

  Long-term AI programming TAM may be limited; commercialization still depends on overseas API expansion and continued model upgrades, and making flagship models closed source may weaken certain developer reach.

  Comparison

  Experts believe Moonshot and DeepSeek and similar pure AI labs are more focused than large platforms; Nomura argues that large platforms still hold capital and data advantages.

  Risks

  API price cuts, intensified competition, offshore demand volatility, difficulty obtaining vertical-industry data, regulatory and compliance constraints, and inference-cost reductions not meeting expectations.

• Alibaba

  Investor in Moonshot AI and a potential LLM competitor, with Qwen used as a benchmark model.

  Strengths

  Has capital, cloud infrastructure, data ecosystem, and strategic necessity for proprietary SOTA LLMs.

  Weaknesses

  Experts believe large platforms covering the full AI value chain may have diluted focus.

  Comparison

  Nomura does not believe large platforms will weaken in-house LLM investment and notes that Qwen’s latest flagship model uses a closed-source premium API strategy.

  Risks

  Model competition, trade-offs between open and closed source strategies, and resource allocation between cloud and model commercialization.

• ByteDance

  A major internet platform competitor, still viewed as an important opponent in China’s LLM market.

  Strengths

  Has massive user scale, data ecosystem, capital, and application scenarios.

  Weaknesses

  Expert views suggest large platforms may have diluted AI focus due to broad business scope.

  Comparison

  Compared with independent AI labs like Moonshot, ByteDance’s advantage leans more toward platform data, capital, and distribution.

  Risks

  Needs sustained investment to maintain SOTA model performance and faces first-mover advantages of independent AI labs in certain vertical segments.

• NVIDIA H20 and domestic inference chips

  Part of Moonshot AI’s inference infrastructure.

  Strengths

  Higher share of domestic inference chips helps lower unit inference costs, while NVIDIA GPUs still support part of the inference workload.

  Weaknesses

  The compute mix and resource utilization affect the pace of margin improvement.

  Comparison

  Cost reductions can partially offset downward pressure on industry API prices.

  Risks

  Chip supply, performance, concurrency efficiency, and utilization may fall short of expectations.

Key data

• Report date2026-07-06Date shown on the front page.
• Moonshot AI mid-2026 ARRUSD400-500mnExperts said monthly ARR is growing about 10-20% month-on-month.
• Moonshot AI end-2026 ARR outlookOver USD1bnThis forecast is based on expert views, mainly relying on overseas API calls, subscriptions, and model upgrades.
• Latest financingUSD2bnAccording to Chinese media reports, completed in May 2026 with Meituan as lead investor.
• Post-money valuationUSD20bnFrom financing information reported in the media.
• K2.7 inference workload gross marginclose to 30%Experts said increasing domestic inference-chip usage, under the current compute mix, helps reduce unit inference costs.
• Potential K3 launch timingAround Sept 2026Experts noted that timing may still be adjusted according to competitive pacing from MiniMax, Zhipu, DeepSeek, and Alibaba Qwen.

Impact & implications

The implication for investment research is that China’s generative AI commercialization is shifting from model launches and user growth toward measurable revenue, API calls, vertical-industry deployment, and inference-cost control. Moonshot AI’s ARR growth reinforces the commercial value of AI programming and overseas developer demand, but long-term competition is not determined only by model capability; proprietary data, compute cost, closed-source pricing strategy, vertical-industry compliance, and strategic investment by large platforms will all shape industry structure.

Risks

• The scale of high-value users in AI programming is limited, so long-term TAM may be smaller than currently expected.
• International players such as Anthropic and OpenAI already occupy some high-value global demand.
• Declining API prices may compress commercialization space, although cost reductions may partially offset the pressure.
• Verticals such as healthcare, education, and finance need high-quality proprietary data and face privacy, compliance, and regulatory constraints.
• Major internet platforms with capital, data, and application ecosystems may erode independent AI labs’ long-term leadership.
• The release timing and performance of new models such as K3 are uncertain and influenced by competitor launch cycles including MiniMax, Zhipu, DeepSeek, and Qwen.

What to watch

• Whether Moonshot AI’s ARR exceeds USD1bn by year-end.
• Whether overseas API calls and international subscription growth remain sustained.
• The release timing, model capability, multimodal, programming, and Agent orchestration performance of K3.
• Whether a tiered strategy of closed-source flagship models and open-source mid-to-low-end models can improve monetization and margins.
• The extent of unit inference cost and margin improvement after increasing domestic inference-chip share.
• Execution progress and data acquisition capability of AI Agents in verticals such as healthcare, education, and finance.
• Release and commercialization strategies of competing models from Alibaba Qwen, DeepSeek, MiniMax, Zhipu, and ByteDance.