Chinese artificial intelligence startups Zhipu AI and MiniMax have provided the first concrete look at their financial performance since going public in Hong Kong this January. The key takeaway: both companies are achieving explosive revenue growth, validating early commercial demand for their AI models, even as the path to profitability remains distant with widening losses.
The Deal: Revenue Soars, Losses Widen
The 2025 financial results reveal the scale and cost of the AI race in China.
- Zhipu AI (Z.ai): Reported revenue of 724.33 million yuan (US$104.8 million) for 2025, a 131.9% year-on-year increase. The Beijing-based company's business is centered on its model-as-a-service platform, serving institutional clients through both cloud and on-premise deployments.
- MiniMax: Reported 2025 revenue of US$79 million, representing a staggering 159% year-on-year jump. The Shanghai-based firm has a more diversified model, combining enterprise AI services with revenue from popular consumer apps: the video generation platform Hailuo AI and the AI companion app Talkie.
Despite these impressive top-line figures, both companies reported widening net losses, a common theme for capital-intensive AI firms investing heavily in research, development, and compute infrastructure.
What the Companies Do: Diverging Paths to Market
As pure-play AI companies, Zhipu and MiniMax are building businesses solely around their AI models, unlike integrated tech giants that bundle AI with cloud services or hardware.
Zhipu AI's Model-as-a-Service Play: Zhipu's strategy is squarely focused on the enterprise. Its core offering is providing access to its proprietary large language models (LLMs) for institutional clients. A key differentiator is offering local, on-premise deployment, which addresses data security and sovereignty concerns critical for government and large corporate clients in China. This B2B focus mirrors the enterprise sales motion of companies like Anthropic.
MiniMax's Hybrid Consumer-Enterprise Strategy: MiniMax has built a multi-pronged revenue engine. Alongside its enterprise AI services, it has successfully launched direct-to-consumer products. Talkie, an AI companion app, taps into the global demand for interactive AI relationships. Hailuo AI, a video generation platform, competes in the rapidly growing AI video synthesis market. This diversification may provide a more resilient revenue base, leveraging consumer adoption to fund advanced model development.
Market Context: Proving Viability in a Crowded Field
The successful IPOs and subsequent revenue growth of Zhipu and MiniMax are significant markers for the global AI industry. They demonstrate that specialized AI model companies can achieve substantial, rapidly scaling revenue outside the shadow of hyperscalers like Alibaba Cloud or Tencent. Investor continued support for their stocks despite losses indicates a belief in the long-term value of their technology stacks and market positions.
This commercial traction comes amidst intense technical competition. As noted in our Knowledge Graph, MiniMax has been aggressively advancing its core models, recently releasing MiniMax M2.7, an AI agent that autonomously rewrites its own operational harness and has shown strong performance on benchmarks like MLE Bench Lite.
gentic.news Analysis
The first earnings from Zhipu and MiniMax are a crucial data point in the narrative of AI commercialization. The >100% revenue growth is undeniably strong, confirming there is a paying market for specialized AI model access in China. However, the widening losses underscore the immense, ongoing cost of the AI arms race. These are not software companies with 80% margins; they are research-intensive organizations where each percentage point of benchmark improvement requires massive capital expenditure on compute.
MiniMax’s diversified strategy, particularly its success with consumer apps like Talkie, is noteworthy. It creates a potentially valuable flywheel: consumer applications generate revenue and, critically, vast amounts of unique interaction data that can be used to refine models, which in turn improves both consumer and enterprise offerings. This contrasts with Zhipu’s pure enterprise focus, which may lead to deeper, stickier contracts but less diversified data intake.
The context from our recent coverage is essential here. Just days ago, we reported on MiniMax M2.7 tying Gemini 3.1 on MLE Bench Lite and its ability to rewrite its own operational harness. This technical prowess is the engine behind the commercial numbers reported today. Investors are funding these losses based on the belief that such technical leads are sustainable and will translate into long-term competitive moats. Furthermore, the broader trend of AGI-focused analysis (appearing in 4 articles this week) and predictions of rapid capability growth fuel the market's willingness to overlook near-term profitability for potential market dominance in a transformative technology.
Frequently Asked Questions
How do Zhipu AI and MiniMax make money?
Zhipu AI primarily makes money through its model-as-a-service platform, selling access to its AI models to enterprise and institutional clients via cloud APIs or on-premise deployments. MiniMax has a hybrid model: it earns revenue from enterprise AI services and from two successful consumer applications—the Hailuo AI video generation platform and the Talkie AI companion app.
Are Zhipu and MiniMax profitable?
No. Both companies reported widening net losses in 2025 alongside their significant revenue growth. This is typical for high-growth AI startups that are investing heavily in research, development, and computing infrastructure (GPU clusters) to advance their models and capture market share, prioritizing scale over immediate profitability.
What are the main consumer apps from MiniMax?
MiniMax operates two main consumer-facing apps: Talkie, an AI companion app where users can interact with AI characters, and Hailuo AI, a platform for generating videos using artificial intelligence. These apps provide MiniMax with a direct-to-consumer revenue stream and a source of diverse user interaction data.
How does Zhipu AI's on-premise deployment work?
Zhipu AI offers clients the option to deploy its AI models locally on the client's own servers (on-premise), as opposed to accessing them via the cloud. This model addresses stringent data privacy, security, and regulatory compliance requirements common among government agencies, financial institutions, and large corporations, particularly in China, by ensuring that sensitive data never leaves the client's controlled environment.
