Chinese AI tigers gain public-market traction

Chinese AI development is moving beyond the largest tech conglomerates as a group of smaller model builders, labeled the "AI tigers," gains public-market footing and investor interest. Two of them, Zhipu and MiniMax, completed Hong Kong listings in January, each raising roughly HKD 4.35-4.82 billion, and their shares have run up several hundred percent since IPO. Market values briefly touched above $40 billion for some names, and as of April 20 their H-share prices traded multiple times above listing levels.

These companies differentiate from hyperscalers by prioritizing narrower, monetizable product pathways and lower-cost inference. Key technical and commercial characteristics to note:

• •Common players include 01.AI, Baichuan AI, MiniMax, Moonshot AI, StepFun, and Zhipu.
• •Business models mix on-premise deployments, cloud model-as-a-service, and verticalized APIs to capture token consumption revenue.
• •Model and agent work centers on GLM-5 and integrations with OpenClaw, with claims of competitive coding and agent benchmarks versus western models.
• •Firms are increasingly adopting domestic chip stacks to meet compute demand and contain export risks.

The rise of these mid-tier players contradicts the narrative that only deep-pocketed giants can drive AI value. Large incumbents are still pouring capital into infrastructure, with ByteDance planning more than RMB 160 billion in AI procurement and Tencent reporting RMB 79.2 billion capex in 2025, but the AI tigers show a different lever for value creation: focused product-market fit plus visible token consumption. China benefits from structural advantages for high-volume token inference, such as lower electricity and data-center costs, which supports a token-economy play. Investor appetite has been amplified by high-profile moments, including positive commentary around OpenClaw from Nvidia CEO Jensen Huang and strong early revenue prints like Zhipu's 724 million yuan in 2025, despite a 3.18 billion yuan net adjusted loss.

These companies are pushing agent-oriented capabilities, tighter vertical integrations, and hybrid deployment paradigms that matter to engineers building enterprise solutions. Expect more packaged inference products, optimized cost-per-token engineering, and stronger local supply chains for accelerators and model ops tooling. For MLOps teams, the emergence of credible Chinese model vendors expands options for on-premise deployment, data residency, and lower inference costs at scale.

Monitor quarterly revenue-to-token metrics, margin progression from price-per-token strategies, adoption of domestic accelerators, and how regulatory scrutiny shapes cross-border expansion. The key open questions are whether strong revenue growth will convert to sustainable profits, and how domestic chip availability will scale with peak inference demand.