Chinas AI Giants and Startups Split Paths

Strategic Divergence Emerges in China's AI Landscape as Model Giants and Startups Forge Separate Paths

In a dramatic surge underscoring the heated state of China's artificial intelligence sector, leading AI model firms Zhipu AI and MiniMax saw their market capitalizations surpass the HK$300 billion mark this week, overtaking established internet giants like Kuaishou and Trip.com in valuation. This surge, triggered by bullish investor sentiment following significant model updates, signals more than fleeting market enthusiasm. It highlights a fundamental and accelerating strategic bifurcation within China's once-unified "hundred-model war," where the race is no longer solely about conversational prowess but about defining sustainable paths to commercial viability and technological leadership.

The 'Hundred-Model War' and the Strategic Fork

The flurry of AI model releases from Chinese companies, often dubbed the "hundred-model war," has entered a new, more nuanced phase. If the initial period was characterized by a scramble to match or replicate the capabilities of OpenAI's ChatGPT, the competitive narrative in 2026 is undergoing a structural divergence. Industry observers note a clear split: while well-funded technology giants are systematically pursuing a vision of becoming AI-era infrastructure providers, a cohort of agile startups is pivoting away from costly consumer-facing "chatbot" battles toward specialized, enterprise-ready applications in coding and AI agents.

This strategic fork represents a maturation of the market, moving beyond pure capability demonstration to address the hard constraints of cost, scalability, and monetization. The recent market performance of companies like Zhipu and MiniMax is a direct reflection of investor confidence in these emerging, more focused strategic pathways.

Giants Emulate Google's Full-Stack Ambition

For China's internet behemoths—Baidu, Alibaba, ByteDance—the prevailing ambition is increasingly articulated as becoming "the Google" of the AI era. This is not a simple analogy but a reference to a specific, integrated strategy encompassing product, foundational model, cloud computing, and proprietary silicon. This full-stack approach, perfected by Google over two decades, is seen as the only sustainable narrative for giants facing the immense computational demands of advanced AI.

The rationale is deeply rooted in economics and control. As models evolve, particularly toward resource-intensive video and image generation, the cost of inference—the compute required to run a model—threatens to overwhelm business models. A company that controls the entire stack, from the AI chips to the cloud platform hosting the models, can optimize costs and performance in a way a pure-play model developer cannot.

ByteDance's recent global impact with its video generation model, Seedance 2.0, exemplifies this logic. Its success is partially attributed to its unique access to vast, real-world video datasets from its TikTok and Douyin platforms, providing invaluable training material. More importantly, as analysis from Huachuang Securities notes, ByteDance's model strategy is geared toward creating low-cost, tool-like accessibility, aiming to reduce content production costs across its ecosystem and, ultimately, enhance advertising efficiency. Similarly, Alibaba's Tongyi Qianwen focuses on high-fidelity image generation tailored for e-commerce, directly boosting its core commercial platforms.

The ultimate goal, as seen with Meta's generative AI tools driving disproportionate advertising revenue growth, is to leverage multimodal AI to create more effective and scalable commercial formats. For these giants, pursuing artificial general intelligence (AGI) is framed less as a standalone goal and more as a byproduct of infrastructure evolution and organizational synergy across their vast product portfolios.

Startups Become Anthropic Disciples

Conversely, for China's AI startups, the reference point has decisively shifted from OpenAI to Anthropic. The constraints of the consumer chatbot route have become starkly apparent. The model-as-a-chatbot business is marked by high inference costs per interaction, low user retention, and a slow, subsidy-heavy path to monetization—a perilous gamble for capital-constrained startups.

Anthropic's path offers a compelling alternative. By focusing on high-value, practical applications like coding, robust APIs for developers, and sophisticated AI agents, Anthropic has demonstrated a viable route to rapid commercialization. Its revenue growth in recent quarters has been astronomical by software industry standards, and its ratio of research and development spending to revenue is notably healthier than many rivals, suggesting a sustainable model of "high-intensity R&D plus commercialization."

Chinese startups are now fervently following this playbook. Following the rise of models like DeepSeek, consumer-focused players like Kimi have reportedly dialed back expensive user acquisition battles to concentrate on high-value scenarios. Zhipu AI has been openly described as walking the "Anthropic route," a perception its leadership has embraced. The recent, nearly simultaneous releases of new coding-focused models—Zhipu's GLM-5 and MiniMax's M2.5—were explicitly benchmarked against Anthropic's Opus series, with both companies declaring their ambition to be China's answer to Anthropic.

The critical distinction from the giants' efforts in coding is the startups' focus. For a ByteDance, coding tools may serve to enrich its cloud ecosystem. For a Zhipu or MiniMax, these tools represent the core commercial model: deeply embedding AI into enterprise workflows where the value is clear, the cost is justified by productivity gains, and the product becomes indispensable.

Market Validation and Future Trajectory

The stock market has delivered a powerful verdict on this strategic pivot. The announcement of their advanced coding models propelled Zhipu and MiniMax shares to record highs. Zhipu's market capitalization skyrocketed, briefly crossing HK$1.7 trillion, while MiniMax approached HK$1.8 trillion. Their combined valuation power now places them among the top tier of Chinese technology companies, a staggering achievement for firms focused on what was once considered a "narrower" segment of the AI market.

This investor enthusiasm underscores a broader recognition: the true watershed in China's AI competition is not the density of model releases, but the clarity of path selection. It distinguishes those still competing for a generic user interface from those building indispensable workflow tools, and those with the balance sheet to bear full-stack costs from those who must quickly prove a profitable business loop.

The competition has evolved from a single race on a unified track to two parallel, high-stakes journeys. On one path, giants marshal their vast resources to build the foundational platforms of the next computing era, betting on vertical integration and scale. On the other, nimble specialists concentrate on solving specific, high-value problems with surgical precision, betting on utility and rapid business validation.

While the long-term winners remain undetermined, the strategic directions are now clearly mapped. China's AI landscape is no longer a monolithic "war" but a complex, bifurcated battlefield where success will be defined not just by technological brilliance, but by strategic discipline and commercial acumen.