Chinas AI Strategy Splits Between Heart and Code

AI Divergence: From Code Companions to Emotional Partners, Chinese Tech Firms Chart New Commercial Paths

The Lunar New Year holiday period has traditionally been a battleground for consumer internet companies vying for user attention with cash incentives and红包 campaigns. This year, however, a different kind of competition has taken center stage in China's technology sector, revealing a strategic bifurcation in the commercialization of artificial intelligence. On one front, AI developers are launching sophisticated, enterprise-grade tools, betting that professionals will pay a premium for enhanced productivity. On another, a burgeoning market for AI-powered emotional companions is tapping into a deep-seated consumer need for connection, offering a glimpse into a future where technology serves not just the mind, but also the heart. This dual trajectory underscores a maturing AI landscape where commercial viability is being tested across vastly different spectra of human need.

Strategic Model Launches Fuel Market Optimism

In a tightly sequenced move just before the holiday closure, two leading Chinese AI firms, Zhipu AI and MiniMax, unveiled significant upgrades to their flagship models, triggering a remarkable rally in their stock prices.

Zhipu AI ignited the activity with the open-source release of GLM-5, a model it positions for "agent engineering" — a paradigm shift from merely assisting with code writing to managing complex, multi-step software projects. The company frames this as the natural evolution beyond "vibe coding," a concept popularized by figures like former OpenAI lead Andrej Karpathy, which emphasizes a developer's role shifting from coder to AI agent orchestrator.

Technically, GLM-5 represents a substantial leap. Its parameter count expanded to 744 billion (with 40 billion activated), and it was trained on 28.5 trillion tokens of data. Crucially, it incorporates two key innovations: an asynchronous reinforcement learning framework called "Slime," designed to maintain consistency in long-horizon tasks, and DeepSeek's sparse attention mechanism (DSA) to improve efficiency and reduce operational costs when processing lengthy contexts, such as large codebases.

Benchmark results released by Zhipu appear robust. GLM-5 achieved top-tier scores among open-source models on programming benchmarks like SWE-bench-Verified (77.8) and Terminal Bench 2.0 (56.2), the latter testing command-line operational proficiency. Its performance on the novel "Vending Bench 2," which simulates running a vending machine business over a simulated year, demonstrated nascent capabilities in long-term planning and resource management.

Within hours, MiniMax responded with the launch of its M2.5 programming model, described as natively designed for agent scenarios and benchmarked against Anthropic's Claude Opus 4.6. Adopting a contrasting "small but beautiful" philosophy, M2.5 operates with a far leaner 10 billion activated parameters, prioritizing inference speed and cost efficiency. While MiniMax has not open-sourced the model's weights, early community feedback suggests it delivers performance competitive with mid-tier international models at a fraction of the cost, with some users reporting speeds twice as fast as Claude Sonnet 4.5 for just 8% of the API cost.

The market reaction was swift and decisive. Zhipu's stock surged over 25% on the announcement day, capping a weekly gain of more than 77% and pushing its market capitalization past HK$170 billion. MiniMax's shares also jumped over 20%, elevating its valuation above HK$180 billion. This explosive growth is particularly notable given Zhipu's IPO valuation stood at just HK$51.8 billion in early January.

Beyond the Hype: The Underlying Business Logic

Analysts caution that the stock rallies are not merely a "new model" hype cycle but are underpinned by tangible business developments. For Zhipu, the launch coincided with a strategic price increase. The company raised the cost of its GLM Coding Plan by over 30%, citing "continuously strong market demand" and the need to support service stability and further optimization. This move is counter to the prevailing industry trend of aggressive price-cutting, exemplified by DeepSeek, and signals Zhipu's confidence in a product-led growth strategy. The company reported that its MaaS platform's Annual Recurring Revenue (ARR) skyrocketed from RMB 20 million to over RMB 500 million within ten months of releasing its previous GLM-4.7 model, suggesting a willing customer base for high-efficacy professional tools.

For MiniMax, the positive investor sentiment was bolstered by a pivotal "overweight" initiation report from J.P. Morgan, which set a HK$700 price target. The bank highlighted MiniMax's distinctive global footprint, with overseas revenue constituting 73.1% of its total in the first three quarters of 2025. This "born-global" positioning, coupled with its cost-advantaged pricing compared to Western rivals like OpenAI and Google, is viewed as a durable competitive moat, especially given the higher willingness-to-pay in overseas markets.

This pre-holiday flurry was part of a broader wave of releases, including models from Alibaba, Moonshot AI, and StepFun, with ByteDance poised to join. As Citigroup noted in a late-January report, the 2026 Lunar New Year represents a critical inflection point for transitioning AI assistants from technical proofs-of-concept to mass-adopted applications, making user habit formation during the extended break a key battleground.

The Rise of the "Emotional Oxygen Tank": AI Pets as Consumer Commodity

Parallel to these advancements in professional AI tools, a distinct and rapidly growing consumer market has emerged: AI-powered companion pets. Products like the viral "Fuzozo" (芙崽), which retails for around RMB 369, or Huawei's "Hanhan," are being marketed as "emotional substitutes" and "electronic ibuprofen" for a generation grappling with urban isolation.

Demographic shifts provide the backdrop. Projections from China's National Bureau of Statistics suggest the solo-living population could reach 150-200 million by 2030. For young professionals like Lin Xia, a designer working late nights in a quiet apartment, the loneliness is palpable. "During those nights when your body is exhausted but your mind is wide awake, the quieter the rental apartment is, the more easily a sense of loneliness suddenly washes over you," she describes. Her AI pet, which responds to touch with following eye movements and, after a "personality cultivation" period, begins to converse and keep a diary, provides a low-stakes, high-certainty source of companionship.

The value proposition is clear: all the perceived benefits of companionship—unconditional attention, interactive play, a sense of being "seen"—with none of the real-world responsibilities, costs, or heartbreak associated with living pets. These devices leverage voice recognition, tactile sensors, and increasingly sophisticated emotion-simulation algorithms to provide customized, evolving interactions. They represent the commodification of emotional labor, packaged as an affordable, durable consumer good.

This "emotional economy" has swiftly crystallized into a market segment worth billions of yuan. Product stratification is already evident, with tiers ranging from simple interactive toys to more complex companions capable of "personality development" based on user interaction, complete with simulated MBTI types.

Converging on Commercialization: Tools for Work and Life

The simultaneous fervor in enterprise AI and emotional AI is not coincidental. It marks a pivotal phase in China's AI industry where companies are moving beyond speculative competition on parameter counts and funding rounds to concretely address monetizable needs. The path is splitting: one road leads to the professional workstation, enhancing cognitive labor and complex problem-solving, while the other leads to the home and the individual psyche, offering solace and connection.

The contrast is stark. Zhipu and MiniMax are betting that developers and enterprises will invest in AI that makes them more capable and efficient, even at a higher price. The AI pet industry, meanwhile, is proving that consumers will pay for AI that makes them feel less alone. Both strategies validate the integration of AI into the fabric of daily life and work, but they answer fundamentally different questions about value.

The challenges ahead are equally divergent. For the coding agents, the test is sustained technical superiority, developer adoption, and the ability to defend premium pricing in a competitive global market. For the emotional companions, the questions are ethical and psychological: to what extent do these relationships represent healthy coping mechanisms versus digital pacifiers, and how will they evolve as the technology becomes more persuasive?

As the Year of the Horse begins, the Chinese AI sector has demonstrated it is no longer running a single race. It is sprinting on two tracks simultaneously—one aimed at empowering the human intellect in professional realms, the other at comforting the human heart in private life. The ultimate commercial winners may be those who can most convincingly prove their utility on either, or perhaps eventually both, of these profound frontiers.