Technical Deep Dive
The valuation of Zhipu AI at over 1.1 trillion HKD is not merely a financial anomaly; it reflects a fundamental reassessment of what constitutes technological moats in the AI era. Zhipu AI's core technical differentiator is its GLM (General Language Model) architecture, particularly the GLM-130B and its successor models. Unlike OpenAI's GPT series, which is based on a decoder-only transformer, GLM employs a unique autoregressive blank infilling objective. This allows it to perform both natural language understanding and generation tasks with a single unified framework, offering competitive performance on benchmarks like MMLU and C-Eval.
From an engineering perspective, Zhipu AI has invested heavily in efficient training and inference. The company open-sourced GLM-130B in 2022, which was one of the first 100-billion-parameter models trained on a Chinese supercomputer. The model uses a 3D parallelism strategy (data, tensor, pipeline) combined with ZeRO optimization to handle the massive memory footprint. On the inference side, Zhipu has developed a custom serving framework that achieves sub-100ms latency for long-context generation, critical for enterprise applications. The company's CodeGeeX model, a code generation tool, has gained significant traction on GitHub, with over 15,000 stars and active community contributions.
Table: Benchmark Performance Comparison (Selected Chinese LLMs)
| Model | Parameters | C-Eval (Avg) | MMLU (Avg) | Cost/1M tokens (CNY) |
|---|---|---|---|---|
| Zhipu GLM-4 | ~130B (est.) | 82.5 | 86.4 | ¥0.80 |
| Baidu ERNIE 4.0 | ~200B (est.) | 79.3 | 82.1 | ¥1.20 |
| Alibaba Qwen2.5-72B | 72B | 80.1 | 85.0 | ¥0.50 |
| ByteDance Doubao | — | 78.9 | 83.5 | ¥0.60 |
Data Takeaway: Zhipu's GLM-4 leads on both Chinese and English benchmarks while maintaining a competitive pricing structure. This performance-to-cost ratio is a key driver of its market valuation, as enterprises seek models that are both capable and economical for deployment.
Key Players & Case Studies
The three events highlight distinct strategies from major players. Alibaba's rice planting is a symbolic gesture, but the company's real AI play is through its Qwen series of open-source models. Qwen2.5-72B has become a favorite among developers for fine-tuning, with over 30,000 GitHub stars and a vibrant ecosystem of adapters. However, Alibaba's cloud business, which provides the infrastructure for AI workloads, is facing margin pressure from aggressive pricing by competitors like ByteDance and Tencent. The rice planting exercise, while mocked, serves as a narrative tool to remind the market that Alibaba is still grounded in operational reality — a contrast to the 'cloud-only' narrative of its rivals.
Zhipu AI, on the other hand, has pursued a strategy of deep vertical integration. It has partnered with state-owned enterprises and government agencies to provide secure, localized AI solutions. This has insulated it from some of the geopolitical risks that affect companies like Baidu, which relies more on international partnerships. Zhipu's valuation is also buoyed by its role in China's national AI infrastructure plan, which prioritizes domestic model development.
Onvo (乐道), NIO's sub-brand, represents the struggle of latecomers in the EV market. The CEO's comment about selling cars being harder than getting into Tsinghua is a stark admission of the market's saturation. Onvo's L60 model, priced around ¥200,000, competes directly with Tesla's Model 3 and BYD's Seal. The company is relying on NIO's battery swapping network as a differentiator, but the network's expansion has been slower than expected.
Table: Competitive Landscape in China's EV Market (Q1 2026)
| Brand | Q1 Sales (units) | YoY Growth | Average Selling Price (CNY) | Key Differentiator |
|---|---|---|---|---|
| BYD | 750,000 | +15% | ¥180,000 | Vertical integration, scale |
| Tesla | 220,000 | -5% | ¥260,000 | Brand, FSD software |
| NIO (incl. Onvo) | 45,000 | +8% | ¥350,000 (NIO) / ¥200,000 (Onvo) | Battery swapping, premium service |
| Xiaomi SU7 | 80,000 | N/A | ¥215,000 | Smartphone ecosystem integration |
Data Takeaway: Onvo's sales are still a fraction of the market leaders. The 'harder than Tsinghua' comment reflects the reality that even with a strong parent brand (NIO) and a unique technology (battery swapping), breaking into the mass market requires either massive scale or a breakthrough in cost or features.
Industry Impact & Market Dynamics
The simultaneous occurrence of these events signals a profound reordering of China's tech hierarchy. The market cap of Zhipu AI exceeding the combined value of Baidu, Meituan, and Kuaishou is a watershed moment. It indicates that investors are now valuing foundational AI capabilities over consumer internet platforms. Baidu's core search business is being eroded by AI-powered answer engines, while Meituan's food delivery and Kuaishou's short-video platforms face margin compression from AI-driven automation. The market is effectively saying that the next generation of value creation will come from models, not platforms.
This shift has immediate consequences. Alibaba, despite its AI efforts, is seen as a legacy player trying to retrofit AI onto an existing e-commerce and cloud business. Its stock has underperformed Zhipu's by 40% over the past year. The rice planting exercise can be interpreted as an attempt to humanize the company and re-establish trust, but the market is voting with its dollars for pure-play AI.
For the EV sector, the 'Tsinghua' comment underscores a broader crisis of differentiation. As battery technology becomes commoditized and autonomous driving features converge, brands are left competing on price and marketing. The result is a brutal price war that is squeezing margins across the industry. Onvo's struggle is a microcosm of this: even with the backing of a well-known EV maker, entering the mass market is a battle of attrition.
Risks, Limitations & Open Questions
Zhipu AI's valuation carries significant risk. The company's revenue is still a fraction of its market cap, with most of its income coming from API calls and enterprise contracts. The AI model market is becoming increasingly commoditized, with open-source models like Meta's Llama and Alibaba's Qwen offering comparable performance for free. If Zhipu cannot maintain its performance lead or build a sticky ecosystem of applications, its valuation could correct sharply.
For Alibaba, the rice planting narrative risks being seen as a gimmick. The company's real challenge is executing its AI strategy while managing a sprawling portfolio of businesses. The 'barbie farm' comparison is a double-edged sword: it generates buzz, but it also highlights the gap between Alibaba's digital past and its physical present.
For Onvo and the broader EV market, the question is whether consolidation will come fast enough. With over 100 EV brands in China, many analysts predict a shakeout that will leave only 5-10 survivors. Onvo's survival depends on NIO's ability to continue funding its losses, which is not guaranteed given NIO's own financial struggles.
AINews Verdict & Predictions
Prediction 1: Zhipu AI will face a valuation correction within 18 months. The current market cap is pricing in perfection — sustained leadership in model performance, rapid enterprise adoption, and immunity from open-source competition. None of these are guaranteed. We expect a 30-40% pullback as the market realizes that AI model companies are not immune to the law of diminishing returns.
Prediction 2: Alibaba's 'back to basics' narrative will fail to move its stock price. The rice planting is a distraction. Alibaba's future depends on its cloud business and its ability to integrate AI into e-commerce. Without a clear AI product that captures the public imagination, the company will continue to trade at a discount to AI-native peers.
Prediction 3: Onvo will either be merged into NIO's main brand or shut down within two years. The EV market is too crowded for a sub-brand that offers marginal differentiation. NIO should focus on its premium positioning and let Onvo go, rather than bleeding cash in a segment it cannot dominate.
What to watch next: The next major catalyst for the Chinese AI market will be the release of GPT-5's Chinese-language performance. If it significantly outperforms domestic models, it could trigger a re-evaluation of Zhipu's moat. For the EV sector, watch for the next round of price cuts from BYD — they will determine whether Onvo and other second-tier players can survive.