ChinaBiz Briefing: Nvidia's China Return, Zhipu's IPO Push, and Long March's Landing Stumble
China's tech landscape is navigating a critical phase of regulatory recalibration and competitive intensity. Nvidia's planned chip exports face mounting obstacles despite approval signals, while domestic AI players like Zhipu sharpen their challenge ahead of capital market debuts. Meanwhile, reality checks in humanoid robotics and aerospace reveal the gap between innovation narratives and industrial execution—underscoring that technical ambition still requires manufacturing maturity.
Nvidia's H200 Return Faces Regulatory Gauntlet and Domestic Competition
Nvidia plans to ship 40,000–80,000 H200 chips to China before the Lunar New Year, targeting $1–4 billion in sales from existing inventory. However, the comeback is far from assured. Chinese authorities have not approved procurement plans, while U.S. agencies continue license reviews amid Congressional opposition. Security concerns linger after China's cybersecurity watchdog flagged vulnerabilities in Nvidia's H20 chips last July. President Trump's mandate requiring Nvidia to remit 25% of revenues to the U.S. government further complicates pricing and margins.
Why it matters: Nvidia's China business collapsed from 95% market share to "zero" following export restrictions, and the window for recapturing ground is narrowing. Domestic players like Moore Threads are closing the performance gap—its new "Huashan" chip reportedly surpasses Nvidia's Hopper series in key specs. Meanwhile, AMD CEO Lisa Su has moved aggressively, securing meetings with Chinese officials and export licenses faster than Nvidia, positioning AMD for potential first-mover advantage. The race reflects a broader shift: China's AI chip market grew 30% in 2025 during Nvidia's absence, with domestic alternatives gaining traction in inference and government projects. Even if H200 exports proceed, Nvidia faces a transformed competitive landscape where ecosystem lock-in and CUDA dominance no longer guarantee market control.
Zhipu Launches GLM-4.7 Ahead of IPO, Challenging Global AI Leaders on Price and Performance
Zhipu AI released its flagship GLM-4.7 model, claiming performance surpassing GPT-5.2 in coding benchmarks while undercutting Western rivals on price. The model scored 42.8% on the "Humanity's Last Exam" reasoning test and topped open-source rankings in Code Arena. Its architecture introduces "Reserved Thinking" and "Turn-level Thinking" mechanisms to improve task stability and reduce inference costs. Annual subscription pricing is reportedly equivalent to one month of competitors' premium plans.
Why it matters: The timing is strategic—Zhipu is preparing for an IPO and needs to demonstrate technical credibility and commercial viability. The aggressive pricing underscores a broader trend: Chinese AI companies are competing not just on capability but on cost-efficiency, a model that could pressure Western players in global developer markets. GLM-4.7's improved visual code understanding and full-stack development capabilities target practical enterprise use cases, moving beyond benchmark performance to real-world application. The model's open-source release also signals a push to build ecosystem adoption before monetization, mirroring strategies that helped Chinese cloud and mobile platforms scale. If Zhipu can maintain this trajectory, it positions itself as a formidable domestic alternative at a moment when geopolitical tensions make supply chain independence a priority for Chinese enterprises.
China's Humanoid Robot Industry Still "Hand-Assembled" Despite Hype
An undercover investigation revealed that China's leading humanoid robot startups assemble products manually in office buildings, not automated factories. Workers use electric screwdrivers and hex keys to build robots from printed A4 manuals, relying on physical skill rather than standardized production lines. The workshop-style approach creates quality inconsistencies and high scrap rates, with senior technicians diagnosing failures by ear rather than digital diagnostics.
Why it matters: Despite investor forecasts positioning humanoid robots as the defining trend of 2025, the sector faces a scalability crisis. The "hand-rubbed" manufacturing model is suited for R&D iteration but cannot support mass commercialization. Unlike automotive or smartphone industries, there's no automation infrastructure to enable exponential production scaling—adding capacity currently means adding headcount linearly. This reveals a disconnect between algorithmic advancement and industrial maturity. The sector is also demand-constrained: current orders sustain small workshops but don't justify capital-intensive automated lines. Until humanoid robots prove utility in real-world applications beyond demonstrations, the industry will remain stuck in high-cost, low-volume production. This reality check is critical for investors assessing valuations premised on near-term mass deployment.
China's Long March 12A Rocket Fails Landing in Reusability Push
China's Long March 12A rocket achieved successful liftoff but failed its first-stage vertical landing attempt during its maiden flight, crashing at the recovery zone in Gansu Province with an explosion described as a "mushroom cloud." The mission marked China's second attempt at orbital launch with booster recovery, a technology critical for reducing launch costs.
Why it matters: The failure underscores the technical challenges China faces in achieving SpaceX-style reusability, even as state-owned aerospace manufacturers aggressively pivot toward this capability. At least five new reusable rocket models are scheduled for flight testing starting 2026, including the Long March 10B (April launch) and Long March 10C, creating intense competition for cost-effective space access. The Long March 12A's use of commercial engines from Jiuzhou Cloud Arrow reflects China's strategy of blending state and private sector capabilities. While the landing failure is a setback, the broader push signals that China's "national team" is treating reusability as essential infrastructure for its space ambitions, with implications for satellite constellation deployment and commercial launch markets.
What to watch: Whether U.S. and Chinese regulators finalize H200 export approvals before Chinese New Year; Zhipu's IPO timeline and valuation; signs of automation infrastructure investment in humanoid robotics; and Long March 10B's April 2026 landing attempt.