Zhipu AI Nears Opus 🧠 Lilly’s $1.9B China Deal 💊 Xiaomi Driverless Lap 🚗

2026-06-29 | subscribe | homepage

Welcome back to China Insights Weekly. Here are some of the key highlights for this week’s edition:

• YMTC climbs to 13% NAND share, as China’s memory-chip race accelerates

• Visa-free travel lifts inbound tourism, with Southeast Asia and the Middle East surging

• Fusion magnets go fully domestic, moving China closer to practical clean power

• Reusable rockets prepare second takeoff, as constellations drive launch demand

Remember to follow us on LinkedIn, X or Facebook.

🚀 Headlines

Bytedance to release Seedance 2.5 upgrade, China’s Zhipu AI is closing in on top US models at 5th of the cost (link, link)

ByteDance has introduced Seedance 2.5, its latest AI video model, during a conference in Beijing. The model can generate 30‑second 4K clips from a single prompt and now supports up to 50 reference pieces (images, videos, or audio), up from 12 in its predecessor. It will launch in China next month, with no timeline given for other markets. The new model is significantly improved from its predecessor, and ByteDance continues to release some of the most advanced video‑generation AI, rivaling OpenAI’s Sora and Google’s Veo 3. The model offers greater creative control. The message is clear: China's AI video race is accelerating, and ByteDance is leading the charge.

Zhipu's GLM 5.2 has landed within a percentage point of Anthropic's Opus 4.8 on a key agentic benchmark, at roughly a fifth of the cost. OpenRouter token traffic climbed faster than after DeepSeek's V4 launch. GLM-5.2 is available under an MIT license and uses a novel IndexShare architecture that improves speculative decoding acceptance length by 20%. On SWE-bench Pro, it scored 62.1, outperforming GPT-5.5 (58.6). It ranks second globally on Code Arena. The open-source model is free to download and fine-tune, putting pressure on closed-source frontier labs. With frontier token spend straining budgets, enterprises are now focused on intelligence per dollar, making Zhipu's model an attractive alternative. The US may have the frontier models, but China is winning the battle for enterprise adoption. The cost gap is decisive.

Eli Lilly reunites with Shanghai’s Abbisko in USD 1.9 billion multiple target R&D deal (link)

Eli Lilly, an American multinational pharmaceutical company, has expanded its partnership with Abbisko Therapeutics, committing up to USD 1.9 billion for the Chinese biotech to apply its discovery and early development capabilities to multiple Lilly-selected targets. The deal follows an earlier 2022 collaboration on a cardiometabolic candidate. Abbisko, a Shanghai-based oncology specialist, also has preclinical obesity and eczema programmes. The move mirrors recent Bristol Myers Squibb and Pfizer deals that channel internal programs through Chinese partners to accelerate early development. Lilly will pay an undisclosed upfront fee, with Abbisko eligible for milestones and royalties. The agreement continues Lilly’s aggressive dealmaking pace, fuelled by GLP-1 revenues. In the past month alone, Lilly has struck deals to acquire three vaccine developers, a neuroscience startup, and to license assets from multiple biotechs. The message is clear: China’s early-stage R&D speed is becoming integral to global pharma strategy. The deeper the ties, the faster the pipeline.

YMTC NAND market share climbs to 13% as global competition intensifies (link)

Yangtze Memory Technologies (YMTC) has raised its global NAND flash market share from 8% to 13% in a year, becoming the fastest-growing player in the sector, according to Counterpoint Research. Samsung still leads with 29%, followed by SK Hynix at 18%, while YMTC now ranks fourth. Korean industry insiders acknowledge that the pace of China’s memory chipmakers has exceeded expectations, directly threatening the market positions of the two Korean giants. YMTC has recorded double-digit growth for three consecutive quarters, with Q1 2026 revenue reaching USD 2.6 billion, a staggering 445% year-on-year increase. The data suggests that China’s memory ambitions are no longer a distant threat; they are reshaping the global NAND landscape in real time. For Samsung and SK Hynix, the window to respond is narrowing. For YMTC, the trajectory is clear: catch up, then overtake. The race for NAND supremacy is now a three-horse contest. And the Chinese contender is gaining ground fast.

Xiaomi YU7 GT secures world’s first driverless Nürburgring lap in 10 minutes (link)

Xiaomi's YU7 GT electric SUV has become the first production‑ready vehicle to complete a fully automated lap of the Nürburgring Nordschleife, recording a time of 10 minutes 29.483 seconds, exactly 3 minutes 7 seconds slower than the manual SUV record of 7:22.755 set earlier by the same model. The vehicle delivers 1,003 hp (738 kW), accelerates from 0‑100 km/h in 2.92 seconds, and achieves a top speed of 300 km/h. Range is 705 km (CLTC), with 570 km rechargeable in 15 minutes. Priced from 389,900 yuan (USD 57,596). The driverless lap sets an initial benchmark for production autonomous software. As one engineer put it: This is not about beating human drivers; it is about proving that autonomous systems can handle the world's toughest track. The real race is just beginning.

China's Lingsheng supercomputer tops global rankings with 2.19 EFLOPS (link)

At the ISC 2026 conference in Hamburg, China’s domestically built Lingsheng supercomputer topped the TOP500 list with a sustained performance of 2.19 EFLOPS, ending a nine-year gap since China last held the No. 1 spot. The system is built entirely with indigenous processors and components, demonstrating that China’s homegrown chips can compete at the highest levels of scientific computing. China had stopped submitting data due to US export controls, but its publicly available research papers provide transparency. The TOP500, maintained since 1993, ranks machines by LINPACK benchmark performance. Lingsheng’s return signals the effectiveness of China’s domestic chip strategy and has implications for AI training, climate modelling, drug discovery, and materials science. China is back at the top, and it is running on homegrown silicon.