{
 "title": "China Dominates Robot Market at 85% Global Share | Robotics & Semiconductor Briefing",
 "description": "China Dominates Robot Market at 85% Global Share | Robotics & Semiconductor Briefing\n\nToday's highlights: Zhipu GLM-5.2 goes open source—1 million token context, MIT license, AI coding tool ZCode 3.0 released simultaneously.\n\nKey scenes covered in the show:\n- Today's top 5 stories\n- Zhipu GLM-5.2 open source and DeepSeek V4 adapting to Chinese chips\n- Roboterra's logistics deployment and humanoid robot IPO rush\n- Enflame Tech IPO approval, China's four GPU leaders gather\n- Changxin Memory HBM3 shipments to Huawei and memory price surge\n\nLeave a comment with a topic from today's news you'd like us to dive deeper into.\nIf you see a Hype button within the last 7 days of uploads, please give it a tap—it helps us reach new viewers.\n\nThis show is brought to you by lawyer Kyungjin Kim.\n\n#AI #ArtificialIntelligence #KyungjinKim",
 "link": "https://kimkj.com/%EC%98%A4%EB%8A%98%EC%9E%90-%EB%89%B4%EC%8A%A4-%EC%98%81%EC%83%81/?mod=document&uid=5949",
 "segments": [
  {
   "id": "S001",
   "slide": 1,
   "speaker": "Host",
   "text": "This show is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S002",
   "slide": 1,
   "speaker": "Guest",
   "text": "This week, Chinese AI stocks rallied across the board. What moved them?"
  },
  {
   "id": "S003",
   "slide": 1,
   "speaker": "Host",
   "text": "On the morning of June 15th, China's CS Artificial Intelligence index jumped 4.61 percent. Guangxun Technology hit the daily limit, Inspur Information climbed 9.7 percent, and Cambricon rose in tandem. According to OpenRouter's tally, global AI model calls hit 44.6 trillion tokens—up 23.5 percent from last week and showing eight consecutive weeks of growth. Behind this momentum, five events converged this week. Zhipu released GLM-5.2 as open source, Enflame Tech cleared its IPO board, Changxin Memory started delivering HBM3 to Huawei, and Roboterra placed humanoid robots in ten logistics centers. AI memory prices are up 682 percent on a DDR5 basis."
  },
  {
   "id": "S004",
   "slide": 2,
   "speaker": "Guest",
   "text": "Zhipu dropped another model. That's the third one in three months—what's different this time?"
  },
  {
   "id": "S005",
   "slide": 2,
   "speaker": "Host",
   "text": "5.0 to 5.1 was a performance upgrade. 5.1 to 5.2 is a shift from internal testing to public release. It supports a million-token context and ships under the MIT license, so anyone can take it and use it commercially. The same day they also launched AI coding tool ZCode 3.0, which uses their in-house-developed ZCode Agent engine to optimize execution paths for large-scale projects. Just as Anthropic is tightening access to their models, Zhipu is releasing an open-source model and a coding agent at the same time—staking out their own territory."
  },
  {
   "id": "S006",
   "slide": 2,
   "speaker": "Guest",
   "text": "DeepSeek V4 is also running on Chinese chips, I hear?"
  },
  {
   "id": "S007",
   "slide": 2,
   "speaker": "Host",
   "text": "The FlagOS community completed multi-chip adaptation for DeepSeek V4. Nine types: Huawei Ascend, Haiguang, Wushi, Biyin, Kunlun, Pingtou Ge, and Tengshu. They proved you can run trillion-parameter models with just Chinese chips. In this flow where software catches up to hardware, a low-cost Chinese open-source coding model could seep into Korean company dev teams. If you send code to an external model, assets leak out. If you put off checking security guidelines, the company ends up liable for the breach."
  },
  {
   "id": "S008",
   "slide": 3,
   "speaker": "Host",
   "text": "This show is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S009",
   "slide": 3,
   "speaker": "Guest",
   "text": "Humanoid robots actually went into a real logistics center—how advanced are they?"
  },
  {
   "id": "S010",
   "slide": 3,
   "speaker": "Host",
   "text": "Roboterra placed humanoid robots in ten logistics centers operated by China Post and SF Logistics. They sort 1,200 packages per hour and work 24/7 without rest. They say the robots hit 85 percent of human work efficiency. In a funding round led by SF Group, they secured 2 billion yuan—about 280 million dollars—with Sequoia China and IDG Capital participating. That said, we also saw cases where human workers processed 192 more packages per hour than the robots. It's moved from demos into real operations, but we're not yet at the replacement stage."
  },
  {
   "id": "S011",
   "slide": 3,
   "speaker": "Guest",
   "text": "Robot companies are also speeding up their IPO push, right?"
  },
  {
   "id": "S012",
   "slide": 3,
   "speaker": "Host",
   "text": "EngineAI quietly filed for a Hong Kong IPO. Three years old, and the company's valued at 10 billion yuan—they're churning out one robot every 15 minutes from a single factory. Unitree passed the Sci-Tech Board review and is getting close to being the first domestically-made robot to list on the A-share market, and LEBOT is pushing for listing at a 6 billion dollar valuation. Over 150 companies are competing, but customer satisfaction sits at just 23 percent. Supply's massively outpacing demand in this market, and only the handful with real revenue will survive."
  },
  {
   "id": "S013",
   "slide": 4,
   "speaker": "Guest",
   "text": "What does Enflame's IPO approval mean for China's GPU landscape?"
  },
  {
   "id": "S014",
   "slide": 4,
   "speaker": "Host",
   "text": "On June 15th, Enflame Tech cleared its Sci-Tech Board IPO review. They filed in January, went through two rounds of questions, and cleared it in five months. They're putting 600 million yuan in fundraising toward developing 5th and 6th generation AI chips. Their revenue forecast for the first half of 2026 is 1.06 to 1.15 billion yuan—a 259 to 289 percent increase year over year."
  },
  {
   "id": "S015",
   "slide": 4,
   "speaker": "Guest",
   "text": "I keep hearing about the 'four GPU dragons'—what does that mean?"
  },
  {
   "id": "S016",
   "slide": 4,
   "speaker": "Host",
   "text": "Cambricon, Haiguang, Wushi, and now Enflame—China's four GPU dragons have all gathered on public markets. What's crucial is that Enflame is a company that's spent eight years building its own 5th generation AI chip architecture independently. This listing means China's AI chip ecosystem has moved beyond the venture capital stage and into capital market validation. When listing rushes bring R&D funding in fast, the competition Korean fabless designers face in the Chinese market gets fiercer."
  },
  {
   "id": "S017",
   "slide": 5,
   "speaker": "Guest",
   "text": "Changxin Memory started shipping HBM3 to Huawei—how serious is that from Korea's perspective?"
  },
  {
   "id": "S018",
   "slide": 5,
   "speaker": "Host",
   "text": "Changxin Memory passed 16-nanometer HBM3 samples to Huawei. Mass production in 2026, HBM3E launch in 2027 is the target. In the second half of 2025, average yield exceeded 80 percent—approaching Samsung's level. They adopted the same MR-MUF packaging technology as SK Hynix, and their Shanghai backend facility is set to start operations at year-end. The tech gap with Korea has narrowed to two to three years. The only way SK Hynix maintains that gap is by moving to HBM4 faster than anyone."
  },
  {
   "id": "S019",
   "slide": 5,
   "speaker": "Guest",
   "text": "How are memory prices moving?"
  },
  {
   "id": "S020",
   "slide": 5,
   "speaker": "Host",
   "text": "As AI data center demand explodes, DDR5 prices jumped 682 percent. NAND flash is up 807 percent. HBM's taking up 23 percent of DRAM wafer capacity, which shrinks the supply for general-purpose memory by that much. Samsung Electronics stock is up 114 percent since the start of the year, SK Hynix is up 186 percent. It's a boom period for Korean memory companies, but if Changxin pulls off HBM3 mass production, Korea's market share in China takes a hit—it's a double-edged moment."
  },
  {
   "id": "S021",
   "slide": 6,
   "speaker": "Host",
   "text": "This show is brought to you by lawyer Kyungjin Kim."
  },
  {
   "id": "S022",
   "slide": 6,
   "speaker": "Guest",
   "text": "I hear Chinese telecom companies are selling tokens as a pricing plan—how does that structure work?"
  },
  {
   "id": "S023",
   "slide": 6,
   "speaker": "Host",
   "text": "According to the National Data Bureau, China's average daily token consumption went from 100 billion in early 2024 to 140 trillion in March 2026. That's 1,400 times in two years. It's the result of the big three telecom carriers rolling out token pricing plans in earnest. In Beijing, you can buy a basic computing package for 5.99 yuan—about 1,100 Korean won. China Mobile set up a token operations alliance with seven partners: Tencent, Alibaba, Huawei, and others. Just like data plans pushed out voice plans, tokens are moving up to become the new unit of product for telecom carriers."
  },
  {
   "id": "S024",
   "slide": 6,
   "speaker": "Guest",
   "text": "MWC Shanghai is next week, right?"
  },
  {
   "id": "S025",
   "slide": 6,
   "speaker": "Host",
   "text": "June 24th through 26th. A new program they created this year is a humanoid robot soccer penalty-kick challenge. Warm-ups on the 23rd, preliminaries on the 24th and 25th. The Mobile AI Innovation Frontier zone has advanced chipsets, mega-scale AI servers, and frontier model demos scheduled. The fact that they added robot soccer to a telecom trade show tells you something—in China's industrial structure, telecom, AI, and robotics move as one package."
  },
  {
   "id": "S026",
   "slide": 7,
   "speaker": "Guest",
   "text": "On the policy side, what's shifted?"
  },
  {
   "id": "S027",
   "slide": 7,
   "speaker": "Host",
   "text": "On June 10th, the Ministry of Industry and Information Technology released a three-year AI+ICT implementation plan. Of 17 initiatives, three are the core: achieve 75 percent one-millisecond latency coverage in metropolitan area compute networks, build 30-plus high-value scenarios, realize autonomous intelligence in telecom networks. By 2030, the end goal is unified integration of network, sensing, computing, and intelligence. Algorithmically-registered generative AI services exceeded 868 by late April. The algorithm registration system is effectively functioning as an entry barrier for China's AI market."
  },
  {
   "id": "S028",
   "slide": 7,
   "speaker": "Guest",
   "text": "How far have US regulations gone?"
  },
  {
   "id": "S029",
   "slide": 7,
   "speaker": "Host",
   "text": "On May 31st, the BIS issued new guidance. If the parent company is in China, overseas subsidiaries are also subject to advanced AI chip export authorization. It's meant to shut down workarounds like buying through Southeast Asian and Middle Eastern subsidiaries. Since January, Chinese customs has effectively blocked Nvidia H200 imports, and in March Nvidia stopped making H200s and reallocated TSMC capacity toward next-gen Vera Rubin. Congress is pushing for an outright ban on semiconductor manufacturing equipment exports to China."
  },
  {
   "id": "S030",
   "slide": 8,
   "speaker": "Guest",
   "text": "Korean semiconductor exports surged—you said it wasn't all good news, though."
  },
  {
   "id": "S031",
   "slide": 8,
   "speaker": "Host",
   "text": "In May, Korea's semiconductor exports to China jumped 243 percent year-over-year. The trade balance with China swung to a 3.8 billion dollar surplus—flipping from a 760 million dollar deficit in December 2025 in just six months. HBM and DDR5 for AI data centers drove it. The risk has two edges. If Changxin pulls off HBM3 mass production, Korea's memory market share in China shrinks, and if the US adds HBM to its China export restrictions list, Korean companies get squeezed between the US and China."
  },
  {
   "id": "S032",
   "slide": 8,
   "speaker": "Guest",
   "text": "What should I check first thing tomorrow morning?"
  },
  {
   "id": "S033",
   "slide": 8,
   "speaker": "Host",
   "text": "Three things to watch. Is SK Hynix's HBM4 transition schedule faster than Changxin's HBM3 mass production? When will the US BIS add HBM to its China sanctions list? Where are the four GPU dragons, including Enflame, channeling their IPO money—which process nodes? Next week's MWC Shanghai has chipset and AI server demos scheduled, so cross-reference the numbers that come out of there with these three variables."
  },
  {
   "id": "S034",
   "slide": 8,
   "speaker": "Host",
   "text": "This show is brought to you by lawyer Kyungjin Kim."
  }
 ]
}