文章

汤臣倍健跨界AI · 谷歌Transformer之父跳槽

#775 · 2026-06-18 · 21ZHAO Blog
Reading Path / ARTICLE 先抓主张，再转成行动 #775 · 21ZHAO Blog · 读完进入产品或下一篇

汤臣倍健跨界AI · 谷歌Transformer之父跳槽

一、 权威必看

EN: On June 18, 2026, Tangchen Biotech announced a strategic financial investment in Yuanli Semiconductor, marking a significant cross-industry move by a traditional health enterprise into the semiconductor sector. The company allocated 50 million yuan to acquire a 0.97% stake in the chipmaker, which specializes in Chiplet-based AI inference chips for edge computing. This decision reflects a broader trend where established manufacturing and consumer goods firms are seeking to establish cognitive windows into frontier technologies to secure long-term value appreciation.

中： 2026年6月18日，汤臣倍健股份有限公司发布公告，宣布以自有资金5000万元投资原粒（北京）半导体技术有限公司，从而持有其0.97%的股权。此举标志着传统保健品巨头正式跨界进入半导体领域，旨在通过财务性投资建立对前沿科技领域的认知窗口。原粒半导体成立于2023年4月，专注于基于Chiplet技术创新的积木式AI推理芯片，为边缘端大模型部署提供高性价比算力方案。尽管本次投资由董事长梁允超配偶间接持股构成关联交易，且原粒半导体2025年净利润为负6295.9万元，但汤臣倍健强调此举符合股东利益及国家创新驱动发展战略，不会影响公司正常生产经营。

二、 深度与多元

EN: The artificial intelligence landscape witnessed a pivotal talent shift on June 18, as Noam Shazeer, the co-lead of Google’s Gemini model and a key figure in Transformer architecture development, announced his departure from Google to join OpenAI. This high-profile move signals intensifying competition for top-tier AI researchers between the two tech giants. Shazeer expressed pride in his time at Google but highlighted his anticipation for collaborating with OpenAI’s team, suggesting that the battle for algorithmic supremacy is now heavily dependent on human capital mobility.

中： 6月18日，AI领域发生重磅人才流动事件：谷歌Gemini模型联合负责人、Transformer架构核心贡献者诺姆·沙泽尔在X平台宣布离职谷歌，正式加入OpenAI。这一举动不仅揭示了科技巨头间对顶尖AI研究人才的激烈争夺，也预示着算法竞赛进入新阶段。沙泽尔在声明中表达了对谷歌团队的自豪，同时表示期待与OpenAI团队共事。作为AI圈的传奇人物，他的跳槽引发了行业广泛关注，反映出当前人工智能竞争已从单纯的算力堆砌转向核心人才与算法创新的深度博弈。这种高层人才的流动往往被视为两家巨头在技术路线和市场竞争中争夺身位的重要信号。

三、 科技与财经

EN: Following reports of a massive financing round exceeding 50 billion yuan, DeepSeek officially launched its image recognition mode on the web. However, early tests revealed significant limitations; when presented with photos of its founder Liang Wenfeng, the model failed to identify him correctly, offering erroneous guesses such as Yang Zhilin or a young Ma Huateng before admitting uncertainty. This incident highlights the gap between DeepSeek’s financial valuation and its current multimodal capabilities, despite the core investment terms prioritizing talent retention over external control.

中： 在超500亿元首轮融资消息曝光后，DeepSeek于6月18日上线网页版识图模式，但实际测试效果引发关注。澎湃新闻记者及多位网友使用创始人梁文锋的照片进行测试，模型在经过几分钟思考后给出了错误答案，甚至将其误认为杨植麟或年轻版的马化腾，最终承认“不推荐猜测名字”。这一现象揭示了DeepSeek在获得腾讯、宁德时代等巨头巨额投资（梁文锋个人出资约200亿元）后，其多模态技术仍处于初始阶段。尽管融资条款中梁文锋强调“不要挖人”，但技术落地的实际表现与资本热度之间仍存在明显落差，显示出AI应用在垂直领域的复杂性。

四、 国际视野

EN: In a move highlighting the global importance of skilled labor in infrastructure, Quan Qingdong, a senior technician from CRRC Dalian, shared his experiences at a press conference organized by the State Council Information Office on June 18. Having spent 25 years in railway maintenance, Quan overcame extreme conditions on the Qinghai-Tibet Plateau and contributed to the Lhasa-Nyingchi Railway project. His work, which improved fault diagnosis accuracy by 80% through self-developed online detection devices, underscores the critical role of grassroots technical innovation in national strategic projects.

中： 6月18日，国新办举行“新征程上的奋斗者”中外记者见面会，中车大连机车车辆有限公司服务经理权庆东受邀出席。权庆东深耕轨道交通售后服务25年，曾在海拔4000米以上的青藏铁路克服严寒缺氧，攻克百余项高原机车技术难题。在拉林铁路国家重点项目中，他搭建全流程运维保障体系，并自主研发在线检测装置，将故障诊断准确率提升80%。权庆东的故事展示了中国一线产业工人在“大国重器”维护中的关键作用，其拥有的6项专利和2项先进操作法，体现了基层技术创新对国家基础设施安全运行的坚实支撑。

五、 青年与生活

EN: JPMorgan Chase significantly revised its global semiconductor equipment market forecast, predicting a 28% growth in 2026 and 29% in 2027 for the WFE market. The bank attributes this surge to cloud giants like Google, Amazon, Microsoft, and Meta accelerating AI infrastructure investments, with their combined capital expenditures expected to reach $575 billion in 2026. This shift indicates that AI-driven demand is fundamentally reshaping the semiconductor supply chain, moving from speculative growth to substantial industrial expansion.

中： 摩根大通在6月18日发布的报告中大幅上调全球晶圆厂设备（WFE）市场预测，预计2026年和2027年增速分别为28%和29%。这一乐观预期主要源于谷歌、亚马逊、微软和Meta四大云服务商正在以前所未有的速度扩大AI基础设施投资。摩根大通测算，这四家公司2026年的资本开支总额将突破5750亿美元，占经营现金流比例将从过去的25%-30%飙升至82%。这表明AI资本开支已进入加速阶段，全球半导体设备市场正从周期性波动转向由AI需求驱动的结构性增长，为相关产业链带来明确的增长信号。

【21ZHAO 综合判断】

EN: The convergence of traditional manufacturing investing in AI chips, top talent shifting between AI giants, and massive capital expenditure by cloud providers paints a picture of an industry in rapid structural transformation. For developers and investors, the key takeaway is that while financial valuations may outpace technical maturity (as seen with DeepSeek’s image recognition), the underlying infrastructure demand is undeniable. The shift of experts like Shazeer to OpenAI suggests that algorithmic breakthroughs remain the primary bottleneck, necessitating a focus on core R&D rather than just application layer hype.

  • For Developers: Focus on edge-computing efficiency and Chiplet integration technologies, as traditional firms like Tangchen Biotech are entering this space, indicating future market opportunities in specialized hardware-software co-design.
  • For Investors: Monitor the capital expenditure trends of the top four cloud providers closely; their 82% capex-to-cash-flow ratio in 2026 signals a sustained boom in semiconductor equipment demand, making WFE stocks a critical sector to watch.

中： 当前，传统制造业跨界投资AI芯片、顶尖人才在巨头间流动以及云厂商巨额资本开支，共同描绘了一个行业正在经历剧烈结构转型的图景。对于开发者和投资者而言，核心启示在于：尽管金融估值可能超越技术成熟度（如DeepSeek识图模式的初期表现），但底层基础设施的需求是确定无疑的。沙泽尔等专家流向OpenAI表明，算法突破仍是主要瓶颈，因此应聚焦核心研发而非应用层的炒作。

  • 对开发者建议： 关注边缘计算效率和Chiplet集成技术，随着汤臣倍健等传统企业进入该领域，专用软硬件协同设计未来可能存在市场机会。
  • 对投资者建议： 密切监测四大云厂商的资本开支趋势；其2026年82%的资本开支占现金流比例信号表明半导体设备需求将持续繁荣，WFE板块值得重点关注。

参考来源