Triangle

OpenAI自研AI芯片媲美英伟达！相关概念股梳理_我的网站

一 |     　　在AI大模型训练与推理需求爆发的推动下，AI芯片行业正经历深刻重构：竞争焦点从单一制程性能转向“专用架构+软件生态+能效效率”的系统级较量。

二 | 而OpenAI自研AI芯片的最新消息，无疑泛起了新的涟漪。    

The lights went down inside the National Speed Skating Oval. A humanoid robot riding on a robotic horse, lifted a bow, drew back the string and let the arrow fly.
The arrow found its mark, and the “Zhixin,” or intelligence chip — this competition’s equivalent to the Olympic flame — lit up above the arena. The venue erupted in cheers.
With that, the 2nd World Humanoid Robot Games opened on Saturday at the Beijing venue better known as the “Ice Ribbon” — the capital’s second robotics showcase in a week, following the 2026 World Robot Conference, which opened on August 19.
The opening ceremony began with demonstrations of the 100-meter sprint and the standing high jump. In the sprint, a humanoid robot clocked 9.39 seconds — faster than the men's 100-meter world record of 9.58 seconds set by Jamaican sprint legend Usain Bolt. In the high jump, a humanoid robot reached 288.43 centimeters. Both marks were set by the Tien Kung Ultra humanoid robot by Beijing-based X-Humanoid. Table tennis, tennis, football and other events were then demonstrated jointly by elite human athletes and robots.
The parade of international delegations and robot contingents followed, including formations of humanoid robots from Booster Robotics, Galbot and Noetix Robotics. The robots performed drills as they marched, all of it done autonomously with no remote control.
The X-Humanoid fielded a 50-strong contingent of its Tien Kung 3.0 robots. Tien Kung 3.0 also served as the only robot flag bearer of the games, marching at the head of all contingents while waving the games flag, and it is also the humanoid robot archer who lighted the “Zhixin”, the company told the Global Times.
More than a thousand robots appeared over the course of the ceremony, in a series of joint performances with humans — an opening overture played by humanoid robot musicians alongside a human orchestra, and segments with pop singers, with the judges of the dance events taking the floor themselves, and with children.
Taishan Sports, the Shandong-based official equipment supplier of the games, provided the running tracks, the artificial turf for the football pitches, the floor surfacing for events including the long jump, high jump and weightlifting, and the impact mats that slow the robots down at the end of the sprint lanes.
"Robots are very heavy, and what they need from a surface in terms of hardness and friction is not what a human runner needs, so we adjusted the material formula specifically," Man Fangfang, the vice director of the company, told the Global Times at the venue. The artificial turf was reworked the same way — robots do not play football the way people do, and the turf has to be non-slip and antistatic. 
The company spent months talking to the competing teams and retesting the gear before the games. Taishan Sports has supplied equipment to eight Olympic Games and more than 3,000 sporting events, "but this is our first time serving robot games," she said.
China's robotics industry is growing fast and robots are moving into every sector, while the sports industry has always lent itself to crossover with other fields, Man said, adding that the company wants to use the games to work out what equipment suits robots and to develop more of it.
Compared with the first edition, the second games have expanded in both scale and quality. A total of 666 teams and 2,056 robots from 16 countries across six continents are competing, with the number of teams up 138 percent from the first edition and the number of robots quadrupling, according to CCTV.
Competition events have doubled from 26 to 51, while the number of matches has more than doubled from 487 to 1,301. More combative and spectator-friendly events such as martial arts and dance have been added to the competitive program.
Notably, the second games set stricter autonomy requirements for the robots. Except for the 100-meter and 400-meter obstacle races, all track events and other competitive events must be completed by robots fully autonomously. In scenario-based events, a task completed fully autonomously carries a final score weighting coefficient of 1, while one completed by remote control carries only 0.5, per CCTV.
Analysts said the higher bar on autonomy is in line with the industry’s development goal of having robots complete tasks without human operation.
。　　据财联社报道，OpenAI自研的推理芯片Jalapeno迎来了新的突破，在测试中的表现优于英伟达当前的产品阵容。　　OpenAI芯片业务负责人Richard Ho在接受采访时表示，Jalapeno芯片与英伟达GB300进行的比较测试中，Jalapeno在两个指标上占据优势：单位功耗所能处理的AI工作量，以及响应的速度。　　需要指出的是，Jalapeno并未与英伟达最新一代的Vera Rubin芯片进行测试，后者刚刚开始出货。此外，Jalapeno也并非为AI模型训练而设计，其主要面向AI推理阶段，即模型完成训练后，根据用户提示生成响应并执行任务的阶段。

三 | 而训练则是英伟达技术最擅长的领域之一。　　公开资料显示，除了英伟达这一行业领跑者之外，OpenAI等全球科技巨头均在积极自研AI芯片，比如亚马逊自研的Trainium系列、Inferentia系列芯片，微软自研的Maia系列芯片，Meta自研的MTIA系列芯片，特斯拉自研的Dojo D1、AI5等芯片，谷歌自研的TPU系列芯片。

四 | 　　国内龙头同样在积极自研AI芯片，比如阿里巴巴的含光、真武系列，百度的昆仑芯，摩尔线程的长江、华山等，快手的SL200，以及字节跳动的SeedChip等等。

五 | 　　头豹研究院表示，巨头们积极自研AI芯片，背后的深层次原因在于美国持续收紧先进芯片及制造设备出口管制，叠加全球AI资本开支向头部云厂商高度集中，加速了算力供应链的地缘割裂，各国纷纷押注自主生态。

六 | 　　亿欧智库表示，当前国产替代已告别零散的技术突破，进入全产业链协同攻坚的新阶段，AI芯片已完成“设计制造-设备-材料”全链条布局。根据三方机构预测，到2030年，中国AI芯片自给率将升至86%，市场规模突破670亿美元。　　从产业链来看，亿欧智库表示，可关注EDA/IP（华大九天、概伦电子、广立微等）、设备（盛美上海、中微公司、华海清科等）、芯片设计（寒武纪、海光信息、云天励飞等）、封装测试（通富微电、长电科技、华天科技等）等核心环节。

七 | （文章来源：东方财富研究中心）。

Current article:http://fcm3.chengjuecannaobianmouqueduoyu.shop/8iw30w9/20260826/7454282.html

Published on:12:42:37