今天早间，特斯拉CEO埃隆·马斯克在X平台发文宣布，随着AI5芯片设计完成，公司将重启超级计算机 项目Dojo 3的开发工作。此同时，他在帖子中发出了一份独特的"招聘启事"，要求应征者用3个要点介 绍自己解决过的最棘手技术难题。 埃隆·马斯克 【来源：CNMO科技】 这一战略转向看似突然，实则早有征兆。马斯克在去年8月就曾暗示："想到Dojo 3和AI6推理芯片，直 觉上我们希望能找到融合点，基本上用同一款芯片就行。"从独立超算到统一架构，特斯拉的AI芯片战 略经历了根本性转变。马斯克此前表示："当所有路径都明确指向AI6时，我不得不关闭Dojo，并做出一 些艰难的人员选择，因为Dojo2现在已经走到了进化的死胡同。"他同时确认，Dojo 3或许会以在一块主 板上集成大量AI6 SoC的形式继续存在下去。 CNMO了解到，就在五个月前，特斯拉全面叫停Dojo超级计算机项目。当时知情人士透露，这一决定由 马斯克亲自下达。当时马斯克对此解释道："特斯拉分散资源并同时开发两种截然不同的AI芯片设计是 没有意义的。" 他补充说，特斯拉的AI5、AI6及后续芯片在推理方面将表现出色，至少在训练方面相当 不错，所有努力都 ...

Core Viewpoint - Tesla is undergoing a significant strategic shift towards becoming an AI and robotics company, moving beyond its identity as merely an electric vehicle manufacturer. This transformation is centered around the Dojo supercomputer and the Cortex computing cluster, which are intended to enhance Tesla's advanced driver-assistance systems and fulfill Musk's vision of turning existing Tesla vehicles into autonomous taxis [1][5]. Group 1: Leadership Changes and Team Dynamics - Pete Bannon, Tesla's VP of Hardware Engineering, is leaving the company, which has led to the closure of the Dojo team and the reassignment of its engineers to other projects [1][4]. - Tesla has experienced a wave of talent departures this year, including key figures such as Milan Kovac, David Lau, and Omead Afshar [3]. Group 2: Dojo Supercomputer and AI Strategy - Tesla is disbanding the Dojo supercomputer project, which was initially seen as a cornerstone of its AI ambitions, and is now increasing reliance on external partners like Nvidia and Samsung for chip production [4][6]. - The Dojo project was intended to process vast amounts of video data for AI training, but discussions around it have diminished as focus shifts to the new Cortex AI training supercluster [5][6]. Group 3: Financial Agreements and Future Plans - Tesla has signed a $16.5 billion agreement with Samsung to produce more of its A16 chips domestically [2]. - Musk indicated that Tesla's AI efforts will continue to evolve, with a focus on integrating the Dojo and AI6 inference chips, which are expected to support various applications from full self-driving (FSD) to the Optimus humanoid robot [6].