馬斯克在韓國招募半導體人才強化AI晶片競爭

• AI5 Specifications: Designed for in-vehicle inference running Full Self-Driving neural networks; targets state-of-the-art performance-per-watt for AI inference with fraction of Nvidia GPU power draw[2] • Manufacturing Strategy: Dual-foundry approach leveraging Samsung's Hwaseong facility and TSMC for parallel production to achieve record-scale volumes[2][5] • Design Methodology: Tesla adopting agile hardware development using advanced electronic design automation (EDA) tools and potentially AI-assisted simulation to compress traditional 18-24 month cycles to 9 months[2][3] • AI6 Specifications: Samsung signed $16.5 billion deal to manufacture AI6 chips at Taylor, Texas fab beginning 2027; designed to power Tesla vehicles, Optimus robots, and enable high-performance AI training in data centers[4] • Dojo 3 Architecture: Restarted training supercomputer now optimized for space-based AI compute infrastructure rather than terrestrial autonomous driving model training[3][4] • Multi-Processor Roadmap: AI5 and AI6 represent milestones in broader roadmap extending to AI7, AI8, and AI9 with nine-month generational cycles, supporting in-vehicle autonomy, robotics processors, and training silicon[2][3]

Tesla's Korea-based recruitment and accelerated chip design cycles position the company to achieve unprecedented vertical integration in AI silicon, potentially reducing dependency on external chip suppliers and enabling tighter control over autonomous vehicle and robotics performance trajectories. The dual-foundry strategy and aggressive 9-month design cadence could establish new industry benchmarks for AI chip development velocity, forcing competitors to reconsider traditional 18-24 month cycles. Success in space-based AI compute via Dojo 3 would expand Tesla's addressable market beyond automotive and robotics into infrastructure computing. However, the realistic near-term outcome may involve hybrid approaches where Tesla expands internal capability while leveraging established compute ecosystems for frontier-model training[1]. For the broader AI chip market, Tesla's vertical integration strategy challenges the fabless model dominance and could inspire other automotive and robotics companies to develop proprietary silicon, fragmenting the market away from Nvidia's current dominance in AI inference and training.