Musk: AI6 Chip Performance Doubles, Will Use Samsung Texas 2nm Process
CEO Elon Musk revealed on social media that the company has completed the tape-out of the AI6 chip, 45 days ahead of schedule. He stated that the new chip uses Samsung Texas 2nm process, achieving a practical doubling of performance compared to AI5, with a comprehensive architecture optimization at the same semi-field area. The subsequent version, AI6.5, will further enhance performance based on TSMC Arizona 2nm process.
Musk pointed out that the AI6 chip integrates LPDDR6 memory design, correcting several design compromises made earlier to accelerate progress. About half of the TRIP AI acceleration units are dedicated to SRAM, resulting in a computing bandwidth in the cache that is an order of magnitude higher than traditional DRAM. This means that AI6 not only doubles performance but also achieves a qualitative change in storage access architecture.
Source: Public Information
ABAB AI Insight
AI6的发布展示了Musk在AI基础算力自研上的战略节奏化部署，其影响不仅限于性能指标，更在于“制造主权”的落地验证。三星德州与TSMC亚利桑那两条产线，象征美国AI算力基础回流本土的地缘产业趋势。Musk借芯片布局同时实现硬件能力与供应链政治的绑定。
架构层面，SRAM高占比设计重塑了AI芯片的带宽极限，预示算力竞争正从“算峰”转向“内存壁垒突破”。这与NVIDIA长期以GPU缓存结构为护城河的模式不同，AI6的路径更接近专用AI阵列的“近算架构”实验。
从更大周期看，这是硅谷AI企业自建硬件的必然阶段：当模型创新边际收益下降，自主芯片带来的性能—成本曲线优化成为关键增量。AI6如果量产成功，将加速算力在美国境内的技术堆栈内循环，重塑AI产业的基础地理版图。