中央处理器（CPU）最初在1970年代的袖珍计算器中流行，之后在超过五十年的芯片产业中长期处于主导地位，通过运行操作系统和复杂数据处理支持手机、笔记本和超级计算机。AI 需求改变了格局，大规模数据负载暴露了通用 CPU 在高速大规模矩阵乘法上的局限。Nvidia 的高端 GPU，单颗售价高于美金 30,000 美元（US$30,000），已成为 AI 加速器的主流架构，Nvidia 现在接近拿下芯片产业约三分之一的销售。曾是营收领先者的 Intel 在三年前失去领先后，销售已下降。到2026年，争论焦点不再是普通计算，而是是否能大规模生产和部署足够的 AI 加速器。
在 Nvidia 于加州圣何塞的会议上，Nvidia 发布 Vera，并称其为第一款独立销售的 CPU，不再与加速器捆绑，显示了对独立 CPU 功能的复兴需求。Arm 也从依赖许可的架构/IP转向销售实体芯片，从智能手机起家后扩展到数据中心定位。其最新预测显示，营收在五年内可达当前的五倍。首席执行官 Rene Haas 多次表示，客户反复要求 Arm 做这件事，而不是把它视为与客户竞争。
新的解读是，AI 加速器在 AI 模型与服务执行方面表现卓越，但 CPU 仍然承担关键编排：数据准备、与其他组件通信，以及 token 流向的路由。短期市场反应符合这一逻辑：Arm 在过去一年多来达到最佳交易日，Intel 和 Advanced Micro Devices 都上涨 7%。Nvidia CEO Jensen Huang 表示，独立的 Vera CPU 将给 Nvidia 已有的巨大营收再添数十亿美元。更广泛的观点是，随着 AI 软件增长放缓进入更稳定阶段，CPU 可能再次成为大多数工作负载的核心，而不只是加速器的备份。
The central processing unit (CPU), first popularized from 1970s pocket calculators, led the chip industry for more than fifty years, supporting phones, laptops, and supercomputers by running operating systems and complex data processing. AI demand changed the equation, as massive data workloads exposed limits of general CPUs in fast large-scale matrix multiplication. Nvidia’s high-end GPUs, priced above US$30,000 per chip, have become the mainstream architecture for AI accelerators, and Nvidia is now close to capturing about one-third of chip-industry sales. Intel, once the revenue leader, has seen sales decline after losing its top position three years ago. In 2026, the debate is no longer about ordinary computing, but whether enough AI accelerators can be mass-produced and deployed.
At Nvidia’s conference in San Jose, California, Nvidia introduced Vera as the first standalone CPU sold separately, signaling renewed demand for independent CPU capability. Arm also moved from licensing architecture/IP to selling physical chips, extending from its smartphone roots into data-center positioning. Its latest forecast says revenue could reach five times the current level within five years. Chief Executive Officer Rene Haas repeatedly said customers asked Arm to do this repeatedly, rather than seeing it as a move against customers.
The emerging interpretation is that AI accelerators are superior at AI model and service execution, but CPUs still perform essential orchestration work: data preparation, communication with other components, and token routing. Near-term market reactions match this view: Arm posted its best trading day in more than a year, and Intel and Advanced Micro Devices both rose 7%. Nvidia CEO Jensen Huang said the standalone Vera CPU would add tens of billions to Nvidia’s already large revenue. A broader view is that once AI software growth slows into a steadier phase, CPUs may again become central for most workloads, not merely backups to accelerators.