
由 AI 智能体根据文末列出的来源撰写。 我们的工作方式
英伟达悄悄向一位美国参议员确认,技术上可以实施《芯片安全法》提出的地理追踪措施,这一进展可能很快影响到购买其先进 GPU 的任何小公司。此消息正值业界对追踪要求的反对声中,业界称这些要求“新颖、未经测试且可能不可行”。
据《MIT Technology Review》报道,英伟达的代表告诉伊利诺伊州议员比尔·福斯特,公司可以实施《芯片安全法》中概述的反走私保障措施。福斯特是该法案的共同提案人,他表示英伟达团队的回应是“是的,当然可以”,尽管他们没有提供书面确认。该法案于2025年5月提出,要求对先进 AI 芯片进行地理追踪,以验证其位置并防止走私至中国。
包括英伟达在内的半导体行业协会于2026年3月公开反对该法案,称追踪要求“新颖、未经测试且可能不可行”。然而,与福斯特的私下讨论表明英伟达已经具备满足法律要求的技术能力。
英伟达自身的产品历史支持了这一说法。2025年12月,公司推出了“Fleet Intelligence”,这是一款让客户监控数据中心各站点芯片性能的工具,并能指示芯片所在的国家,路透社已证实。当《MIT Technology Review》询问 Fleet Intelligence 是否可以重新用于阻止走私时,英伟达未直接回答,而是发表声明称“政府和媒体发现受控美国产品的转移微乎其微——相较于中国已有的大量国内算力,这只是杯水车薪”。
《芯片安全法》将允许三种验证方式:设施检查、公司声明或在芯片通电时广播其位置的“基于 ping 的”定位信号。安全与新兴技术中心在2026年9月将基于 ping 的方法确定为最具成本效益的方案。如果该法案作为参议院《国防授权法案》的一部分通过,追踪仅会显示芯片位置;不会自动禁用走私硬件。福斯特计划在明年推出更严格的立法,要求在 GPU 连接前进行许可证验证。
依赖英伟达高端 GPU 的小公司——无论是用于 AI 模型训练、边缘计算还是数据中心服务——可能很快面临此前可选的强制合规步骤。实施基于 ping 的追踪可能需要固件更新、与监控平台的集成以及向监管机构持续报告。对于 IT 预算有限的初创企业来说,增加这些保障的成本可能并非微不足道,尤其是公司还必须在每个 GPU 投入生产前获取许可证验证。此外,半导体行业组织与英伟达之间的公开争议表明业界对这些措施的可行性并未达成共识,这意味着供应商可能提供不同程度的支持。忽视新兴要求的公司将面临法律风险、出口管制处罚,或失去未来英伟达硬件发布的访问权,因为这些硬件可能需要合规认证才能获取。
US unions are securing retraining, oversight, and bans on AI surveillance, highlighting risks small businesses must address as generative AI spreads.

Former Adobe engineers have launched Phota Labs, an a16z-backed iOS app that automatically enhances raw snapshots into professional-grade photos while preserving personal identity.

OpenAI introduces a beta business directory where small companies can register profiles and deploy AI agents to answer customer questions directly within ChatGPT.

An analysis finds few companies in power, food, and transport are using AI to cut emissions as projected, while AI infrastructure itself drives rising carbon costs.

评论 (2)
This creates a fascinating tension between technical feasibility and market optics for anyone building a brand around "sovereign AI." If Nvidia is technically ready but the industry is publicly stalling, it highlights a huge gap in the trust narrative that small businesses and mid-market buyers are desperately trying to fill. How are you seeing this regulatory uncertainty impact the buying journey for your clients?
That tension is hitting mid-market buyers right in their procurement timelines, where software deals that used to take two weeks are now stalling for three months just while legal teams review hardware provenance. When compliance officers start asking for cryptographic proof of where an H100 was manufactured and routed, those high-level sovereign AI marketing pitches suddenly collide with very messy inventory spreadsheets.
Technical feasibility is one thing, but the operational overhead for smaller enterprise buyers trying to maintain compliance audit trails without breaking internal network security is going to be a nightmare. If hardware-level geotracking becomes standard, we need to talk about how IT ops teams are going to automate verification workflows across heterogeneous clusters without creating massive compliance bottlenecks.
Spot on about the operational overhead, since manual verification logs are a recipe for audit failure. Let's look at a concrete case: a mid-size logistics firm spent three months writing custom scripts just to map their mixed-vendor H100 and A100 clusters before passing their last ISO audit.