
该公司正在加强对在东南亚销售的人工智能加速器的监控,这是限制半导体走私的战略举措,此前该公司因未能采取足够措施阻止芯片非法运往中国企业而受到批评。
Corp. is stepping up monitoring of AI accelerators sold in Southeast Asia, a strategic step to curtail semiconductor smuggling after criticism it hasn’t done enough to stop shipments from illegally reaching Chinese companies.
知情人士透露,这家芯片制造商现在要求该地区的客户加入其所谓的车队情报系统——该系统除其他指标外,还追踪参与公司拥有的人工智能加速器的位置——以便成为英伟达官方云合作伙伴(简称NCP)。
The chipmaker now requires customers in the region to enroll in its so-called fleet intelligence system — which tracks the location of AI accelerators owned by participating companies, among other metrics — in order to become official Nvidia Cloud Partners, or NCPs, according to people familiar with the matter.
对于人工智能云计算提供商来说,NCP称号备受青睐,它既能带来公开的官方认可,又能从这家在行业中举足轻重的公司获得优先获取芯片的资格。知情人士称,英伟达的想法是利用这一头衔的诱惑,来鼓励更广泛地采用目前仍属自愿的芯片追踪计划。由于讨论的是一项机密的新举措,知情人士要求匿名。
The NCP designation is highly coveted for AI cloud computing providers, granting both a public stamp of approval and preferred chip access from a company that plays kingmaker to the industry. Nvidia’s idea is to use the carrot of that label to encourage broader uptake of what remains a voluntary chip tracking program, said the people, who requested anonymity to discuss a confidential new measure.
此举可能会为英伟达硬件在东南亚的流向提供更高的透明度,该地区长期以来一直是受限制商品的转运点。美国政府全面禁止中国企业在未获许可的情况下购买英伟达的高级人工智能芯片,不过当局认为,中国企业依然通过主要经由东南亚的走私网络,获取了价值数十亿美元的此类硬件。
That could provide more visibility into how Nvidia hardware moves through a region that’s long been a transshipment point for restricted goods. The US government broadly prohibits Chinese companies from purchasing advanced Nvidia AI chips without a license, though authorities believe firms there have nonetheless gotten their hands on billions of dollars’ worth of the hardware, primarily via smuggling networks in Southeast Asia.
随着此类案例不断增加,英伟达与部分美国当局就谁该对解决这一问题负责发生了争执。华盛顿的一些角落对这家芯片制造商屡屡忽视危险信号感到沮丧,而英伟达则反唇相讥称,自己一直严格按字面法律行事,这使得甄别违规者的责任落在了政府官员身上。与此同时,这些官员对如何采取适当的应对措施意见不一,特朗普政府内部以及国会山正酝酿着不满情绪。
As cases mount, Nvidia and some US authorities have sparred over who is responsible for solving the problem. Corners of Washington are frustrated that the chipmaker has repeatedly missed red flags, while Nvidia has shot back that it follows laws as written — putting the onus on government officials to identify bad actors. Those officials, meanwhile, disagree on the appropriate tack for their own response, with frustration brewing in the Trump administration and on Capitol Hill.
美速国际私人有限公司(Megaspeed International Pte.)就是这一困境的最好例证。该公司已遭到美国和新加坡调查一年多,上周因《纽约时报》报道称华盛顿怀疑该公司一名高管为北京方面从事间谍活动、并利用她与英伟达(Nvidia)首席执行官黄仁勋(Jensen Huang)的密切关系而再次成为焦点。
There is no better illustration of the quagmire than Megaspeed International Pte., which has been under US and Singapore investigation for more than a year and returned to the limelight last week after The New York Times reported that Washington suspects one of the company’s top executives of spying for Beijing, leveraging her close relationship with Nvidia Chief Executive Officer Jensen Huang.
对此,黄仁勋表示,他与美速国际的互动“完全正常”,而英伟达的一名发言人则表示,这家芯片制造商“确信,如果美国政府认为任何人对美国工业构成威胁,它会让我们知道”。这与英伟达此前对芯片走私嫌疑的立场一致;这家芯片制造商去年表示,它曾询问美国政府是否需要改变与美速国际的关系,得到的回答是不需要。美速国际已否认存在不当行为。
In response, Huang said his interactions with Megaspeed were “completely normal,” while an Nvidia spokesperson said the chipmaker is “confident that if the US government believes that any person is a threat to American industry, it would let us know.”That aligns with Nvidia’s posture on earlier chip diversion suspicions; the chipmaker said last year that it asked the US government whether it needed to alter its relationship with Megaspeed, and was told no. Megaspeed has denied wrongdoing.
英伟达的回应并未让国会山满意,众议院的几位关键共和党议员本周要求该首席执行官进行简报。
Nvidia’s response hasn’t satisfied Capitol Hill, where key House Republicans this week requested a briefing from the CEO.
英伟达自身对美速国际的尽职调查状况——涉及走私嫌疑,而非间谍活动——目前也不明朗。黄仁勋上周表示,“我们已经研究了他们对我们GPU的所有部署,并且清点了每一个,”他指的是图形处理器。这比英伟达此前就此事发表的声明更进一步。英伟达去年在被追问美速国际数据中心足迹存在差异的问题后曾表示,它已查明了发往该公司的“绝大部分”产品。
The status of Nvidia’s own due diligence on Megaspeed — concerning the smuggling suspicions, rather than espionage — is also unclear. Huang said last week that “we’ve studied all their deployments of our GPUs and counted every one,” referring to graphics processing units. That goes a step beyond previous statements on the matter from Nvidia, which said last year — after getting pressed about discrepancies in Megaspeed’s data center footprint — that it had identified “substantially” all the products sent to the company.
英伟达还在2025年12月承诺,将在不久的将来再次访问Megaspeed的数据中心,但此后却多次拒绝置评此类检查是否已经进行。此前不久,Megaspeed的主要计算枢纽房东刚取消了其租户资格,东南亚的另外两家数据中心运营商也对英伟达关于Megaspeed芯片所在地的说法提出了质疑。本周再次被问及英伟达是否进行过任何能够支持其首席执行官关于清点“每一个”处理器这一断言的额外检查时,一位发言人拒绝置评。
Nvidia also pledged in December 2025 to visit Megaspeed’s data centers again in the near future, but since then has repeatedly declined to say whether such checks have occurred. That’s even after Megaspeed’s main computing hub landlord dropped it as a tenant and two other data center operators in Southeast Asia disputed Nvidia’s account of the location of Megaspeed’s chips. Asked again this week whether Nvidia had conducted any additional checks that would support its CEO’s assertions about counting “every” processor, a spokesperson declined to comment.
然而与此同时,英伟达已开始尝试自我监管。其中最大的例子就是车队智能(fleet intelligence)——该功能首次亮相是在近一年前,当时正值美国开始起诉涉嫌走私芯片的人员、华盛顿对Megaspeed的疑虑公之于众之后。这主要是一个数据中心管理工具,可收集有关英伟达硬件部署的一系列信息,对运营商非常有用。不过,由于它还会收集单个部件的地理位置信息,因此它也兼具了应对美国对转运担忧的功能。
In the meantime, however, Nvidia has begun flirting with self-regulation. The biggest such example is fleet intelligence — first debuted nearly a year ago, after the US began prosecuting alleged chip smugglers and Washington’s suspicions about Megaspeed became public. It’s primarily a data center management tool, collecting a range of information about deployments of Nvidia hardware that’s useful for operators. By also gathering geolocation information for individual parts, though, it doubles as a response to US diversion concerns.
该产品既突显出英伟达拥有对其芯片进行地理定位的工具,也表明其在使用这些工具时仍然保持谨慎。一方面,车队智能默认并不要求参与者与英伟达或美国政府共享位置数据。英伟达的一位发言人此前曾向彭博社表示,如果华盛顿监管机构想要数据库中的信息,他们将直接与用户沟通,但没有进一步阐述。
The product underscores both that Nvidia has the tools to geolocate its chips and that it remains cautious about using them. For one, fleet intelligence does not by default require participants to share location data with Nvidia or the US government. An Nvidia spokesperson previously told Bloomberg that should Washington regulators want information from the database, they would communicate directly with users, without elaborating further.
关于最新更新的一个重大问题与NCP(网络云提供商)指定有关,即英伟达最终是否可能要求现有NCP——例如在2023年成立后一周内就获得该标签的Megaspeed——必须注册才能维持其特权地位。英伟达拒绝就这个问题置评,也拒绝回答本文的其他问题,包括新的NCP要求是否会适用于东南亚以外的地区。
A big question for the latest update, related to NCP designations, is whether Nvidia may ultimately require existing NCPs — such as Megaspeed, which won the label within a week of its 2023 creation — to sign up to maintain their privileged status. Nvidia declined to comment on this question or answer others for this story, including whether the new NCP requirement would be applied in regions beyond Southeast Asia.
除了舰队情报之外,英伟达(Nvidia)还采取了其他几项举措,以限制中国公司获取其硬件。据彭博社报道,在美国最大的芯片走私起诉案发生后,该公司今年早些时候将其预先批准可购买芯片的亚洲企业数量削减了一半。
Outside of fleet intelligence, Nvidia has taken several other steps to limit Chinese companies’ access to its hardware. In the wake of the biggest chip smuggling indictment in the US, the company earlier this year removed half of the Asian firms it had pre-approved for chip purchases, Bloomberg has reported.
最近,该公司开始告知买家,不得将超过一半的英伟达计算能力出租给中国公司——尽管华盛顿总体上允许中国公司使用海外第三方拥有的加速器。英伟达的一位发言人对率先报道这一举措的《金融时报》表示:“长久以来的美国政策鼓励每一个云服务都建立在美利坚基础设施之上,并在他们能够触及的每个地方赢得业务。”这位发言人没有解释英伟达这一新举措背后的原因。
More recently, it began telling buyers not to rent more than half of their Nvidia computing power to Chinese companies — even though Washington generally allows Chinese firms to use accelerators owned by others overseas. “Longstanding US policy encourages every cloud to build on American infrastructure and to win business everywhere they can,” an Nvidia spokesperson told The Financial Times, which first reported that move. The spokesperson didn’t explain the reasoning behind the new Nvidia measure.