Apple 在规划面向开发者、企业和政府的 AI 服务器，每台 2 或 4 颗 M8 Ultra；最早 2029 年，仍可能取消。
Apple 在放弃 Xserve 业务 15 年多之后，正在考虑重返专用服务器市场：公司在探索出售以自研处理器为核心的高性能人工智能系统，而这些系统可能要用老对手 Nvidia 的网络技术连接起来。
Apple 盯上企业级 AI 硬件
据 The Information（付费墙）的最新报道，Apple 一直在规划一款企业级 AI 服务器，面向 AI 开发者、企业和政府客户。拟议中的系统可能搭载 2 颗或 4 颗 Apple 规划中的 M8 Ultra 处理器，让公司的自研芯片战略远远超出目前主要使用它们的 Mac、iPhone 和内部云基础设施。
这个项目将是 Apple 硬件野心的一次惊人扩张。Apple 上一次卖专用服务器产品，还是 Xserve 系列；随着公司越来越专注于消费设备和面向企业的常规 Mac 硬件，Xserve 在 2011 年停产。
据 The Information 报道，新服务器预计不会早于 2029 年推出，项目仍在变化，仍可能被取消，或者在不用 Nvidia 技术的情况下上市。
02
意外的一环：找 Nvidia 谈互连
Apple 考虑用 NVLink Fusion 连接多颗 M8 Ultra；两家关系僵了近二十年，AI 基建的规模可能在逼它重估技术边界。
意想不到的 Nvidia 关联
最值得注意的一点，是 Apple 与 Nvidia 讨论使用 NVLink Fusion。这是 Nvidia 用来连接处理器、让它们以极高速度交换数据的技术。
据报道，Apple 曾考虑用 NVLink Fusion 连接拟议服务器里的多颗 M8 Ultra 处理器。这类高带宽互连在 AI 计算中越来越重要，因为大型推理负载可以分布到多颗处理器上，而这些处理器必须作为一个协调的系统运转。
考虑到两家公司的不愉快历史，合作将意义重大。在 Mac 时代，Apple 和 Nvidia 关系密切，但在部分 MacBook 所用的 Nvidia 图形处理器出问题后，关系恶化。此后 Apple 逐渐不用 Nvidia 的硬件，并靠 Apple 芯片最终替换了大部分 Mac 产品线里的第三方处理器。Reuters 指出，两家公司的关系已经紧张了将近二十年。
这些讨论表明，现代 AI 基础设施的规模，可能正在迫使 Apple 重新考虑一些长期坚持的技术边界。
03
从自用服务器到对外卖
Apple 已用自研芯片跑部分云端 AI，更重的负载要借外部算力；这次可能直接把算力系统卖给外部机构，进入 Nvidia 和云厂商的地盘。
从内部 AI 服务器到商业产品
Apple 已经在把自己的芯片更深地推进数据中心。公司用自己的处理器处理部分云端 AI 基础设施的工作，而要求更高的负载则需要借助外部算力。
The Information 此前报道，Apple 也一直在开发专门的服务器芯片，寻找更高效地运行越来越吃力的 AI 模型的办法。拟议中的 M8 Ultra 服务器似乎把这一战略带向了新方向：公司可能不再只为自家服务设计基础设施，而是直接把计算系统卖给外部机构。
这会让 Apple 进入一个越来越由 Nvidia，以及包括 Amazon、Microsoft 和 Google 在内的服务器厂商和云服务商塑造的市场。
04
意义：Apple 芯片变企业平台
意义不在复活 Xserve，而在把 Apple 芯片变成企业计算平台，借 Nvidia 的互连专心做处理器。但离兑现还远，两家拒评，Reuters 没能核实。
为什么重要
更大的意义也许不在于复活 Xserve，而在于把 Apple 芯片变成一个企业计算平台。
Apple 的芯片战略已经从 iPhone 扩展到 iPad、Mac 和数据中心。一台 M8 Ultra 服务器会把这套架构延伸到商用 AI 基础设施，为 Apple 设计的处理器再开辟一个潜在市场，同时让开发者和企业用上 Apple 在各类设备上使用的同一套芯片生态。
使用 Nvidia 的网络技术，也可能让 Apple 专注于它越来越视为核心优势的事，也就是设计处理器，同时依靠成熟的技术在数据中心规模上把这些芯片连起来。
这个项目还要好几年，而且远未确定。Apple 和 Nvidia 都拒绝向 Reuters 置评；Reuters 表示无法独立核实 The Information 的报道。
但如果 Apple 推进下去，它重返服务器市场的方式将与 Xserve 时代大不相同。这一次的机会不是传统的企业计算，而是 AI。Apple 似乎在考虑，驱动其消费设备生态的处理器，能否成为一项全新企业业务的基础。
Even Apple can't build the interconnect layer itself: the most vertically integrated company in the industry is talking to Nvidia about NVLink to link its own chips.
How to read this
A second-hand quick take, still to be checked: Patently Apple relays a paywalled report from The Information; Apple and Nvidia declined to comment, and Reuters says it couldn't verify it independently. It's also very early: no sooner than 2029, the project is fluid, it could be canceled, and it might ship without Nvidia. Its value is as a signal, not as evidence.
Per The Information, Apple is planning an AI server for developers, companies and governments, with two or four M8 Ultras each; not before 2029, still fluid, and it could be canceled.
Apple is considering a return to the dedicated server market more than 15 years after abandoning its Xserve business, as the company explores selling high-powered artificial intelligence systems built around its own processors — potentially linked together using networking technology from longtime rival Nvidia.
Apple Eyes Enterprise AI Hardware
According to a new report from The Information (paywall), Apple has been developing plans for an enterprise AI server aimed at AI developers, corporations and government customers. The proposed systems could contain either two or four of Apple’s planned M8 Ultra processors, extending the company’s custom silicon strategy well beyond the Macs, iPhones and internal cloud infrastructure where its chips are primarily used today.
The project would represent a striking expansion of Apple’s hardware ambitions. Apple last sold a dedicated server product through its Xserve line, which was discontinued in 2011 as the company concentrated increasingly on consumer devices and conventional Mac hardware for businesses.
The new server is not expected to arrive before 2029, and the project remains fluid. It could still be canceled or reach the market without Nvidia technology, according to The Information.
02
The surprise: talks with Nvidia on interconnect
Apple has considered Nvidia's NVLink Fusion to link multiple M8 Ultras. After nearly two decades of strained relations, the scale of AI infrastructure may be forcing Apple to rethink its technology boundaries.
An Unexpected Nvidia Connection
One of the most notable elements is Apple’s discussion with Nvidia about using NVLink Fusion, Nvidia’s technology for connecting processors so they can exchange data at very high speeds.
Apple has reportedly considered NVLink Fusion as a way of connecting multiple M8 Ultra processors inside the proposed server. Such high-bandwidth interconnects have become increasingly important in AI computing because large inference workloads can be distributed across several processors that must operate as a coordinated system.
A partnership would be significant given the companies’ difficult history. Apple and Nvidia had a substantial relationship during the Mac era, but ties deteriorated following problems involving Nvidia graphics processors used in some MacBooks. Apple subsequently moved away from Nvidia hardware and, with Apple silicon, eventually replaced third-party processors across most of its Mac lineup. Reuters noted that relations between the companies have remained strained for nearly two decades.
The discussions suggest that the scale of modern AI infrastructure may be forcing Apple to reconsider some longstanding technology boundaries.
03
From internal servers to a product
Apple already runs parts of its cloud AI on its own chips and needs outside compute for heavier loads; now it might sell computing systems directly to outside organizations, on Nvidia's and the cloud providers' turf.
From Internal AI Servers to a Commercial Product
Apple has already been pushing its silicon deeper into the data center. The company uses its own processors to handle parts of its cloud-based AI infrastructure, while more demanding workloads have required outside computing resources.
The Information previously reported that Apple has also been developing specialized server silicon as it looks for more efficient ways to run increasingly demanding AI models. The proposed M8 Ultra server appears to take that strategy in a new direction: rather than designing infrastructure solely for Apple’s own services, the company would potentially sell computing systems directly to outside organizations.
That would put Apple into a market increasingly shaped by Nvidia, along with server manufacturers and cloud providers including Amazon, Microsoft and Google.
04
Why it matters: Apple silicon as an enterprise platform
The point isn't reviving Xserve but turning Apple silicon into an enterprise computing platform, with Nvidia's interconnect letting Apple focus on processors. It's far off; both companies declined to comment, and Reuters couldn't verify it.
Why It Matters
The larger significance may be less about reviving Xserve than about transforming Apple silicon into an enterprise computing platform.
Apple’s chip strategy has already expanded from the iPhone to the iPad, Mac and data center. An M8 Ultra server would extend that architecture into commercial AI infrastructure, creating another potential market for Apple-designed processors while giving developers and businesses access to the same silicon ecosystem Apple uses across its devices.
Using Nvidia networking technology could also allow Apple to concentrate on what it increasingly considers a core strength — designing processors — while relying on proven technology to connect those chips at data-center scale.
The project remains years away and far from guaranteed. Apple and Nvidia declined to comment to Reuters, which said it could not independently verify The Information’s report.
But if Apple proceeds, its return to servers would be substantially different from the Xserve era. This time, the opportunity isn’t traditional corporate computing. It is AI — and Apple appears to be considering whether the processors powering its consumer-device ecosystem could become the foundation of an entirely new enterprise business.
Disclosure: The owner of Patently Apple holds Nvidia Stock
Where Indigo landsFurther
Indigo's conclusion
The signal is in the interconnect: at data-center scale, even Apple, the company least willing to depend on Nvidia, is going back to talk about NVLink. But it's triple second-hand news about 2029 and beyond; a signal, not a conclusion.
What to remember
Even Apple can't build the interconnect itself: at data-center scale, the most vertically integrated company is talking about NVLink Fusion.
NVLink Fusion is Nvidia's horizontal hook, pulling other companies' chips into its interconnect ecosystem.
For Apple it's a distant possibility: after 2029, possibly canceled, possibly without Nvidia; no near-term effect.
Triple second-hand; check the originals first: The Information's report and any Reuters follow-up.
Back on the long-running theses
confirms
Bottlenecks are shifting from algorithms to physics “Even Apple can't build the interconnect layer” is another concrete landing point for this view: the bottleneck sinks from algorithms to interconnect and bandwidth, and even the ultimate vertical integrator can't route around it. Second-hand, to be checked.
adds to
Dally and Jouppi on chip architecture The architects say energy and memory bandwidth are hard limits; this report makes the interconnect bottleneck concrete, with even Apple renting NVLink.
For Nvidia: if true, a first-hand signal that even Apple has to rent its interconnect moat, reinforcing the central-bank view. For Apple: its chips becoming an enterprise compute platform is a distant possibility, after 2029 and possibly canceled, with no near-term effect.
What would change my mind
Apple shipping an AI server before 2029 on its own high-bandwidth interconnect, or officially confirming it walked away from Nvidia entirely.
Finished. Indigo's take on this piece is in two places: