拉上英伟达硬刚苹果 微软的高端AI PC赌局-36氪Joining Forces with NVIDIA to Take On Apple: Microsoft's High-End AI PC Gamble - 36Kr


AI free token/AI免费积分

拉上英伟达硬刚苹果,微软的高端AI PC赌局

硅碳变量·2026年10月26日 08:55
AI PC的竞争,早已不再停留在有没有NPU的浅层比拼。

在AI PC赛道热闹了两年之后,Windows阵营终于拿出了真正对标Mac生态的旗舰筹码。

微软联合英伟达推出搭载RTX Spark超级芯片的Surface Laptop Ultra,这不再是微软单独下场做硬件的一次常规迭代,而是Windows生态首次拿到英伟达这款一体化顶级SoC,依靠端侧大算力补齐长期存在的短板。

微软Windows与设备执行副总裁帕万·达武卢里周三在旧金山举行的一场公司活动上表示,Surface Laptop Ultra现已开放预购,并将于10月16日开始发货。该电脑配有磁吸USB-C端口,类似苹果的MagSafe充电接口。

Windows借英伟达顶级SoC补齐端侧AI核心短板?

过去两年大量推向市面的AI PC,大多建立在传统PC硬件底座之上,更偏向存量硬件的附加改造。

无论是英特尔、AMD的集成NPU思路,还是高通移植至Windows的Arm版本,核心思路都是在成熟硬件框架之上增加轻量化AI加速单元,很少针对高阶生成式AI任务,完成端侧算力体系的底层重构。

这也使得绝大多数Windows AI PC存在显著局限:端侧专用AI加速单元算力天花板不高,仅合适文本摘要或简单修图这类轻度AI任务。

一旦尝试百亿级大模型本地推理、超长上下文解析、高清图文视频生成、多步骤AI智能体自主执行等高负载情形,算力与内存的约束会快速显现。

这也成为长期以来在高端创作者与开发者市面,Windows生态竞争力不及Mac的关键因素。

苹果依托Apple Silicon自研SoC,达成CPU、GPU与统一内存的深度架构耦合,从硬件底层完成端侧AI全链路优化,为创作者和开发者搭建起稳定高质的本地AI算力底座。

反观Windows阵营,长期受硬件碎片化困扰,跨设备的软硬件协同优化难度更高。在这之前,行业尚未形成一套专门适配高阶AI任务的完整端侧硬件思路,复杂AI任务大多仍要依托云端Copilot完成,本地算力更多只能支撑轻量化AI需要。

拉上英伟达硬刚苹果 微软的高端AI PC赌局-36氪

英伟达RTX Spark超级芯片的落地,有望改写高端AI PC的竞争逻辑,也意味着微软在AI PC赛道不再独自承担平台层的攻坚工作,联合英伟达与一众整机厂商搭建起Windows高端 AI 生态的核心同盟。

作为英伟达面向个人AI智能体构建的旗舰级超级芯片,RTX Spark整合Grace架构CPU与Blackwell架构RTX GPU,最高可配备128GB统一内存,FP4算力最高可达1Petaflop,可以在本地承载120B参数大模型、支撑百万级Token超长上下文窗口,并原生兼容全套 CUDA生态。

在FP4精度AI推理情形下,这套硬件规格让消费级轻薄本具备接近专业AI工作站和小型服务器的本地端侧算力,足以承载过去只能在大型设备上运行的大模型任务。

过去多数PC厂商是在芯片思路定型之后开展整机开发,工作集中在主板、外观与散热调校层面,芯片资源调度和底层平台适配的调整空间有限。

而在RTX Spark项目中,微软全程参与平台定义与打磨,从整机散热思路、供电逻辑、功耗调控,延伸到Windows底层算力调度、内存分配和硬件协同机制,开展全维度联合优化。

这套思路的目标是在该平台上让操作平台可以精准高质调用芯片CPU、GPU与统一内存资源,缓解Windows生态长期存在的软硬件适配割裂难题。

对英伟达而言,RTX Spark的价值不只是一款高端PC芯片,它也是英伟达将CUDA生态向高端消费PC情形延伸的关键布局。

借助引入与云端、专业工作站同源的CUDA软件栈到个人终端,开发者可在本机完成AI大模型原型调试、3D渲染、AI视频生成与智能感受证。依托这套环境,开发者可以在本地完成模型原型搭建、小规模微调与部署验证,大幅降低对云端实例的依赖,缩小消费端与云端AI开发环境之间的迁移成本。

依托英伟达的顶级算力与完整CUDA生态,Windows在这条高端硬件线上,补上了此前难以在轻薄形态下运行高阶端侧AI任务的短板,大幅降低对云端AI的调用需要。

微软瞄准创作者基本盘,用差异化感受完成阵地争夺

跳出硬件参数来看,Surface Laptop Ultra在不少感受外观上吸收了Mac已经被市面验证的高端创作设备思路,但微软并没有简单复刻,而是保留Windows生态独有的差异化长处。

最有代表性的是Magnetic Connect磁吸USB-C接口,这是笔记本原厂首次把磁吸结构集成在标准USB-C端口上。外观思路瞄准Mac磁吸充电防拉扯的使用痛点,同时保留USB-C通用拓展能力,替代了微软沿用多年和生态封闭的Surface Connect私有接口,兼顾接口安全与通用性。

除此之外,PixelSense Ultra触控屏峰值HDR亮度可达2000尼特,高于M5 Pro版本MacBook Pro 的1600尼特峰值,同时大尺寸触控板也加入了触觉反馈。

不过微软没有全盘照搬Mac的取舍逻辑,保留了Windows受众看重的实用属性:机身配齐HDMI、USB-A、SD读卡器、3.5毫米耳机口,拓展能力完整,在满足配套线材与规格前提下,最多可同时驱动三台4K表明器;SSD兼容使用者自行拆装更换,为创作者和AI开发者预留硬件扩容空间。

这套外观取舍,瞄准的核心目标人群颇为明确:创作者与AI开发者。

这部分群体恰好是苹果在高端生产力市面里最稳固的基本盘。长期以来,大量视频剪辑、3D建模、AI模型调试从业者挑选Mac,不只是芯片算力的长处,更是统一内存架构、整机稳定性、软硬件深度协同共同构成的综合使用感受。

微软的竞争思路也由此清晰:依托RTX Spark带来的完整CUDA生态,以及特定情形下更突出的AI视频生成表现,再搭配对标Mac水准的精致硬件感受,尝试撬动苹果长期把持的高端创作使用者群体。

这一系列操作释放的信号颇为明确:微软不再满足于守住大众PC市面的基本盘,试图向高端创作和本地AI开发这一苹果长处领地发起正面角逐。

2599美元的起售价低于同档位16英寸MacBook Pro,这套定价策略本质是生态卡位,而非单纯降价让利。

微软想要借助这款标杆机型证明:Windows高端AI PC,可以在本地AI任务上给出持平甚至更强的算力表现,同时保留Windows生态的软件兼容性、CUDA开发能力,以及外设拓展和硬件可升级的自由度。

但放到更大的产业视角看,仅凭这一台旗舰硬件很难直接推动Windows在高端创作者市面重新占据主流。

拉上英伟达硬刚苹果 微软的高端AI PC赌局-36氪

高端生产力市面的使用者迁移,从来不是单一硬件参数或者一款标杆机型就能促成。创作者群体的迁移成本很高:工程文件、项目模板、插件生态、团队协作规范,都深度绑定现有工作流。

很多团队挑选Mac不只是看重本地AI算力,还有跨设备协同和长期版本稳定性,以及大量专业创作软件针对Apple Silicon的原生持续优化。这是苹果多年积累形成的软件生态壁垒,短期很难靠一台Windows旗舰抹平。

另一个现实约束来自机型定位,由于RTX Spark属于高端SoC,成本高且产能有限,短期只会落地在少数旗舰机型,很难快速下沉到主流价位段PC。绝大多数普通Windows AI PC依旧沿用英特尔、AMD、高通 NPU 思路,硬件碎片化和软硬件协同参差不齐的老难题不会消失。

也就是说,这套出色的端侧 AI 底座,短期内无法普惠整个 Windows 阵营。Windows 仍旧会展现两极分化:少数旗舰机型具备媲美 Mac 的一体化 AI 能力,海量主流机型仍然停留在轻量化 NPU 的阶段。

更大的考验落在Windows Agent的落地质量上,这套旗舰硬件所承载的AI价值,最终要依靠平台层智能体来兑现。倘若Agent在多步骤任务调度、本地混合模型推理、应用深度联动上的感受不及预期,再亮眼的芯片规格也容易沦为纸面参数。

而智能体感受的打磨,需要海量软件开发者参与适配、大量第三方应用深度改造,这类生态培育的周期,会远长于硬件研发周期。

从全局看,这台设备的战略定位更像Windows高端AI生态的能力样板。它的目标并非立刻转化大量Mac存量使用者,而是向开发者和创作者证明:Windows平台有潜力搭建起承载高阶本地 AI、大模型调试、AI 视频生成的一体化硬件底座。

硬件能不能撑起Windows Agents?

整台Surface Laptop Ultra的硬件外观,核心取舍都围绕一个目标展开:支撑Windows Agent,也就是可以在本地运行的个人AI智能体。

其和当下人们熟悉的Copilot存在显著层级差异。

Agent不再局限于简单问答交互,目标是跨应用自动串联多步骤任务:读取本地文档、调用绘图工具、剪辑视频、管理本地文件,依托多模型协同完成完整复杂工作流,并且尽可能在终端本地完成运算,减少云端依赖,以此保障数据隐私。

这类自动化智能任务对硬件提出了严苛门槛:大容量统一内存、持续稳定的AI推理算力、CPU与GPU之间的高速互联通道,这正是RTX Spark芯片的外观原点。

英伟达还配套推出OpenShell,为智能体搭建本地安全运行沙箱,借助权限策略管控,让Agent在约束之下安全调用本机各类硬件与软件资源。

但硬件就位并不代表Windows Agent就能顺利落地,中间横亘着两层现实鸿沟。

第一层鸿沟,来自软件生态的适配缺口。

算力底座只是基础,Agent想要跑通全链路离不开 Windows底层API、第三方应用、模型调度框架的多方协同。当下Adobe已宣布针对RTX Spark重构Photoshop、Premiere底层架构,优化AI加速能力,但海量中小开发者与垂直行业软件,仍需要漫长的迁移与改造周期。

即便硬件足以承载120B参数大模型,如果应用层缺少配套的Agent联动能力,使用者拿到手,它大概率依旧只是一台高表现创作本,而非具备自主执行能力的AI智能体设备。

第二层鸿沟,则落在功耗控制、平台稳定性与真实使用感受层面。

持续运行由大模型驱动的智能体会让整机长期处于高负载状态,对散热外观、平台调度算法以及电池管理形成持续考验。发布会的演示情形属于理想环境,在长时间叠加多任务的真实情形下,是否会浮现表现降频和电池续航大幅缩水,仍有待真实使用者长期使用来验证。

硬件只是赋予可能性。Windows Agent能否从发布会Demo,蜕变为日常稳定可用的生产力工具,很大程度上取决于后续Windows平台迭代以及整个软件生态的建设进度。

随之而来的争议是:Surface Laptop Ultra会不会只是一台曲高和寡的标杆演示机型,很难达成大规模放量?

定价已经说明了它的定位:2599美元起,128GB顶配版本达到5899美元,对标专业工作站,核心面向AI开发者和高端创作者,并非瞄准大众消费市面的走量机型。微软也并未指望依靠这款设备吸纳海量使用者,它更核心的使命是完成整套平台定义。

微软与英伟达已经官宣,华硕、戴尔、惠普、联想、微星等PC厂商将在今年秋季陆续推出搭载RTX Spark的轻薄本与小型主机,宏碁、技嘉也会后续跟进。

不过大众AI PC的普及,依旧要面对多重约束。

首先是成本难题,RTX Spark作为旗舰 SoC,芯片制造良率和整机BOM成本都偏高,短期内很难下探到万元以内的主流消费区间。

其次是功耗的取舍难题,面向普通使用者的轻薄机型必须在AI 算力、电池续航、噪音之间寻找平衡点,很难复刻旗舰机型的满血算力规格。

最后是应用生态的普及逻辑:如果普通使用者无法直观感受到端侧AI带来的实际价值,便很难为AI能力对应的机型溢价买单。

微软联合英伟达推出RTX Spark旗舰PC本质并非一场单纯的硬件参数竞赛,而是 Windows在个人AI时代重构高端生产力生态的一次关键押注。

短期来看,这款旗舰机型更像是开发者与创作者的试验田。中长期成败系于两件事:一是微软能否把Windows Agent从演示特性打磨成稳定可靠的平台原生能力;二是PC厂商和软件开发者能否围绕 RTX Spark 持续多元配套应用情形。

AI PC的竞争,早已不再停留在有没有NPU的浅层比拼。

本文来自微信公众号“硅碳变量”,作者:温宁,36氪经授权发布。

+1
1

好文章,需要你的鼓励

你可能也喜欢这些文章
谷歌开源TorchTPU,TPU成PyTorch原生设备,CUDA最深的护城河开始松动
谷歌为AI电力想出妙招,构建虚拟电网
人类的科技树,终于点亮了炼金术
Muse刚火,小扎就催它挣钱了
硅谷医疗科技的荒诞现实:近半创始人没有医疗经验
Meta拒绝投放广告,TikTok将广告预算转向YouTube
英伟达杀入,CPU四强争霸
马斯克画饼「全民高收入」,AI干了我的活,还能给我分钱?
Anthropic 招股书最新披露:营收163亿、账面亏损高达807亿美元

专注硬科技全产业链观察 拆解工艺革新、资本动向与产业变局

最近内容

1元/Wh这道坎,把储能行业劈成了两半
拉上英伟达硬刚苹果,微软的高端AI PC赌局
AI存储困局:良率天花板为何卡住了HBM供给?
选靠谱AI,看真实实测
查看
36氪AI测评官方交流社区
加入
咨询项目审核和入驻
联系
36氪项目建议订阅号
关注
马斯克:在AI面前,人类就是黑猩猩

AI 不需要消灭人类,AI 很可能懒得理你

12小时前

意见反馈
36氪APP让一部分人先看到将来
36氪
鲸准
氪空间
拉上英伟达硬刚苹果 微软的高端AI PC赌局-36氪

推送和解读前沿、有料的科技创投资讯

本文来源:https://www.36kr.com/p/4040520626689666
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

Microsoft’s high-end AI PC bet: teaming up with NVIDIA to take a hard line against Apple

Silicon-Carbon Variables · October 26, 2026 08:55

Competition in AI PCs has long moved beyond the superficial contest of whether a device has an NPU.

After two years of heated activity in the AI PC track, the Windows camp finally has a flagship offering that truly targets the Mac ecosystem.

Microsoft and NVIDIA have jointly launched the Surface Laptop Ultra powered by the RTX Spark super chip. This is not a routine iteration of Microsoft building hardware on its own; it marks the first time the Windows ecosystem has obtained NVIDIA’s all-in-one top-tier SoC, using large on-device compute to shore up a long-standing weakness.

Pavan Davuluri, Microsoft’s Executive Vice President of Windows and Devices, said at a company event in San Francisco on Wednesday that the Surface Laptop Ultra is now available for preorder and will begin shipping on October 16. The laptop features a magnetic USB-C port, similar to Apple’s MagSafe charging connector.

Is Windows closing the gap in on-device AI core capabilities by leveraging NVIDIA’s top-tier SoC?

Most AI PCs pushed to market over the past two years were built on traditional PC hardware foundations, leaning more toward incremental retrofits of existing hardware.

Whether it’s Intel and AMD’s integrated-NPU approach or Qualcomm’s Arm build ported to Windows, the core idea has been to add lightweight AI acceleration units on top of mature hardware frameworks, rarely completing a bottom-up reconstruction of the on-device compute system for advanced generative AI tasks.

That has also led to significant limitations for the vast majority of Windows AI PCs: the compute ceiling of the on-device dedicated AI acceleration unit is not high, making it suitable mainly for light AI tasks such as text summarization or simple image editing.

Once you try high-load scenarios—local inference for tens-of-billions-parameter large models, ultra-long context parsing, high-resolution image/text/video generation, or multi-step autonomous execution by AI agents—compute and memory constraints quickly become apparent.

This has long been a key factor behind why, in the high-end creator and developer market, the Windows ecosystem has been less competitive than the Mac.

By relying on Apple Silicon’s in-house SoCs, Apple achieves deep architectural coupling among the CPU, GPU, and unified memory, completing full-stack on-device AI optimization at the hardware foundation and building a stable, high-quality local AI compute base for creators and developers.

By contrast, the Windows camp has long been plagued by hardware fragmentation, making cross-device software-hardware coordination and optimization more difficult. Until now, the industry had not formed a complete on-device hardware approach specifically tailored to advanced AI tasks, so complex AI work has largely relied on cloud-based Copilot, while local compute could mainly support lightweight AI needs.

Joining Forces with NVIDIA to Take On Apple: Microsoft's High-End AI PC Gamble - 36Kr

The arrival of NVIDIA’s RTX Spark super chip is expected to reshape the competitive logic of high-end AI PCs. It also means that in the AI PC track Microsoft is no longer shouldering platform-layer breakthroughs alone, but has formed a core alliance with NVIDIA and a range of OEM partners to build a high-end Windows AI ecosystem.

As NVIDIA’s flagship super chip built for personal AI agents, RTX Spark integrates a Grace-architecture CPU with an RTX GPU based on the Blackwell architecture. It can be configured with up to 128GB of unified memory and deliver up to 1 Petaflop of FP4 compute, enabling it to run 120B-parameter large models locally, support million-token ultra-long context windows, and natively compat with the full CUDA ecosystem.

Under FP4-precision AI inference, these hardware specs allow consumer thin-and-light laptops to have on-device compute close to that of professional AI workstations and small servers—sufficient to handle large-model tasks that previously could only run on large-scale equipment.

In the past, most PC vendors began whole-system development after the chip direction was finalized, focusing on motherboard design, exterior design, and thermal tuning, with limited room to adjust chip resource scheduling and low-level platform adaptation.

In the RTX Spark project, however, Microsoft participated throughout in platform definition and refinement, carrying out full-dimensional joint optimization spanning system thermal design, power delivery logic, and power management, as well as Windows’ low-level compute scheduling, memory allocation, and hardware coordination mechanisms.

The goal of this approach is to enable the operating platform to call CPU, GPU, and unified-memory resources on this platform with precision and high quality, mitigating the long-standing split between software and hardware adaptation in the Windows ecosystem.

For NVIDIA, the value of RTX Spark is not only as a high-end PC chip; it is also a key move to extend the CUDA ecosystem into high-end consumer PC scenarios.

By bringing the same-origin CUDA software stack used in the cloud and professional workstations into personal terminals, developers can complete AI large-model prototyping and debugging, 3D rendering, AI video generation, and experience validation on their local machines. With this environment, developers can build model prototypes, perform small-scale fine-tuning, and carry out deployment validation locally, greatly reducing reliance on cloud instances and lowering migration costs between consumer-side and cloud AI development environments.

With NVIDIA’s top-tier compute and the complete CUDA ecosystem, Windows, on this high-end hardware line, has addressed the previously difficult challenge of running advanced on-device AI tasks in a thin-and-light form factor, significantly reducing the need for cloud AI calls.

Microsoft targets the creator core base, using differentiated experiences to fight for territory

Looking beyond hardware specs, the Surface Laptop Ultra incorporates many user-experience and design ideas that Mac has already validated in the market for high-end creative devices. But rather than simply copying, Microsoft retains the distinctive strengths unique to the Windows ecosystem.

The most representative example is the Magnetic Connect magnetic USB-C port—the first time an OEM laptop has integrated a magnetic mechanism into a standard USB-C port. The design targets the real-world pain point of Mac’s magnetic charging avoiding cable tug, while preserving the extensibility of USB-C, replacing Microsoft’s Surface Connect proprietary interface (used for many years and closed to the ecosystem), balancing port security with universality.

In addition, the PixelSense Ultra touchscreen’s peak HDR brightness can reach 2000 nits, higher than the 1600-nit peak of the M5 Pro version MacBook Pro, and the large touchpad also adds haptic feedback.

However, Microsoft has not wholesale adopted Mac’s trade-off logic, and has kept practical features valued by Windows users: the chassis includes HDMI, USB-A, an SD card reader, and a 3.5mm headphone jack, offering complete connectivity. Provided the required cables and specs are met, it can drive up to three 4K monitors simultaneously; the SSD is compatible with user removal and replacement, reserving hardware expansion space for creators and AI developers.

These design and feature trade-offs point to a clearly defined core target audience: creators and AI developers.

This group happens to be Apple’s most solid foundation in the high-end productivity market. For a long time, large numbers of video editors, 3D modelers, and AI model fine-tuning practitioners have chosen Mac—not only because of the advantages of chip compute power, but also because of the comprehensive user experience formed by the unified memory architecture, overall system stability, and deep hardware-software integration.

Microsoft’s competitive logic also becomes clear: leveraging the complete CUDA ecosystem brought by RTX Spark, along with more prominent AI video generation performance under certain scenarios, and pairing that with refined hardware experiences on par with the Mac level, it is attempting to pry loose the high-end creative user base that Apple has long dominated.

The signal released by this series of moves is quite clear: Microsoft is no longer satisfied with defending its core position in the mass-market PC segment, and is trying to launch a direct contest against Apple in the high-end creation and local AI development arena where Apple excels.

The $2,599 starting price is lower than that of the 16-inch MacBook Pro in the same tier; this pricing strategy is essentially an ecosystem positioning play, rather than a simple price cut to concede margin.

Microsoft wants to use this flagship model to prove that high-end Windows AI PCs can deliver compute performance that matches or even exceeds Apple on local AI tasks, while retaining the Windows ecosystem’s software compatibility, CUDA development capabilities, and the freedom of peripheral expansion and hardware upgradeability.

But from a broader industry perspective, a single flagship hardware product alone is unlikely to directly drive Windows to regain mainstream status in the high-end creator market.

Joining Forces with NVIDIA to Take On Apple: Microsoft's High-End AI PC Gamble - 36Kr

Migration of users in the high-end productivity market has never been achieved by a single hardware parameter or one flagship model. The switching costs for creators are high: project files, project templates, plugin ecosystems, and team collaboration standards are all deeply tied to existing workflows.

Many teams choose Mac not only because of local AI compute power, but also because of cross-device coordination and long-term version stability, as well as sustained native optimization of many professional creative applications for Apple Silicon. This is a software ecosystem moat that Apple has built over many years, and it is unlikely to be leveled in the short term by a single Windows flagship.

Another practical constraint comes from the product positioning: because RTX Spark is a high-end SoC with high costs and limited production capacity, it will only land on a small number of flagship models in the short term, making it difficult to quickly move down to mainstream-priced PCs. Most ordinary Windows AI PCs still follow the Intel, AMD, and Qualcomm NPU approach, and the old problems of hardware fragmentation and uneven hardware-software coordination will not disappear.

In other words, this excellent on-device AI foundation cannot benefit the entire Windows camp in the short term. Windows will still show a polarized pattern: a small number of flagship models will have integrated AI capabilities comparable to the Mac, while a vast number of mainstream models will remain at the lightweight NPU stage.

A bigger test falls on the implementation quality of Windows Agents; the AI value carried by this flagship hardware must ultimately be realized through platform-layer agents. If the agent experience in multi-step task orchestration, local hybrid model inference, and deep application integration falls short of expectations, even the most eye-catching chip specs can easily become nothing more than paper specifications.

And polishing the agent experience requires massive participation from software developers in adaptation and deep rework by a large number of third-party applications; the cycle for such ecosystem cultivation will be far longer than the hardware development cycle.

From a holistic perspective, this device’s strategic positioning looks more like a capability showcase for Windows’ high-end AI ecosystem. Its goal is not to immediately convert a large number of existing Mac users, but to demonstrate to developers and creators that the Windows platform has the potential to build an integrated hardware foundation that can support advanced local AI, large model fine-tuning, and AI video generation.

Can the hardware support Windows Agents?

The hardware and design choices of the entire Surface Laptop Ultra revolve around one core objective: to support Windows Agent—that is, a personal AI agent that can run locally.

It differs significantly in capability level from the Copilot that people are familiar with today.

Agent is no longer limited to simple question-and-answer interactions; its goal is to automatically chain multi-step tasks across applications: reading local documents, invoking drawing tools, editing videos, and managing local files. Leveraging multi-model collaboration, it completes entire complex workflows, and as much as possible runs computation locally on the device to reduce cloud dependence, thereby protecting data privacy.

This kind of automated intelligent task imposes strict hardware requirements: large-capacity unified memory, sustained and stable AI inference compute power, and high-speed interconnect channels between the CPU and GPU—this is exactly the design starting point of the RTX Spark chip.

Nvidia has also launched OpenShell as a companion, creating a local secure sandbox for agents to run in. With policy-based access control, it enables agents to securely invoke various hardware and software resources on the machine within constraints.

But hardware being in place does not mean Windows Agents can be smoothly implemented; there are two real gaps in between.

The first gap comes from shortfalls in software ecosystem adaptation.

A compute foundation is only the baseline; for Agent to run through the full chain, it relies on multi-party collaboration among Windows low-level APIs, third-party applications, and model orchestration frameworks. At present, Adobe has announced that it will rebuild the underlying architectures of Photoshop and Premiere for RTX Spark and optimize AI acceleration, but a vast number of small and medium developers and vertical-industry software still need a long migration and rework cycle.

Even if the hardware is capable of supporting a 120B-parameter large model, if the application layer lacks matching agent integration capabilities, what users get will most likely still just be a high-performance creation laptop, rather than an AI agent device with autonomous execution capabilities.

The second gap lies in power control, platform stability, and real-world user experience.

Continuously running agents driven by large models will keep the whole system under high load for long periods, posing ongoing challenges for thermal design, platform scheduling algorithms, and battery management. The demo scenarios at the launch event represent an ideal environment; under real-world conditions of long hours with multiple tasks stacked, whether performance throttling and significant battery-life reduction will emerge still needs to be verified through long-term use by actual users.

Hardware only creates possibilities. Whether Windows Agent can evolve from a launch-event demo into a stable, everyday productivity tool depends largely on subsequent iterations of the Windows platform and the progress of building the entire software ecosystem.

The controversy that follows is: will Surface Laptop Ultra end up being just a high-end, niche showcase model that is difficult to achieve large-scale shipments?

The pricing already reveals its positioning: starting at $2,599, with the top-tier 128GB configuration reaching $5,899, it targets professional workstations and is aimed primarily at AI developers and high-end creators, rather than a volume product for the mass consumer market. Microsoft is also not counting on this device to attract massive numbers of users; its more central mission is to complete a full platform definition.

Microsoft and Nvidia have officially announced that PC makers including ASUS, Dell, HP, Lenovo, and MSI will roll out thin-and-light laptops and small desktops equipped with RTX Spark this fall, with Acer and Gigabyte also expected to follow later.

However, the widespread adoption of consumer AI PCs still faces multiple constraints.

First is the cost challenge: as a flagship SoC, RTX Spark faces high chip manufacturing yields and elevated overall BOM costs, making it difficult to drop into the mainstream consumer price bracket of under 10,000 yuan in the short term.

Second is the power consumption trade-off: lightweight models aimed at general users must find a balance among AI computing power, battery life, and noise, making it difficult to replicate the full-performance computing specifications of flagship models.

Finally, there is the adoption logic of the application ecosystem: if general users cannot intuitively experience the tangible value brought by on-device AI, they will hardly pay for the price premium associated with these AI capabilities.

Microsoft and Nvidia's launch of the RTX Spark flagship PC is not merely a hardware specification race; it is a critical bet by Windows to restructure the high-end productivity ecosystem in the era of personal AI.

In the short term, this flagship model is more like a testing ground for developers and creators. Its medium- and long-term success hinges on two things: first, whether Microsoft can refine Windows Agent from a demo feature into a stable and reliable platform-native capability; and second, whether PC manufacturers and software developers can continuously build diverse supporting application scenarios around RTX Spark.

The competition in AI PCs has long moved beyond the shallow race of simply having an NPU.

This article is from the WeChat official account "Guitan Bianliang" (Silicon-Carbon Variables), author: Wen Ning. 36Kr authorized publication.

1

Good article, need your encouragement

You may also like these articles

Google open-sources TorchTPU, making TPU a native PyTorch device; CUDA's deepest moat begins to loosen

Google comes up with a brilliant trick for AI electricity, building a virtual power grid

Humanity's tech tree finally lights up alchemy

Muse just got popular, and Zuckerberg is already urging it to make money

The absurd reality of Silicon Valley's health tech: nearly half of the founders have no medical experience

Meta refuses to run ads, TikTok shifts its ad budget to YouTube

Nvidia enters the fray, sparking a four-way battle for CPU supremacy

Musk promises "universal high income": AI does my work, and still gives me a share of the money?

Anthropic's latest prospectus disclosure: revenue hits 16.3 billion, with a book loss of up to 80.7 billion dollars

Dedicated to observing the entire hard tech industry chain: unpacking process innovations, capital trends, and industry shifts

Recent content

The 1 yuan/Wh hurdle splits the energy storage industry in two

Teaming up with Nvidia to take on Apple: Microsoft's high-end AI PC gamble

The AI storage dilemma: why is the yield ceiling stifling HBM supply?

Choose reliable AI, look at real-world tests

View

36Kr AI Review Official Community

Join

Project review and onboarding inquiry

Contact

36Kr Project Recommendation Subscription

Follow

Musk: In the Face of AI, Humans Are Chimpanzees

AI Doesn't Need to Eliminate Humans; AI Probably Just Won't Bother With You

12 hours ago

About 36Kr

City Partnership

Project Suggestions

I Want to Join

Investor Relations

Business Cooperation

About the Author

Contact the Author

Join the Author

36Kr Europe 36Kr Europe 36Kr Europe AI Model Daily Report

Online Rumor Reporting Portal

Partners

高德

36Kr APP Download

iOS & Android

Feedback

36Kr APP Let Some People See the Future First

36Kr

Jingzhun

KR Space

Delivering and Interpreting Cutting-Edge, Insightful Tech Venture Capital News

Joining Forces with NVIDIA to Take On Apple: Microsoft's High-End AI PC Gamble - 36Kr

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry