When Microsoft unveiled its agentic Windows strategy at Build 2026 and the October Surface event, the demos centered on Surface Laptop Ultra and the seamless integration between Windows, Microsoft Foundry, and NVIDIA RTX Spark. The story was compelling — a vertically integrated hardware-software stack where Microsoft controls the silicon partnership, the operating system, the AI runtime, and the developer toolchain.
But Microsoft does not build most Windows PCs. Dell, HP, Lenovo, ASUS, Acer, MSI, and Gigabyte collectively ship the overwhelming majority of Windows devices worldwide. These companies must navigate a platform transition they did not architect, on hardware timelines they do not control, while competing against the very company defining the platform standard.
This article examines how third-party PC manufacturers are adapting to Windows’ transformation into an Agent OS — the technical challenges, the strategic responses, and the unresolved tensions that will shape the competitive landscape through 2027.
1. The Structural Asymmetry: What Microsoft Has That OEMs Don’t
1.1 Full-Stack Control
Microsoft’s position in the agentic Windows era is structurally different from any previous platform transition. The company controls every layer of the stack:
- The operating system: Windows 11 24H2 or later is required for Foundry Local, MXC, and most Windows AI APIs
- The security layer: Microsoft Execution Containers (MXC) is a Windows-native containment system generally available since October 2026
- The inference runtime: Windows ML, Foundry Local, and the Windows AI APIs are all Microsoft-managed
- The development framework: The Windows Agent Framework (WAF) is MIT-licensed and maintained by Microsoft
- The governance layer: Agent 365 and Windows 365 for Agents provide centralized agent management under Intune policy control
- The hardware reference: Surface Laptop Ultra serves as the reference implementation, with NVIDIA RTX Spark providing the silicon foundation
An OEM shipping a Windows laptop can use all of these components. But they cannot modify them, replace them with alternatives, or differentiate on their fundamentals. Every Windows PC maker is building on the same platform foundation.
1.2 The Timing Problem
The agentic Windows transition requires hardware that meets specific thresholds. Windows AI APIs require a Copilot+ PC specification: NPU with 40+ TOPS, 16GB+ RAM, and 256GB+ storage. Foundry Local supports any Windows hardware but delivers optimal performance on Copilot+ PCs or RTX Spark devices. MXC containment requires Windows 11 24H2 or later.
Many third-party OEM laptops currently in market do not meet these specifications. The 8GB RAM configuration — which has re-emerged as a mainstream option due to memory price increases — falls below the 16GB threshold. This creates a two-tier market: devices that can run the full agentic Windows stack, and devices that cannot.
2. The Hardware Adapter Layer: How OEMs Are Meeting the Silicon Challenge
2.1 The RTX Spark Coalition
NVIDIA’s RTX Spark platform, announced at COMPUTEX 2026, represents the most significant silicon opportunity for third-party OEMs. Unlike Microsoft’s earlier Copilot+ PC push — which relied on Qualcomm Snapdragon X Elite processors initially — RTX Spark is explicitly a multi-vendor platform.
The OEM roster for RTX Spark systems is broad: ASUS, Dell, HP, Lenovo, MSI, Acer, and Gigabyte are all shipping RTX Spark devices in Fall 2026. Systems range from laptops to compact desktops designed for continuous operation. Over 30 laptop and desktop models are expected across these brands.
The specification envelope is consistent across OEMs:
| Component | Specification |
|---|---|
| GPU | Blackwell RTX, up to 6,144 CUDA cores |
| CPU | NVIDIA Grace, up to 20 cores |
| Interconnect | 600 GB/s NVLink-C2C |
| Unified Memory | Up to 128GB LPDDR5X |
| AI Performance | Up to 1 petaflop FP4 |
What differentiates OEM implementations is not the silicon — it is thermal design, form factor, display quality, software integration, and price positioning.
2.2 The Copilot+ PC Generation
For OEMs not shipping RTX Spark devices, the Copilot+ PC specification remains the primary target. Microsoft’s requirements are explicit: NPU with 40+ TOPS, 16GB+ DDR5 or LPDDR5 memory, and 256GB+ storage. Lenovo was the first OEM to explicitly label its product specs to this standard, distinguishing 8GB non-Copilot+ configurations from 16GB Copilot+ models even when both use the same AMD Ryzen AI chip.
Acer’s Copilot+ lineup — the Swift Go 16 AI, Swift Go 14 AI, and Aspire C AI — uses AMD Ryzen AI 300 Series processors with XDNA 2 NPU architecture delivering up to 50 TOPS. ASUS offers Snapdragon-powered devices like the Zenbook A14 with up to 45 TOPS NPU capability and claims up to 32 hours of battery life.
2.3 The AMD and Intel Path
For OEMs building systems around AMD and Intel silicon, Windows ML provides the framework for hardware-accelerated inference across CPU, GPU, and NPU. AMD has collaborated with Microsoft since Build 2025 to optimize Windows ML for AMD hardware accelerators, delivering improvements including:
- 4×8 overlay performance improvements
- Support for general asymmetric quantization, enabling third-party quantized models to run on NPU
- Improved support for A8W8 and A16W8 QDQ models
- 3× average compile time improvements for BF16 models
Intel’s OpenVINO execution provider is distributed through Windows Update, ensuring that Intel-based systems receive optimized inference runtimes without OEM intervention. Qualcomm’s Neural Network execution provider is similarly maintained for Snapdragon X Series NPUs.
This execution provider architecture is significant for OEMs because it decouples hardware support from application development. A developer building on Windows ML does not need to create separate builds for AMD, Intel, NVIDIA, and Qualcomm platforms. Windows ML dynamically acquires the appropriate execution provider for the target hardware.
3. The Software Differentiation Problem
3.1 The Platform Uniformity Trap
Every Windows PC running 24H2 or later has access to the same AI stack: Windows ML, Foundry Local, Windows AI APIs, MXC containment, and the Windows Agent Framework. The operating system provides the runtime. The silicon vendor provides the execution provider. The developer provides the agent.
Where does the OEM add value?
This is the central strategic question for third-party PC makers in the agentic era. In previous platform transitions — from DOS to Windows, from HDD to SSD, from integrated to discrete graphics — OEMs differentiated through hardware configuration, industrial design, and price. Those levers remain, but they are increasingly commoditized at the platform level.
3.2 The OEM AI Assistant Strategy
Each major OEM has responded by building proprietary AI assistants that sit above the Windows platform layer. These assistants are not replacements for Microsoft’s agent infrastructure. They are applications that leverage the platform to deliver OEM-specific value.
Dell: Dell Optimizer and Pro AI Studio
Dell Optimizer 6.x, which replaced both the older Dell Optimizer and MyDell applications, provides AI-powered optimization for power, battery, display, collaboration touchpad, and presence detection. It is Dell’s system-level management layer, available across Latitude, Precision, XPS, Inspiron, and Vostro product lines.
More strategically, Dell Pro AI Studio offers an AI toolkit with validated tools, frameworks, templates, and models. Dell claims this can reduce AI development and deployment time by as much as 75% — from six months to as little as six weeks. The positioning is silicon-agnostic: developers and IT administrators can build and manage AI software independent of the underlying hardware.
Dell’s XPS series demonstrates local AI capability independent of the RTX Spark platform. XPS 16, XPS 14, and XPS 13 models with Intel Core Ultra processors can run 20-billion-parameter language models locally and execute AI image generation in seconds, supporting over 100 local AI applications.
HP: AI Companion and ZBook Ultra
HP AI Companion is preloaded on select HP next-generation AI PCs and requires an NPU supporting 40–60 TOPS with 16GB or more of memory. The assistant provides three core capabilities: intelligent search, data analysis, and automatic performance optimization based on usage habits. It also maintains an optimal set of system drivers, BIOS, and firmware for peak performance.
On the hardware side, HP’s ZBook Ultra features up to 16 desktop-class CPU cores, discrete-like integrated graphics, and up to 128GB unified memory with the ability to assign up to 96GB RAM to the GPU. The Z2 Mini and ZBook Ultra are positioned for 3D design, graphics-intensive rendering, and local LLM workloads.
Lenovo: Qira and AI Now
Lenovo has taken the most ambitious approach to cross-device AI integration. Qira, described as “personal ambient intelligence,” runs across Lenovo Windows PCs and Motorola Android smartphones with shared unified memory across devices. It is integrated at the system level — users do not need a separate app to operate it. Qira replaces both the AI Now feature on Lenovo PCs and the Moto AI feature on Motorola smartphones.
Lenovo AI Now, built on a local LLM using Meta’s Llama 3.0, stores and processes all user data locally. On the ThinkPad X9, it delivers on-device intelligence for document organization, device management, and natural language queries across locally stored files.
Acer: Copilot+ Integration and Developer Focus
Acer’s approach emphasizes Windows platform integration. The company has publicly committed to making Windows “a great home for developers” and highlighted MXC containerization as a key capability for enterprise agent deployment. Acer’s Copilot+ lineup uses AMD Ryzen AI 300 Series processors with XDNA 2 NPU architecture.
3.3 The Integration Tension
A significant unresolved question is how OEM AI assistants interact with Microsoft’s agent infrastructure. Lenovo’s Qira and HP’s AI Companion are system-level agents that could potentially operate as agents within the Windows Agent Framework. But they are also proprietary platforms that compete for user attention with Copilot and third-party agents.
The Windows Agent Store, announced at Build 2026 with an 85% developer revenue share, creates a distribution channel for agents. Whether OEM assistants will be distributed through this store, or remain pre-installed applications outside Microsoft’s governance model, remains unclear.
4. Enterprise Deployment: The OEM’s Role in Agent Governance
4.1 MXC and Intune Integration
For enterprise customers, the value of MXC containment is that it integrates with existing Intune and Entra infrastructure. IT administrators define which files and networks agents can access, and MXC enforces those policies at runtime. Agents can be isolated through process containers, session isolation, WSL containers, virtual machines, or Windows 365 for Agents cloud PCs.
This is where OEMs have a genuine value-add opportunity. Enterprise fleet management — deployment, driver validation, BIOS configuration, and lifecycle management — is a core OEM competency that Microsoft does not replicate. Dell Command | Power Manager and Dell Pro AI Studio, HP’s AI Companion driver management, and Lenovo’s commercial fleet tools are examples of OEM software that extends platform capabilities for managed environments.
4.2 Agent 365 and Shadow AI
Agent 365, generally available since May 2026, addresses the shadow AI problem: employees installing and running agents on their Windows endpoints without IT approval. The platform detects agent activity, applies organizational policies, monitors risk, and provides a unified view of agent behavior across the fleet.
For OEMs selling into enterprise accounts, Agent 365 compatibility is a requirement rather than a differentiator. The question is whether OEM management tools will integrate with Agent 365’s registry sync — which already supports third-party platforms including AWS Bedrock and Google Vertex — or whether they will operate as separate management planes.
4.3 Windows 365 for Agents
Windows 365 for Agents provides a managed cloud PC environment for agent workloads, operating under Intune policy control. This creates an alternative deployment path for organizations that prefer cloud-based agent execution rather than local execution on employee devices.
For OEMs, this represents both an opportunity and a risk. The opportunity is that Windows 365 for Agents complements local device deployment, potentially driving demand for thin clients and cloud-managed endpoints. The risk is that organizations may choose to run agents exclusively in the cloud, reducing the requirement for high-spec local hardware.
5. The Memory Cost Crisis: A Structural Constraint
5.1 The 8GB Regression
One of the most significant challenges facing third-party OEMs is the memory cost environment. As DRAM prices have risen, the PC market has seen a regression toward 8GB configurations — precisely the threshold below which Windows AI APIs and the full Copilot+ PC experience are unavailable.
Microsoft has acknowledged this tension. The company has committed to optimizing Windows 11 to reduce memory footprint while simultaneously pushing local AI capabilities that require more memory. The result is a bifurcated market: 8GB PCs for general productivity, 16GB as the entry point for AI experiences.
Lenovo was the first OEM to explicitly encode this distinction in its product specifications, labeling 8GB configurations as Non-Copilot+ even when they use the same AMD Ryzen AI chip as 16GB Copilot+ models. This is a pragmatic response to market reality, but it also creates consumer confusion about what “AI PC” actually means.
5.2 The RTX Spark Premium
RTX Spark systems are premium devices. Entry points cluster around $2,599–$3,799, with high-memory configurations exceeding $4,500. This positions them for developers, creators, and enterprises — not mainstream consumers.
For OEMs, this creates a portfolio challenge. The RTX Spark segment offers differentiation and margin, but limited volume. The Copilot+ PC segment offers broader reach but depends on NPU-equipped processors from AMD, Intel, and Qualcomm — all of which compete with NVIDIA’s Grace CPU in the RTX Spark platform.
5.3 The x86 vs. Arm Question
RTX Spark is an Arm-based platform. Windows on Arm AI notebook penetration is projected to rise from 1.2% in 2025 to 3.2% in 2026, reaching 11.5% by 2029. x86-based AI notebooks are expected to hold approximately 50.7% of the total notebook market by 2029.
This means most third-party OEM revenue will continue to come from x86 platforms through the forecast period. But the highest-profile agentic AI devices — Surface Laptop Ultra, ASUS ProArt RTX Spark, Lenovo Yoga Pro 9n — are Arm-based. OEMs must maintain dual-platform strategies, investing in Arm64 Windows compatibility testing and driver development while continuing to ship x86 systems for the majority of their volume.
6. The Open-Source Leverage Strategy
6.1 Windows Agent Framework as Neutral Ground
The MIT-licensed Windows Agent Framework provides OEMs with a vendor-neutral platform for agent development. Because WAF decouples agent definition from runtime — a single YAML manifest deploys to Windows 11, Windows Server 2026, Windows IoT, Windows 365 Cloud PCs, and via Docker to Linux — OEMs can build agents that run across their product portfolios without platform-specific modifications.
This is significant for OEMs with diverse hardware lineups. An agent built for a Lenovo ThinkPad can run on a Lenovo Yoga, a Lenovo Legion gaming laptop, or a Lenovo ThinkCentre desktop without re-engineering. The same applies to Dell’s Latitude, Precision, XPS, and Inspiron families.
6.2 MXC Cross-Platform Containment
MXC works across operating systems, not just Windows. The containment technology supports process containers on Windows, macOS, and Linux; WSL containers on Windows 11; and micro-VMs on Windows 11 and Linux. The SDK provides Rust, .NET, Node.js, and TypeScript bindings.
For OEMs with heterogeneous fleets — Windows laptops, Linux workstations, and macOS devices — MXC provides a unified containment model. An agent’s policy can be defined once and enforced consistently across platforms.
6.3 Foundry Local Hardware Agnosticism
Foundry Local supports over 20 open-source LLMs and speech models through an OpenAI-compatible API on any Windows hardware. This means OEMs are not locked into specific silicon requirements to offer local AI capabilities. A laptop with an integrated NPU can run Foundry Local models, just as an RTX Spark device can run larger frontier models.
7. What OEMs Are Actually Shipping: A Model Survey
7.1 Dell
- Dell XPS 16 Creator Edition: RTX Spark, up to 128GB unified memory, 3.2K Tandem OLED, $3,799.99 entry
- Dell XPS 14/16 (2026): Intel Core Ultra, 20B parameter local LLM, AI image generation in seconds
- Latitude 7420/7430/7440: Enterprise fleet, Dell Optimizer support, MXC compatible
- Precision 3580/5570/7680: Workstation-class, full thermal and battery feature support
- Inspiron 15 3535: Budget mainstream, Battery Extender and Thermal Management support
7.2 HP
- HP OmniBook Ultra 16: RTX Spark, 16-inch 3K OLED, 99Wh battery, 17-hour runtime, $3,199.99
- HP OmniBook X 14: RTX Spark, $2,999.99 entry
- HP ZBook Ultra: AMD Ryzen AI Max PRO, 128GB unified memory, 96GB GPU-allocatable RAM
- HP Z2 Mini: Compact workstation, local LLM capability
7.3 Lenovo
- Lenovo Yoga Pro 9n: RTX Spark N1X, 20-core CPU, 6,144-core GPU, 128GB unified memory, available October 16
- ThinkPad X9: Lenovo AI Now, on-device Llama 3.0, local document search and device management
- IdeaPad Vibe: Copilot+ PC label for 16GB+ configurations only; 8GB labeled Non-Copilot+
7.4 ASUS
- ProArt P14: RTX Spark, 1.48kg, 13.9mm, 90Wh battery, $3,099.99 entry
- ProArt P16: RTX Spark, 4K 120Hz VRR, G-Sync, €3,399
- Zenbook A14: Snapdragon X Series, 45 TOPS NPU, up to 32 hours battery
7.5 Acer
- Swift Go 16 AI: AMD Ryzen AI 300, 50 TOPS XDNA 2 NPU, up to 24.9 hours video playback
- Swift Go 14 AI: Same platform, 14-inch form factor
- Swift Blade 14: Sub-799g, carbon fiber and magnesium alloy construction
7.6 MSI and Gigabyte
Both vendors are shipping RTX Spark systems in the Fall 2026 window. Specific model details and pricing have not been widely disclosed, but both are participating in the launch ecosystem alongside the larger OEMs.
8. The Unresolved Tensions
8.1 Platform Dependency vs. Differentiation
The core tension for third-party OEMs is that the agentic Windows platform is Microsoft-controlled. OEMs can build on Windows ML, MXC, Foundry Local, and the Windows Agent Framework. But they cannot own the platform. Their differentiation must come from hardware design, thermal engineering, software integration, enterprise management, and price — not from the AI runtime itself.
This is a familiar dynamic from the x86 era, where Intel and Microsoft controlled the platform and OEMs competed on configuration and price. The difference is that AI agents create new value pools — enterprise governance, developer tooling, local inference — where OEMs currently have limited footprint.
8.2 The Surface Competitive Dynamic
Microsoft’s Surface division has historically been positioned as a reference design rather than a volume competitor. But Surface Laptop Ultra at $2,599 — with RTX Spark, 128GB memory, and full Windows agent integration — competes directly with premium OEM devices from Dell, HP, and Lenovo.
The OEM response has been to emphasize form factor differentiation (ASUS ProArt P14 at 1.48kg, Acer Swift Blade 14 at sub-799g), enterprise management capabilities (Dell Optimizer, HP AI Companion), and price segmentation. Whether these advantages are sufficient against Microsoft’s vertical integration remains an open question.
8.3 The Developer Relations Gap
Microsoft’s pitch for the agentic Windows era is explicitly developer-focused. Build 2026 was a developer conference. The Windows Agent Framework is MIT-licensed and open-source. Foundry Local adds 20MB to an application. The message is that Windows is the best platform to build and deploy agents.
But developers building AI agents today predominantly use MacBook Pros or Linux workstations. The Verge’s assessment of the October event noted that the demos were compelling for developers and creators, but did not fully establish the value for mainstream Windows users. OEMs face the same challenge: convincing developers that their hardware is the best platform for agent development, when the platform itself is Microsoft-controlled and available on any Windows device.
Conclusion: Building on Someone Else’s Platform
The agentic Windows era presents third-party PC manufacturers with a paradox. The platform transition creates genuine opportunities — RTX Spark hardware, MXC enterprise deployment, local AI inference — that drive premium device sales and open new enterprise conversations. But the platform itself is Microsoft-controlled, and every OEM is building on the same foundation.
The companies best positioned to succeed are those that treat the Windows AI stack as infrastructure rather than product. The value proposition is not “our laptop runs agents” — every Copilot+ PC can make that claim. The value proposition is form factor, thermal design, enterprise management, developer tooling integration, and price positioning that differentiates a specific device for a specific use case.
Dell’s silicon-agnostic AI Studio, HP’s cross-device AI Companion, Lenovo’s Windows-Android unified memory with Qira, and ASUS’s sub-1.5kg RTX Spark workstation are examples of this approach. They do not compete with Microsoft’s platform. They build on it.
What remains unresolved is whether the premium hardware required for frontier local inference — RTX Spark systems at $2,599 and above — will achieve sufficient volume to sustain OEM investment, or whether the market bifurcates permanently into a premium developer tier and a mainstream tier where local AI capabilities are limited to smaller models on NPU-equipped devices.
For now, the strategy is clear: ship the hardware, integrate with the platform, build the management layer, and let the agent ecosystem determine which workflows justify the investment. The operating system’s role is changing. The OEM’s role is still being defined.
This analysis is based on publicly available announcements, product specifications, and platform documentation from Microsoft, NVIDIA, Dell, HP, Lenovo, ASUS, Acer, MSI, and Gigabyte as of October 2026. Feature availability, pricing, and system requirements are subject to change.