Summary of Surface and Windows 7/10 events: Surface Laptop Ultra and a series of AI upgrades on Windows 11
On October 7 in San Francisco, Microsoft held an event to introduce a series of technology upgrades surrounding the Windows 11 operating system, artificial intelligence models running locally and new hardware devices in cooperation with NVIDIA.
This article will summarize for you all that Microsoft announced at that event.
First of all, currently more than 40% of personal computers produced to serve corporate customers belong to the Copilot+ PC standard. This number shows that the business laptop segment is shifting extremely quickly to devices that integrate specialized NPU hardware to directly handle internal AI tasks. The strong popularity of Copilot+ PC in professional working environments confirms that local security features, energy optimization and AI task automation capabilities are gradually becoming mandatory hardware standards for new generation computers.
Microsoft Execution Containers Platform
Microsoft officially released the Microsoft Execution Containers feature, abbreviated as MXC, widely on Windows 11. This is a flexible sandbox environment at the execution level that helps isolate the tasks of AI agents. The key to MXC lies in its ability to separate identities between users and agents right within the operating system. Instead of forcing all tasks to run in the same environment, the system breaks down isolation into many levels: from the lightest such as Process (direct process), Session (user session), WSLc (WSL-based container for Linux), to hardware-level isolation with Micro VM (micro virtual machine with separate kernel), full VM virtual machine and the highest is pushing straight to the cloud via W365.
When Microsoft Defender detects malicious instructions from a website to take control of an agent, the system will immediately block that agent's access without interrupting the user's work. Businesses can set up centralized management policies through the Agent 365 toolkit, applying uniformly to all devices in the network.
Hybrid Intelligence và các mô hình AI chạy trực tiếp trên máy
Hybrid Intelligence is Microsoft's hybrid AI architecture orientation, flexibly combining local processing capacity on the device (local) and large-scale computing power from the cloud (cloud). Instead of depending entirely on one party, the system will automatically distribute: tasks that need immediate response and private data security will run directly using the internal NPU/GPU, while complex analysis problems that require huge models will be sent to the cloud for processing. This model helps optimally balance performance, latency and operating costs. Users can not only take advantage of the device's new hardware power (like Copilot+ PC), but also have no capacity limitations when needing to exploit the power of remote AI super servers.
Microsoft brings the 130 billion parameter scale MAI-Code 1.1 Flash programming model to run directly on the device using 3-bit quantization technique, reducing memory capacity by nearly 80% but still maintaining a context window of 256,000 tokens. Besides, Nvidia's Neatron model with more than 70 billion parameters at 2-bit level only takes about 20 GB of memory, while DeepSeek Flash V4 with 284 billion parameters at 1.6-bit level only requires 60 GB of memory to operate on personal computers.
The Windows ML library is also integrated with Llama.cpp to expand compatibility with many open source models on CPU, GPU and NPU.
Upgraded taskbar experience and Copilot features
The Copilot application is extended with three capabilities including reading local file context, operating directly on the system, and allocating tasks to the local model. The Copilot Code feature supports creating native Windows applications from a description statement, then running tests in MXC's secure environment. The Copilot Autopilot feature can automatically browse emails, gather tax records scattered on hard drives, rename files, compress data and prepare letters to send to accountants without uploading personal data to the cloud.
Surface Laptop Ultra and RTX Spark platform officially go on sale
Microsoft introduces the Surface Laptop Ultra model aimed at programmers and content creators. The machine is designed based on the NVIDIA RTX Spark superprocessor, made up of two large chips put together. This hardware possesses an AI computing capacity of 1 Petaflop, is equipped with up to 128 GB of unified memory and supports DirectX, OpenGL graphics scripts and the CUDA platform. The body is less than 18 mm thin, weighs less than 2.04 kg, has a cooling system that increases efficiency by 2.5 times and maintains more than 99% of performance when running on battery. The PixelSense Ultra Touch screen supports high-brightness HDR display, and comes with a Magnetic Connect magnetic charging port integrated into the USB-C port.
Included with this model is a developer device called Surface RTX Spark Dev Box with pre-installed Project Zenith environment.
In addition to targeting programmers and creators, RTX Spark will support many games based on NVIDIA's existing platforms without the need for developers to work much on compatibility issues. In the immediate future, in 2027, there will be Call of Duty coming to RTX Spark.
DGX Station PC and workstation ecosystem for Windows
Major hardware partners simultaneously introduced laptops running NVIDIA RTX Spark including Lenovo Yoga 9 2-in-1, Asus ProArt, Dell
In the professional segment, Microsoft and NVIDIA bring Windows to DGX Station desktop workstations such as Dell Precision Pro and HP ZGX Fury AI Station. This configuration supports the simultaneous operation of up to 32 AI agents, serving to locally run models with a scale of more than 1,000 billion parameters such as DeepSeek V4 Pro, Llama V4 Maverick and Kimi K2.6 without needing to rent a cloud server cluster.