Project Zenith: Microsoft Defines the Local-AI PC With 64GB of Memory

Microsoft introduced Project Zenith in early September 2026: a Windows configuration designed to build and run AI models directly on the PC, without relying on the cloud. The minimum requirement says everything about the ambition: 64GB of unified memory and 250GB/s of memory bandwidth. The first devices use AMD’s Ryzen AI Halo silicon, and Nvidia keeps RTX Spark on the roadmap for this fall.

Quick answer: what is Project Zenith?

It is a Windows hardware-and-software profile for local AI: PCs with at least 64GB of unified memory (shared by CPU, GPU and NPU) and 250GB/s of bandwidth, capable of running large models without sending data to servers. The first devices use AMD Ryzen AI Halo silicon. It is the tier above the Copilot+ PC, which only requires a 40-TOPS NPU.

Why 64GB of unified memory?

Language models must fit in memory to run fast. A 30-billion-parameter model at reduced precision takes about 20GB; agents that load several models plus long context easily exceed 40GB. With unified memory, CPU, GPU and NPU access the same pool without copies — the architecture Apple has used in Macs for years and the one Nvidia adopted in RTX Spark with 128GB.

Copilot+ PC vs. Project Zenith

What about Nvidia RTX Spark?

Unveiled at Computex in June, RTX Spark is Nvidia’s superchip for Windows laptops and desktops: an Arm CPU, a Blackwell GPU and 128GB of unified memory, promising to turn Windows into an “agentic AI OS.” Since then, the company has confirmed one piece of the puzzle per month — agent security in July, graphics software in August and a reaffirmed fall schedule in September — while holding back the two details buyers care about most: official pricing and an exact ship date.

Why this matters to you

If you are shopping for a laptop, the message is that 16GB is no longer enough for serious local AI in 2027, and 32GB has become the floor. For developers, Zenith means training and fine-tuning models without paying for cloud time by the hour — which changes the cost of prototyping. And for privacy-sensitive businesses (legal, healthcare, finance), local AI is the only way to use advanced models without data ever leaving the building.

Frequently asked questions

Is Project Zenith a new version of Windows?

No. It is a configuration (hardware and software profile) of Windows 11 for on-device AI development and execution.

Which PCs qualify for Project Zenith?

The first devices use AMD Ryzen AI Halo, with at least 64GB of unified memory and 250GB/s of bandwidth. Nvidia RTX Spark machines are expected in late 2026.

When will Nvidia RTX Spark launch?

Nvidia maintains a fall 2026 schedule but has not announced pricing or an exact date.

At DigitalRadar, we follow the evolution of the AI PC. Stay on the radar so you do not miss the next launch.