NVIDIA is expected to announce the N1X laptop SoC at Computex 2026 in Taipei (June 2 to 5). This is NVIDIA's first system-on-chip designed for laptops: an ARM CPU paired with a Blackwell GPU, delivering CUDA in a thin-and-light form factor for the first time.
Sources: NVIDIA N1X preview | Computex 2026 coverage | Dell N1X listing leak
What the N1X Is
The N1X is a single-chip SoC combining CPU and GPU on a 3nm process, co-developed with MediaTek. Key specs based on confirmed information and supply chain leaks:
| Component | Spec |
|---|---|
| CPU | 20 ARM cores (10 high-performance + 10 efficiency) |
| GPU | 48 Blackwell SMs, 6,144 CUDA cores |
| Memory | Up to 128 GB LPDDR5X (unified, CPU + GPU shared) |
| Architecture | Blackwell (same gen as H100/H200 datacenter GPUs) |
| Process | 3nm |
| TDP | ~140W |
| Partner | MediaTek (CPU design and manufacturing) |
Source: ChatForest N1X preview | HotHardware Dell listing
Why This Matters for AI Development
The N1X changes a fundamental trade-off in local AI hardware. Before N1X, developers had two options for local model work:
Option A: MacBook Pro with Apple Silicon. Up to 128 GB unified memory. Runs large models (70B Q4 fits at ~35 GB). Great battery life. But limited to MLX and MPS for framework support. No CUDA.
Option B: Windows gaming laptop with NVIDIA RTX. Full CUDA support. But discrete GPU VRAM caps at 24 GB (RTX 5090). Cannot fit 70B models. Heavy, hot, short battery life.
With N1X: CUDA on an ARM laptop with 128 GB unified memory. A 70B Q4 model fits comfortably at ~35 GB. Full Flash Attention, TensorRT, and llama.cpp CUDA optimizations run natively. In a thin-and-light form factor.
| Platform | Max RAM | CUDA | 70B Q4 Possible | Portability |
|---|---|---|---|---|
| NVIDIA N1X | 128 GB | Yes | Yes | High |
| Apple M4 Max | 128 GB | No (MPS) | Yes | High |
| RTX 5090 Laptop | 24 GB VRAM | Yes | No | Low |
| Qualcomm Snapdragon X Elite | 64 GB | No | No | High |
Source: ChatForest N1X vs competition table
AMD Strix Halo: The Competitor Already Shipping
While N1X is not expected until late 2026, AMD's Strix Halo (Ryzen AI Max+ 395) is already shipping in laptops today. It is a direct competitor on unified memory and local AI capability.
| Spec | AMD Strix Halo (Ryzen AI Max+ 395) |
|---|---|
| CPU | 16-core Zen 5, up to 5.1 GHz |
| GPU | RDNA 3.5, 40 Compute Units (Radeon 8060S) |
| Memory | Up to 128 GB LPDDR5x (up to 96 GB allocatable as VRAM) |
| Memory bandwidth | 276 GB/s |
| NPU | XDNA 2, 50 TOPS |
| AI framework | ROCm (CUDA alternative) |
| Availability | Shipping now |
| Price | From $1,999 (ASUS ROG Flow Z13) |
Source: AMD official benchmarks | Strix Halo laptop review
AMD's claims on AI performance:
- 2.2x faster than a desktop RTX 4090 on Llama 70B Nemotron inference (96 GB VRAM fits the whole model, the RTX 4090's 24 GB cannot)
- 3.9x faster than MacBook Pro M4 Pro (48 GB) on Stable Diffusion 3.5
- Runs 70B parameter LLMs locally
The trade-off: AMD's ROCm ecosystem is still catching up to CUDA in framework breadth. Memory bandwidth (276 GB/s) trails Apple M4 Max (546 GB/s) by roughly 2x, which directly impacts LLM token generation speed. For developers dependent on CUDA-specific libraries like Flash Attention or TensorRT, Strix Halo does not help.
On the other hand, Strix Halo ships today. N1X does not. For anyone who needs 128 GB unified memory on a Windows or Linux laptop in 2026, Strix Halo is the only shipping option.
Source: AMD technical article
Competitive Comparison
| Platform | CPU | GPU | CUDA | Max Unified RAM | Availability |
|---|---|---|---|---|---|
| NVIDIA N1X | 20-core ARM | 6,144 CUDA (Blackwell) | Yes | 128 GB | Late 2026 |
| AMD Strix Halo | 16-core Zen 5 | 40 RDNA 3.5 CUs | No (ROCm) | 128 GB | Now |
| Apple M4 Max | 16-core Apple | 40-core GPU | No (MPS) | 128 GB | Now |
| Qualcomm Snapdragon X Elite | 12-core Oryon | Adreno | No | 64 GB | Now |
The Catch: Performance and Timing
Performance. The integrated GPU runs at lower clock speeds with shared memory bandwidth compared to a discrete RTX 5070 (which has the same 6,144 CUDA core count). Expect roughly 20-25% of discrete RTX 5070 AI compute performance. Still meaningfully ahead of any current Windows ARM laptop, and sufficient for running 13B to 70B models locally at usable speeds.
Source: ChatForest realistic expectations
Timing. First devices expected October 2026. Broad availability in early 2027. Dell and Lenovo have both been confirmed through leaks as OEM partners. Eight specific Dell and Lenovo models are expected. Asus is also reportedly developing N1X designs.
Source: Igor's Lab Computex timeline | Lenovo N1X leak
OEM Partners and Device Categories
Lenovo's internal ADFS authentication system has been spotted referencing an "NVIDIA N1x Portal," confirming active development. Earlier support page leaks listed several unreleased Lenovo systems:
- Legion 7 15N1X11 (gaming laptop)
- Yoga Pro 7
- IdeaPad Slim 5
- Yoga 9 2-in-1
Dell has also shipped engineering samples. A Dell DVT (Design Validation Test) unit with the N1X was logged in November 2025, suggesting the hardware has been in validation for months.
Source: TweakTown Lenovo leak | HotHardware Dell DVT unit
What Still Needs to Happen
N1X is not a shipping product yet. Several open questions remain:
Software validation. PyTorch, llama.cpp, TensorRT-LLM, and ComfyUI all need to certify on the new architecture. NVIDIA has a strong track record of day-one framework support, but this is a new architecture class where the validation pipeline has not been exercised before.
Windows on ARM maturity. The ecosystem is still at 4-6% PC share. Gaming anti-cheat support and driver maturity remain open issues.
DRAM pricing. A global DRAM shortage has pushed memory prices up. A unified-memory SoC with 256-bit LPDDR5X feels that pain directly.
Source: ChatForest risks and open questions | WinBuzzer software concerns
Summary
| Question | Answer |
|---|---|
| What is it? | NVIDIA's first laptop SoC with ARM CPU + Blackwell GPU |
| When does it ship? | First devices October 2026, broad availability early 2027 |
| Key advantage | CUDA on ARM with up to 128 GB unified memory |
| Key risk | First-gen ARM + CUDA in a laptop: software validation unproven |
| Who is building it? | Dell, Lenovo, Asus confirmed through leaks |
The N1X is the most significant PC hardware announcement for local AI development since Apple Silicon. Whether NVIDIA can execute the software side as cleanly as the hardware side will determine whether this is a watershed moment or a promising but rough first generation. Early 2027 will tell the story.
Sources: ChatForest N1X preview | Igor's Lab Computex coverage | Lenovo N1X portal leak
Learn More About N1X
NVIDIA's Computex 2026 keynote covers the N1X and next-gen AI hardware.
Watch the Keynote — Free