NVIDIA confirms the first RTX Spark laptops for fall 2026: Grace CPU, Blackwell GPU, 128GB unified memory
NVIDIA has put a date on its most ambitious PC project: the first RTX Spark laptops arrive in fall 2026, and we now know who is building them. ASUS and MSI are first out of the gate, with Acer and GIGABYTE following and HP and Microsoft preparing their own machines. The pitch is a single device that fuses AI workstation, content creation rig and gaming laptop into one slab.
Key Takeaways
- RTX Spark laptops launch in fall 2026, led by ASUS and MSI, with Acer, GIGABYTE, HP and Microsoft to follow.
- The platform pairs a 20-core NVIDIA Grace CPU with a Blackwell RTX GPU and up to 128GB of unified memory.
- NVIDIA claims up to one petaflop of AI performance and 100FPS at 1440p in modern AAA games.
- Built with MediaTek, Spark marks NVIDIA’s return to making the whole computer, not just the GPU.
What RTX Spark actually is
Spark is NVIDIA’s answer to a question the industry has been circling since Apple Silicon: what happens when you design the CPU, GPU and memory as one system? The platform combines a 20-core Grace CPU with a Blackwell-generation RTX GPU, sharing up to 128GB of unified memory, the architecture that lets large AI models, giant textures and heavy timelines all live in one pool instead of being shuffled between separate CPU and GPU memory. NVIDIA claims up to one petaflop of AI performance in laptop form, a number that was rack-scale not long ago.
Unveiled at Computex 2026 and developed in partnership with MediaTek, Spark is also strategically notable: it is NVIDIA building the whole computer again, and it brings MediaTek into NVIDIA’s AI ecosystem for the first time.
One device, three jobs
The target user is the developer or creator who runs local AI workloads, edits or renders professionally, and games, and is currently carrying compromises for all three. Local AI is the differentiator: running serious models on-device needs memory capacity and bandwidth that conventional laptops cannot touch, and 128GB of unified memory is the entire argument. Content creation benefits from the same pool. Gaming is the flex on top, with NVIDIA’s CEO promising at least 100FPS at 1440p across modern AAA titles.
That gaming claim deserves its asterisk. Frame rate promises ahead of launch are marketing until independent benchmarks say otherwise, and we will test exactly that when hardware lands, the same way we approach every claim in our GPU reviews. But a Blackwell GPU with that memory subsystem hitting triple digits at 1440p is plausible, not fanciful.
Why the partner list matters
ASUS and MSI leading is the expected play: both have the premium gaming laptop engineering to absorb a first-generation platform. Acer and GIGABYTE following quickly suggests confidence in volume. The interesting names are HP and Microsoft. Microsoft building Spark hardware would put NVIDIA’s architecture inside the Surface line and signal that this platform is aimed at the mainstream premium market, not just the gaming fringe. The partner spread implies NVIDIA expects this to be a category, not a curiosity.
Who this machine is really for
Strip away the launch framing and the target user sharpens quickly. The developer running local models gets 128GB of unified memory, the single biggest bottleneck for serious on-device AI work, in a laptop for the first time. The creator gets a Blackwell GPU with a memory pool that laughs at 8K timelines. The gamer gets what is, on paper, a very fast gaming laptop as a side effect. The person who gets nothing is the casual buyer, because nothing about this platform is aimed at them, and the price will make that explicit.
That focus is the correct strategy for a first generation. New platforms survive on early adopters who have a burning need, and the local-AI developer community has exactly that. Win them, and generation two earns the right to chase everyone else.
The unified memory lesson
Spark is also the PC industry’s clearest answer yet to the architecture Apple normalized: one memory pool, engineered as a system, instead of CPU and GPU memory islands with a bus between them. The difference is scale and openness. NVIDIA is offering the unified model with PC upgrade culture, multiple OEMs and a gaming ecosystem attached. If it works, expect the rest of the Windows laptop industry to follow within two product cycles, and the old CPU-versus-GPU memory split to start looking like a legacy quirk.
The questions fall will answer
Price is the big one, and nobody is talking about it yet, which usually means brace yourself. A 20-core Grace CPU, Blackwell GPU and 128GB of unified memory is not a bill of materials that ships cheap, and first-generation platforms carry early-adopter tax. Thermals are the second question: a petaflop in a laptop chassis is a cooling statement waiting to be tested. Battery life is the third, because all this capability means nothing if it is chained to a wall.
The platform also lands in a genuinely changed laptop market, where efficiency-first designs have reset expectations for what portable performance looks like. Spark is NVIDIA’s argument that you can have the efficiency conversation and the power conversation in the same machine.
The fall lineup itself will be telling. ASUS has the ROG engineering pedigree and will likely position Spark as a creator-gaming flagship; MSI has spent years chasing the premium workstation crowd. How each partner differentiates on the same platform, thermals, displays, chassis and price, will reveal how much room NVIDIA leaves OEMs to compete and how tightly the reference design is controlled. First impressions of a new platform tend to stick, and NVIDIA knows the launch reviews of these machines are effectively reviews of Spark itself.
Platform details and partner announcements are on NVIDIA’s official site.
The bottom line
RTX Spark is the most interesting laptop platform in years: NVIDIA’s CPU, GPU and unified memory in one design, with real OEM breadth behind it. If the performance, thermals and price land anywhere near the pitch, fall 2026 redraws the premium laptop map. If not, it will still be the machine every developer wants on their desk.
Would you buy a first-generation Spark laptop? Tell the hardware desk.