There’s still a lot we don’t know about NVIDIA’s RTX Spark AI chip – we’re still waiting for deeper technical details and pricing for the first batch of systems – but it has a good chance of completely changing the way we think about Windows PCs. RTX Spark should deliver the raw performance that the first batch of Copilot+ systems lacked, and it will also push Microsoft to make the Windows experience even better for Arm processors (something I argued was the real highlight of the entire Copilot+ initiative, more so than AI support).
With Qualcomm’s latest Snapdragon chips powering low- and mid-range Windows PCs, NVIDIA has the opportunity to fill the gap at the high-end, delivering powerful (and likely very expensive) PCs that will appeal to content creators, developers and memory-hungry AI enthusiasts.
Windows could finally have its Apple Silicon moment this year – a bigger effort to optimize the aging operating system for more efficient Arm chips, while rethinking what’s possible when you can give GPUs direct access to insane amounts of memory. Comparisons to the MacBook Pro were immediately clear from the first batch of RTX Spark machines like the Surface Pro Ultra and ASUS’ new ProArt systems. Of course, powerful Windows laptops are nothing new, but they usually had tradeoffs like thick, heavy cases or huge battery-draining power demands. What if you could get a powerful PC that doesn’t look like a brick and lasts as long as a low-power ultraportable?
This is the promise of RTX Spark systems, but many details still remain up in the air. Based on NVIDIA’s description of the chip, RTX Spark appears to be the same hardware found in the DGX Spark AI workstation that launched last year for $3,999, but now sells for $4,699. This makes me think that early RTX Spark systems could start around $4,000 or more. Its 20-core processor is made up of 10 Cortex This hardware is currently only found in the Mediatek Dimensity 9500 chip, which powers the new Oppo Find X9 Pro. NVIDIA’s Arm cores are also slower than the latest Snapdragon Elite Oryon cores, as well as Apple’s M5 chips. If NVIDIA charges extra for RTX Spark systems, we would expect them to use at least CPU cores from 2026.
The real stars of the RTX Spark systems will be their 6,144 RTX Blackwell GPU cores, identical to the RTX 5070 desktop GPU, as well as their support for up to 128 GB of unified memory. Conceptually, it’s not far from what Apple did with its M-series chips, which also combined Arm cores, beefy GPUs, and tons of unified memory. Apple Silicon has given Macs a huge performance and efficiency boost over PCs, and the entire Windows ecosystem is struggling to keep up.
The only competition for NVIDIA hardware on the PC side is AMD’s Ryzen AI Ultra line, which are also complex system-on-a-chip designs with powerful CPU cores, integrated graphics, and a powerful NPU. And built-in memory. Last year’s flagship Ryzen AI Max+ 395 (aka “Strix Halo”) chip supported up to 128GB of RAM, and the new AI Max 400 series takes that up to 192GB of unified memory.
The Ryzen AI Max line is still based on the x86-64 architecture, giving them full compatibility with older Windows software, but none of the power efficiency benefits of Arm cores. NVIDIA RTX Spark systems, like Snapdragon X Elite Copilopt+ machines, will need to run x86 software through Microsoft’s Prism emulator. I found it to be a pretty seamless experience when reviewing the Surface Pro Copilot+ in 2024, and Microsoft says it’s also working with NVIDIA to further optimize Windows 11 for the RTX Spark.
Last year, Microsoft’s Copilot+ initiative for AI PCs seemed like a dud, but I argued at the time that it at least pushed PC makers to solidify new standards for their high-end laptops. Eventually, almost everyone started offering their PC with at least 16 GB of RAM. The real failure of Copilot+ was that it relied on AI features that few people wanted and which posed many privacy concerns, like Microsoft’s Recall.
Things are a little different now, as many developers and power users are adopting AI agents like OpenClaw. NVIDIA is clearly trying to target this crowd with the RTX Spark systems – people willing to spend a lot of money to get PCs with lots of RAM so they can use the latest AI tools and models. (These people also explain why the Mac Mini has been out of stock for months, and you can partly blame them for the death of the $599 model.)
Beyond AI enthusiasts, I could also see non-AI content creators and developers tempted by the possibility of an exorbitant amount of unified memory. RTX Spark systems would theoretically be able to handle massive video editing jobs much better than existing laptops. Their Blackwell GPUs also mean you’ll have full access to NVIDIA’s DLSS AI scaling and frame generation for gaming (and you’ll likely have more than enough overhead for serious gaming while working on the AI ​​and doing background rendering work).
As powerful as RTX Spark systems may be, they will ultimately be limited by their price. From what we know so far, these machines will be incredibly expensive for deep-pocketed consumers. There is hope for the future, however: rumors indicate that NVIDIA is working on stripped-down versions of the chip for cheaper systems and potentially true ultraportables. And hopefully at that point the RAMaggedon will be over and more people will actually be able to afford RTX Spark systems.
