Product1 distinct publisher3 min readPublished
The first RTX Spark laptop got a hands-on at IFA 2026, where the figure that decides whether an agent can run on your own desk turns out to be the shared memory ceiling rather than any price, since Wired reports none.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Someone speccing a machine for a team that wants agent loops running in-house will click the 16-inch model, because that is what the top of the range is for, and land on a 64 GB ceiling [8]. The 15-inch Yoga Pro 9n, which Wired says otherwise seems to perform the same, configures to 128 GB [9], twice the room [17]. On this chip that number carries more weight than usual: the Grace CPU, the Blackwell GPU and the memory sit in one package, closer to Apple silicon than to a Windows laptop with a discrete card [5], so the weights and the working context draw on the same pool as everything else the machine is doing. Which model stays resident is settled at order time, in a dropdown.
The physical envelope is the genuinely new part. Wired puts the 16-inch chassis at 0.69 inches and calls that 0.03 inches shy of a 16-inch MacBook Pro [4], which makes the reference machine roughly 0.72 inches [18]. A box the size of a premium creative laptop is now being sold as somewhere to park an agent. Missing from the hands-on is the other half of that pitch: no price for any announced laptop or mini PC, and no sustained throughput figure [19]. Wired's own performance estimate comes from core counts rather than a benchmark, landing the integrated GPU between a laptop RTX 5070 Ti and an RTX 5080 [6], and no battery-life claim was offered [7].
Wired frames the demand here as rising interest in running agentic workflows like OpenClaw locally [13]. Teams tell themselves that means the whole loop moves in-house. What a launch like this actually establishes is that a certain quantity of memory is purchasable in a certain thickness. Tokens per second at minute sixty of a long run, on the configuration you can order rather than the one on the show floor, is what decides whether the local step replaces the API call or merely precedes it.
The forcing function has two axes. First, does your model plus its working context fit under the ceiling you can buy, 64 or 128 GB. Second, is the work bursty and interactive, or a loop that grinds for an hour unattended. Fits and bursty is the laptop. Fits and long-running is the Mac-mini-sized box, and there are several now: Acer's SFF RTX Spark and Asus's ProArt Mini PC at up to 128 GB with a claimed petaflop of AI performance [12], plus Lenovo's AMD-powered ThinkCentre X Ultra at the same memory figure on a Ryzen Max+ Pro 495 [15]. Does not fit means the API bill stays, whatever the deck promises. The vendors have converged on 128 GB as the useful ceiling; the comparison against your current invoice is still yours to build, because nobody in Berlin published a price [19].
Ranked by verification strength, evidence, and original report placement.
Nvidia unveiled the RTX Spark superchip and a handful of future laptops using it three months before IFA 2026.
At IFA 2026 in Berlin, Wired had its first hands-on look at one of the RTX Spark laptops, alongside announcements of RTX Spark-powered mini desktop PCs.
The laptop shown was the Lenovo Yoga 9n 2-in-1, a 16-inch machine with a 360-degree hinge and a 2880 x 1800 OLED touch display at 120 Hz.
The Yoga 9n 2-in-1 is 0.69 inches thick, which Wired describes as 0.03 inches shy of the 16-inch MacBook Pro.
The RTX Spark chip combines an Arm-based Grace CPU with up to 20 cores and a Blackwell RTX GPU with up to 6,144 cores, packaging CPU, GPU and memory into a single system-on-a-chip that shares more with Apple silicon than a conventional Windows laptop.
Wired reports there was no mention of battery-life claims for the Spark laptops.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · September 3, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Amazon stopped eating the memory bill, and the Echo Dot went up 60%1 distinct publisher
product
Google's Googlebook merges Android and ChromeOS at last, with the Gemini terms still blank1 distinct publisher
security
Eclypsium counted 1,051 AI-infrastructure vulnerabilities across 12 vendors in 38 days1 distinct publisher
product
Nvidia's server partners are passing the DRAM bill straight to Microsoft, Google and Oracle2 distinct publishers
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One reporter, one machine actually touched
Wired handled the 16-inch Yoga 9n and reports it to the hundredth of an inch, so the chassis and display details are as solid as this kind of coverage gets. The 15-inch Pro that carries the decisive 128 GB option was never in the room, the core counts arrive inside Nvidia's own quotation marks, and the petaflop belongs to Acer and Asus. Precise where measured, borrowed everywhere else.
Spec sheets and a show floor
Nothing here has reached a buyer. The Dell, Asus, Microsoft and HP machines are dated only to "this fall"; Acer's mini PC and Lenovo's AMD desktop exist as announcements; the single hands-on lasted an afternoon in Berlin. The nearest thing to a market signal is a price on the AMD box, and that machine does not use the chip this story is about.
Headline runs ahead of the piece under it
"The First AI PCs Are Here" sits above a report that concedes price is the crucial missing element and that nobody mentioned battery life. The GPU class is arithmetic, the petaflop is marketing, and no model was run on any of these machines. The overstatement is real but shallow, because Wired keeps flagging its own gaps as it goes.
Access granted by the people being covered
IFA hands-on time is given, not obtained: Nvidia, Lenovo and Acer chose which machine could be touched and which numbers went out, and the two a buyer would actually weigh — cost and sustained performance — are precisely the two withheld. To Wired's credit it names the gap and reaches for Apple's $6,139 128 GB MacBook as a yardstick, then closes with a subscription pitch of its own.
Nothing to check it against
A single account of a single afternoon, with no second outlet to test the specification sheet. The measurements will survive; the 64-versus-128 GB split holds only until Lenovo publishes a configurator, and the performance framing stands as a placeholder until somebody actually loads a model onto one of these.