Acer’s smallest AI bets yet showed up at IFA 2026 in Berlin. The Acer SFF RTX Spark design squeezes NVIDIA’s RTX Spark superchip into a compact desktop. The Veriton RI110 AI Mini Workstation takes the Intel route instead. Engadget surfaced both announcements, yet the substance comes from Acer’s own newsroom. Together they frame Acer’s answer to agentic AI: tiny boxes that run big models at home.
One petaflop in a shoebox
The SFF RTX Spark is a design showcase, not a shipping product. Acer says availability will be announced at a future date. Still, the spec sheet reads like a workstation. It pairs a 6,144-core Blackwell RTX GPU with a 20-core NVIDIA Grace CPU. The NVLink-C2C interconnect ties the two together. As a result, the design targets up to one petaflop of AI compute. It also carries up to 128 GB of unified memory, enough to feed local agents without a cloud round trip.
NVIDIA unveiled RTX Spark with Microsoft at GTC Taipei in May. The platform bundles CUDA, RTX, DLSS, TensorRT, OptiX, Reflex and G-SYNC on one chip. Designed for creators, AI developers and gamers, NVIDIA RTX Spark brings NVIDIA’s full-stack AI platform and full suite of RTX technologies to compact desktops. New tricks arrive too, including DLSS 4.5 Ray Reconstruction and RTX Video with 4x Frame Generation. Fall laptops and desktops arrive from ASUS, Dell, HP, Lenovo, Microsoft Surface and MSI. So Acer’s box joins a crowded first wave. Meanwhile the Arm-on-Windows push keeps widening, from this superchip to Qualcomm’s Snapdragon C laptops.
The Veriton RI110 bets on Intel
The Veriton RI110 is the concrete product. It pairs an Intel Core Ultra X7 processor 358H with Intel Arc B390 graphics. Acer says it supports AI models of up to 120 billion parameters. Up to 96 GB of LPDDR5X memory and up to 4 TB of PCIe Gen 4 storage back that claim. It ships as a Copilot+ PC on Windows 11 Pro. Microsoft’s assistant is therefore baked in from the first boot.
Acer also wants the RI110 to host third-party agents. Its Qubi Claw software runs local workflows inside an isolated sandbox. Setup takes one click, and ready-made skills cover research, travel planning and audio transcription. Instead of locking buyers in, the box lets them download and deploy their preferred agentic AI solution. Acer also says Qubi Claw will keep adding compatible models, skills and channels over time. For heavier work, an OCuLink port adds up to 64 Gbps of bandwidth for external GPUs and NVMe storage. Acer is not new to this shelf, either. Its Veriton GN100 already runs NVIDIA’s GB10 Grace Blackwell superchip today.
Availability and the fine print
The Veriton RI110 reaches North America in the fourth quarter of 2026. EMEA follows in the first quarter of 2027. Acer has not released pricing, and exact specifications will vary by region. The SFF RTX Spark, by contrast, remains a promise: a design study that proves the superchip fits a shoebox. Copilot+ branding now spans everything from Samsung’s Galaxy Book6 to this mini workstation. So the label alone will not sell it.
Connectivity is genuinely pro-grade, however. Wi-Fi 7 and Bluetooth 5.4 handle wireless duty. Dual LAN covers 10 GbE and 2.5 GbE, and two USB 4 Type-C ports plus HDMI 2.1 round out the back. All of it fits a chassis 138.5 mm wide that weighs about a pound and a half. A fingerprint reader guards sign-in. Thus the RI110 aims at offices and creators who want agents local, private and expandable. Whether the petaflop promise ships as rated will wait for independent tests, since NVIDIA’s numbers are FP4 peak claims. For now, Acer has two small boxes with big ambitions. One of them is already on the calendar.
