Nvidia Enters the PC Chip Market at Computex 2026 — RTX Spark Combines ARM CPU, Blackwell GPU, and 1 Petaflop of AI

Computex 2026, running June 1 through 5 in Taipei, produced a hardware announcement that changes who competes in the PC chip market. Nvidia unveiled RTX Spark — a superchip combining an ARM-based CPU with a Blackwell GPU on a single package — and immediately secured partnerships with Dell, HP, Lenovo, ASUS, and Microsoft for devices shipping this fall. The company that has dominated AI infrastructure for three years is now directly in the consumer PC market, competing with Intel, AMD, and Qualcomm for the laptop and desktop customers it previously only reached indirectly through discrete GPUs.

RTX Spark is designed as an "agentic AI PC" platform. The architecture connects a 20-core ARM CPU — ten high-performance Cortex-X925 cores paired with ten efficiency Cortex-A725 cores, clocked up to 4.1 GHz — to a Blackwell GPU with 6,144 CUDA cores via NVLink-C2C at 600 GB/s chip-to-chip bandwidth. Unified memory reaches 128 GB of LPDDR5X with 300 GB/s bandwidth. AI performance is rated at 1 petaflop — enough to run 120-billion-parameter language models with context windows up to one million tokens on the device, without a cloud connection. The CPU was co-developed with MediaTek.

What is actually shipping and when

The device lineup announced at Computex is substantial. Over 30 laptops and 10 desktops are confirmed for fall 2026. Microsoft's Surface Laptop Ultra carries a 15-inch mini-LED display at 2880×1920 and 2,000 nits peak brightness, weighing under 4.5 pounds. ASUS's ProArt P16 and P14 bring OLED displays, 99.9 Wh batteries, and configurations up to 128 GB of unified memory. Dell is bringing an RTX Spark variant of the XPS 13. Lenovo's Legion 7 applies the chip to gaming with a 245W power supply for sustained boost performance. HP and MSI have confirmed RTX Spark product lines without announcing specific models yet.

Thin laptop targets are aggressive: Nvidia describes chassis as slim as 14 mm and as light as three pounds. The power envelope runs from 45W to 80W TDP with transient boost support. Three generations of the platform are already on the roadmap — current Blackwell with LPDDR5X, followed by Vera Rubin Spark with LPDDR6, then Rosa Feynman Spark with next-generation memory.

Pricing has not been disclosed by any OEM. Based on unified-memory configurations and premium positioning, industry estimates for Surface Laptop Ultra range from $3,000 to $7,000 depending on configuration. These are not mass-market budget devices.

The performance picture — honest benchmarks

Early benchmark data from Computex gives a clearer picture than press materials alone. RTX Spark beats the standard Apple M5 by 54% in Clang compiler throughput — a CPU-intensive task where the ARM core count and architecture should be a genuine advantage. It comes in slightly behind the M5 Pro in the same test. In broader multi-threaded comparisons, RTX Spark runs roughly a third slower than the M5 Pro and M5 Max, which remain the performance leaders in the thin-and-light segment.

The GPU side is where RTX Spark makes a more compelling argument. The 6,144-core Blackwell GPU outperforms integrated graphics on any competing platform for creative workloads, gaming at lower resolutions, and — most relevantly for 2026 — local AI inference. A chip that can run a 120-billion-parameter model on-device without cloud latency or API costs addresses a real workflow need that M5 Pro cannot match at equivalent price points, regardless of which compiler benchmark wins.

One headwind remains significant: ARM Windows application compatibility. Not every Windows application runs natively on ARM, though Microsoft has made substantial progress on x86 emulation performance and more developers are compiling ARM-native builds. The platform's long-term success depends partly on how aggressively the Windows developer ecosystem moves to ARM-native, and that transition has historically been slower than optimistic timelines suggested.

Vera CPU: the data centre play running in parallel

RTX Spark is Nvidia's consumer story at Computex. The more financially significant announcement for Nvidia's business is the Vera CPU, an ARM-based data centre processor targeting agentic AI workloads. Vera is deployed as part of the Vera Rubin rack platform — meaning hyperscalers that purchase Rubin AI infrastructure automatically acquire Nvidia CPUs as part of the package, bypassing Intel and AMD in the data centre purchasing decision rather than competing against them directly in a CPU procurement.

Nvidia is projecting $20 billion in Vera CPU revenue in the current fiscal year from a zero baseline. Early customers include OpenAI, Anthropic, SpaceX, Dell, Oracle, and CoreWeave. All four major cloud providers — AWS, Google Cloud, Microsoft Azure, and Oracle — are confirmed early deployers of the Vera Rubin platform. Nvidia's performance claims for Vera show a greater than 55% advantage over Intel's flagship data centre CPU and an 11% lead over AMD's top chip on agentic AI workloads, with token generation running 1.8 times faster than x86 alternatives.

AMD and Intel also showed up

Computex is not a Nvidia-only show. AMD launched two new processors and a globally-available discrete GPU. The Ryzen 7 7700X3D for the AM5 platform offers eight cores, 104 MB of total cache, and a 4.5 GHz boost clock at $329, launching July 16. The Ryzen 7 5800X3D 10th Anniversary Edition — a final salute to the AM4 platform — arrives at $349 on June 25, notable because AMD simultaneously confirmed AM5 socket support through 2029, giving users a clear upgrade path. The Radeon RX 9070 GRE, an RDNA 4 discrete GPU with 12 GB of memory claiming 21% average faster performance than competitors at 1440p, reached global availability this week at $549 MSRP.

Intel's showing centred on Panther Lake, its first processor built on the 18A process node — the manufacturing technology Intel is using to pitch external foundry customers. The handheld gaming PC variants of Panther Lake, shown at Computex with OEM partners including MSI, OneXPlayer, and GPD, deliver 180 total TOPS of AI performance and a 60% multi-threaded improvement over the previous generation at equivalent power. A potential Microsoft Xbox-branded handheld using Panther Lake was mentioned in the context of OEM partnerships without official confirmation.

What the week means for the PC market

The PC market has been waiting for an architectural inflection point that would give consumers a reason to upgrade beyond marginal year-on-year improvements. RTX Spark offers a genuine step change in on-device AI capability — 1 petaflop is meaningfully different from the 100 to 200 TOPS that previous "AI PC" marketing described, and the ability to run frontier-class models locally is a capability shift rather than a spec bump. Whether that capability translates into purchase decisions depends on how many workflows people actually want to run locally rather than through cloud APIs, and on whether the Windows ARM ecosystem matures fast enough to make the transition invisible.

What is unambiguous is that Nvidia has entered the PC silicon market with credible hardware, serious OEM partners, and a roadmap. The company that was a GPU supplier to laptop manufacturers is now a chip platform competing for design wins against the processors those laptops previously used. Jensen Huang's stated goal of owning every layer of the AI stack is now visible in consumer products as well as data centre infrastructure.