Nvidia unveils Vera Rubin and RTX Spark at GTC Taipei launch; a day later, in a media Q&A, Jensen Huang related it all to an “agent computing instance” running from the data center to your next laptop. The consumer chip is shipping this fall — pricing hasn’t been announced, and the only gaming numbers so far are from Nvidia itself.
The hardware was unveiled at Nvidia’s GTC Taipei opening ceremony: Vera Rubin, the company’s next-generation multi-rack system is now in full production, and a new chip called the RTX Spark brings the same idea to Windows PC. But the next day was a media Q&A where Jensen Huang made a bigger argument – that computing is moving to an “agent” model where AI agents, not humans, are the primary users of software, and every class of machine (PC, car, data center, robot) needs to be redesigned around it. He continued to return to five growth drivers; here’s what each one means to a buyer in GCC.
“Agent computing instance” and PC as your agent
Huang’s framework is that the old model—program code running inside an application in an operating system—gives way to one in which a large language model (or several) sits inside a “harness” that calls tools and runs on a distributed infrastructure. It describes an agent as a combination of a model, trailers, tools, and runtime, with the model underlying and the trailer connecting everything. The practical conclusion that Nvidia wants you to draw is that the computer is no longer a tool you operate, but an assistant that manages the tools on your behalf.
This is a review, not a shipping feature, and it’s worth separating the two. The agent-first computer is the direction of travel; This year, the ships are really silicone.
Vera Rubin reaches volume and squeezes the supply chain
Huang calls Vera Rubin the most ambitious project in the company’s history, emphasizing that it’s not a single GPU, but a pod-scale system. Its NVL72 configuration combines 36 Vera CPUs and 72 Rubin GPUs over sixth-generation NVLink, with BlueField DPUs, NVLink and Spectrum-X, which Nvidia says now assembles in about five minutes versus two hours for Grace Blackswellra.
It is a more immediate signal supply for the regional reader. Huang flew to Taiwan to meet with TSMC’s chairman Vera Rubin about production capacity, and the ramp-up is straining the Taiwanese supply chain. Combine that with a confirmed global memory shortage, and the reading is simple: consumer GPU and RAM prices are tightening instead of easing as the world’s most profitable customers adopt advanced wafers and DRAMs. Expect to see this push at retail – in graphics cards and memory kits.
Vera CPU goes after x86’s home turf
The most directly competitive move was the Vera CPU, billed as a “CPU for agents”: a monolithic 88-core Arm design with high per-core bandwidth and LPDDR5X memory built to overcome CPU bottlenecks that limit GPU usage. Huang’s argument is that CPUs have historically been built “for humans,” while agents run in nanoseconds and are impatient with tool calls — and the orchestration layer of agent AI is CPU-heavy territory owned by Intel and AMD. Nvidia claimed that SQL workloads ran nearly three times faster on Vera. Whether hyperscalers really displace x86 in scale is an open question; it’s not a foregone conclusion, but a valid opening shot.
RTX Spark: agent PC and the part you will actually buy
This is the consumer side of the strategy. The RTX Spark is built on the N1X superchip — a 20-core Arm Grace CPU co-designed with MediaTek and a Blackwell RTX GPU in a single TSMC 3nm class package, combined with Nvidia’s associated NVLink-C2C interconnect. It carries up to 128GB of unified LPDDR5X memory shared between the CPU and GPU, Nvidia claims a peak bandwidth of 600GB/s and 6,144 CUDA cores – the same fraction Nvidia uses as performance benchmarks for the desktop RTX 5070. The proposition for gamers and creators is compatibility: Huang claimed that the full Nvidia software stack and every application that Windows has ever run will run on the chip, and will launch when Adobe Photoshop and Premiere are launched.
Two caveats. First, the gaming claim is Nvidia alone. The company says the platform is good for “100 FPS 1440p gaming” based on DLSS upscaling and Multi-Frame Generation, and that the RTX Spark has no standalone, portable-laptop benchmarks since launch. We won’t make any judgments on real-world performance until PCMag Labs gets its hands on a unit. Second, the Arm angle is a total gamble: Nvidia is entering Windows-on-Arm territory as soon as Qualcomm’s exclusivity ends, and Arm software loopholes have sunk the idea before.
In the roadmap, Nvidia laid it out by GPU architecture rather than in a pure “N1X → N2X → N3X” sequence: Grace Blackwell (the N1X part) will launch in 2026, the Vera Rubin generation with LPDDR6 in 2027/28, and the Feynman generation in 203.
RTX Spark devices—more than 30 laptops and 10 desktops from Microsoft, Dell, HP, ASUS, Lenovo, and MSI—are expected this fall. Pricing is set by the OEM and not announced, so we don’t have anything concrete yet. For now, the only honest performance framework is the RTX 5070-class, so if you’re buying a gaming machine today, Our recommended gaming laptops in UAE and Saudi Arabiatogether with we test the current RTX 50 seriesremain safer reference points than the uncompared Arm part.
Physical artificial intelligence: the cars and robots that make it happen
The fifth driver extended the idea of an agent to the physical world. Along with platform and CPU news, Nvidia unveiled new models including the Alpamayo 2 for self-driving, and Huang repeatedly returned to humanoid robots and “smart cars” as the next surfaces for agent computing. The brief’s “skill file” concept—robots that download packaged skills that can be transferred rather than being reprogrammed for each task—fits Huang’s description of agents as portable workers, though it’s a guiding idea, not a shipping product.
Bottom line
Separate the stage and Computex 2026 was a single, connected bet: the agent is the new computing unit, and Nvidia intends to sell the silicon for it at every layer – in the rack, in the desktop, in the car and in the robot. Half of the data center is real and shipped; The Vera Rubin volume and serious Arm server CPU are concrete, and the supply strain behind them is the first part buyers will feel as upward pressure on GPU and memory prices.
The consumer half is the real bet. RTX Spark is the most reliable Windows-on-Arm effort yet — backed by a multi-year roadmap and real OEM commitment that previous efforts lacked — but it asks gamers to rely on Arm GPUs for first-party numbers without independent benchmarks and pricing. Our opinion is to watch the data center and wait for a tested RTX Spark unit before believing the 1440p story. The vision is Nvidia’s most ambitious yet; Whether your next PC will actually run on it is a question for consideration this fall.
