Nvidia is pushing far beyond the familiar limits of the GPU market. At Computex Taipei 2026, the company used its main stage to show that its next phase reaches into AI PCs, autonomous vehicles, and humanoid robotics as parts of one connected strategy.
That broader direction does not mean gaming is being left behind. Nvidia still treats the PC gamer market as a core priority, but it is now pairing that business with systems designed for AI agents, physical machines, and software that can move between digital and real-world tasks.
RTX Spark becomes the clearest sign of Nvidia’s new PC direction
One of the most notable announcements was RTX Spark, a superchip Nvidia says is meant to “reinvent the PC.” The chip is described as having up to 128 GB of memory and one petaflop of computing power, with AI at the center of its design.
Nvidia says systems built on RTX Spark will target AI agents and will be developed with Microsoft, with Windows as the main focus. Even so, the company still positions the chip for gamers, saying it can run modern AAA games at 1440p with ray tracing and DLSS at more than 100 fps.
DLSS 4.5 extends the company’s rendering push
Nvidia also gave DLSS 4.5 a major role in its presentation. The next step in its AI rendering stack is Ray Reconstruction, following DLSS 4’s Multi Frame Generation, which can generate up to five frames per rendered frame.
DLSS 4.5 Ray Reconstruction uses neural rendering on modern GeForce RTX GPUs to improve ray tracing quality. Nvidia says the result includes cleaner particle effects, less ghosting, and more accurate lighting through better spatial awareness.
The company says DLSS 4.5 will arrive in August 2026 for all RTX GPUs. At launch, only a small number of games will support it, but several upcoming titles are already scheduled to add compatibility.
Game engines and partners are being pulled into the same plan
Nvidia’s effort is not limited to hardware or visual features. The company says the team behind games such as Marvel Rivals is working to update its engine so it can use DLSS 4.5.
That signals a broader attempt to expand adoption across the PC gaming ecosystem. Nvidia appears to be aligning its hardware and software roadmap so that game developers can build around its AI rendering tools more directly.
Autonomous driving gets a new reasoning model
Away from gaming, Nvidia also showed Alpamayo 2 Super for autonomous vehicles. The model is an open 32-billion-parameter reasoning VLA framework that Nvidia says is designed for the infrastructure behind future robotaxis.
Nvidia describes Alpamayo 2 Super as capable of “humanlike perception, reasoning, and action.” It will connect with AlpaGym and AlpaSim to train autonomous vehicle AI, with the goal of reducing cascading errors during training.
Jensen Huang said manufacturers such as Nissan, Hyundai, and Mercedes-Benz will use Nvidia Drive Hyperion systems built with Alpamayo 2 Super. He also said around 80 percent of the world’s carmakers have signed up to build Nvidia Drive Hyperion vehicles.
Humanoid robotics and physical AI move closer to center stage
Nvidia’s robotics push was led by Isaac GR00T Reference Humanoid Robot, a reference design for humanoid systems. It uses a Unitree H2 Plus chassis, Sharpa Wave five-finger hands, and a Jetson Thor module running Isaac GR00T software.
The project is meant to support more open research in humanoid robotics while helping drive physical AI across industries. That makes the system part of Nvidia’s effort to build machines that can understand and interact with the physical world more directly.
Nvidia also introduced Cosmos 3 as what it called the world’s first fully open omnimodel. The model can translate text, images, video, and audio into actions in the physical world.
The company says Cosmos 3 is meant to address the challenge of training physical AI when data is limited and simulation stacks remain fragmented. It is also designed to help platforms understand object interaction, motion, space-time relationships, and trajectory prediction with physics support.
Several companies are already said to be joining efforts to train robots, autonomous vehicles, and other physical AI systems using the model.