5 of the coolest reveals from Nvidia’s Computex 2026 event

If you want to get a glimpse of the future of various technologies and industries, just check out exhibitions and conventions. If you’re looking for the latest IT technology, look no further than Computex Taipei. The expo has been going strong for several decades and to open the 2026 convention, Nvidia founder and CEO Jensen Huang took the stage as the keynote speaker.

If you play PC games, you probably know Nvidia for its graphics cards. The company sells incredible flagship GPUs, the RTX 5080 and 5090, but is also known for technologies like Deep Learning Super Sampling (DLSS) and G-Sync. The former uses AI to improve image quality without sacrificing frame rates, while the latter works in tandem with compatible displays to reduce input lag and image tearing. For two hours, Huang discussed Nvidia’s upcoming products, including new game rendering technologies and AI. Here are the reveals we think are the coolest.

Nvidia launched its RTX line of graphics cards in 2018. These GPUs were the first cards capable of delivering real-time ray tracing, a rendering technique that simulates the behavior of light for improved visual fidelity, to the general public (the “RT” in “RTX” stands for “Ray Tracing,” after all). And now the next evolution of the RTX lineup is upon us.

One of the highlights of Nvidia’s Computex 2026 event was the RTX Spark, a “superchip” that Jensen Huang said will “reinvent the PC.” This new chip will provide up to 128 GB of memory and a whole petaflop of computing power. AI will be at the forefront of chip design, and computers that use RTX Spark will be geared toward “AI agents.” Additionally, Nvidia is working closely with Microsoft to develop this component and its capabilities with Windows in mind. However, Nvidia isn’t about to abandon the demographic that made the company a household name: PC gamers.

Ultimately, the RTX Spark will bring more powerful RTX graphics, which should satisfy most PC owners. Huang promises that the RTX Spark will power modern AAA games at 1440p resolution, with ray tracing and DLSS, all at over 100fps. And as Nvidia and Microsoft collaborate on the chip, the Xbox brand will benefit from a backlash. According to Xbox’s VP of Next Gen, Jason Ronald (yes, that’s his title), the RTX Spark will offer more access to Xbox titles on computers using the superchip.

Nvidia’s DLSS has been a key feature of its RTX GPU lineup since September 2018, and the company has been iterating on the system ever since. The latest version, DLSS 4, introduced multi-frame generation, which uses AI to “increase the frame rate” and “generate up to five frames per rendered frame.” Although DLSS 5 is nowhere to be seen, Nvidia is ready to move forward with DLSS 4.5.

Nvidia announced DLSS 4.5 Super Resolution at CES 2026, but Jensen Huang used Computex 2026 to announce the next big feature in DLSS 4.5: ray reconstruction. The technology takes the neural rendering capabilities of all modern GeForce RTX GPUs and applies them to ray tracing. According to videos released by Nvidia, DLSS 4.5 will provide sharper particle effects with less ghosting on titles such as “Alan Wake 2” and “Pragmata.”

DLSS 4.5 would bring much more than just improved ray tracing. The update will also boost computing capabilities across the board, as well as better spatial awareness that will allegedly improve lighting accuracy. Nvidia GPU owners can expect DLSS 4.5 to roll out in August 2026 for all RTX GPUs. Although only a handful of titles will use DLSS 4.5 at launch, many upcoming games are expected to release with DLSS 4.5 support, and the teams behind games such as “Marvel Rivals” are working with Nvidia to upgrade their engines to include DLSS 4.5 features.

Several Silicon Valley companies are developing a kind of robotaxi, including Nvidia. While the company is most often associated with PC gaming, Nvidia plans to soon compete with Elon Musk’s Tesla through autonomous vehicle technology. And Jensen Huang presented the next phase of these plans at Computex 2026.

Instead of building its own cars, Nvidia will provide the tools needed to make autonomous vehicles safer and more capable. One such tool will be the recently announced Alpamayo 2 Super, a “32 billion parameter open reasoning VLA model” for robotaxis that will serve as a framework for future autonomous vehicle infrastructures. According to Nvidia, the Alpamayo 2 Super will offer “human perception, reasoning and action” so that cars can reason as effectively and quickly as many human drivers.

Alpamayo 2 Super will also tie into another upcoming AI system, AlpaGym. While Alpamayo will run on robotaxis, AlpaGym will help train its AI models using “AlpaSim”. This program continually executes various actions and decisions through a model that determines all the ways they can change the local environment, potentially reducing the cumulative errors of competing training models. Jensen Huang said manufacturers including Nissan, Hyundai and Mercedes-Benz will install models of Nvidia’s Drive Hyperion – the company’s autonomous vehicle platform – made with Alpamayo 2 Super. According to his calculations, the manufacturers of about 80% of the world’s cars are committed to building Nvidia Drive Hyperion cars.

What’s cooler than a self-driving car? And why not an autonomous robot? Yes, there are fears that future robot companions will actually be piloted by aliens, but Nvidia has a potential solution.

At the Nvidia event, Jensen Huang announced the Nvidia Isaac GR00T benchmark humanoid robot. This is not a solo Nvidia project, but a true concert of technologies working in harmony. The robot takes a Unitree H2 Plus humanoid robot chassis, gives it Sharpa Wave five-finger touchscreen hands, and installs the Nvidia Jetson Thor module loaded with Nvidia Isaac GR00T software.

True to its name, the Isaac GR00T humanoid reference robot is not intended to be its own robot, but an “open humanoid robot reference design” intended to help “democratize research in cutting-edge humanoid robotics.” Nvidia wants the Android to help bring “physical AI” (AI designed to interact with the physical world) to various industries and “make groundbreaking discoveries in general-purpose physical intelligence,” all without relying on proprietary technology. Imagine a future where a lunar colony exists and is run by robots trained using Isaac GR00T. The less oxygen its workers need (because they’re robots), the more lunar dust can be turned into rocket fuel.

As you may have gathered, AI was a recurring topic among Nvidia’s announcements at Computex 2026, in keeping with the exhibition’s theme for the year, “AI Together.” Although Nvidia’s AlpaGym is designed to train robotaxis AI models, it’s not the company’s only AI intended to help autonomous devices understand the physical world.

Announced at Computex 2026, Nvidia Cosmos 3 is the “world’s first fully open omnimodel” that natively translates text, images, video and sounds into actions in the physical world. The modality (type of data) doesn’t matter, because Cosmos 3 can “generate and reason with it.”

So what’s so impressive about Cosmos 3? According to Nvidia, the model is designed to address what the company calls a “fundamental challenge” of training physical AI: helping diverse platforms “generalize to the real world with limited training data and fragmented simulation stacks.” The model combines reasoning with generation transformers to help platforms understand how objects interact, move, and interact with each other in space-time. And that’s before you start using physics to predict trajectories. Several companies have already signed up to train robots, autonomous vehicles and everything else on Cosmos 3, and are also using it to develop even more advanced physical AI models.