NVIDIA’s GTC Berlin Keynote Frames Tokens as a Path From Data to Action

The keynote presents tokens as “the building blocks of AI” and describes their role as a sequence: data becomes knowledge, and knowledge becomes action. Its examples make that sequence concrete in two ways. Some put information into view—organizing scans, forecasting crowds, labeling anatomy. Others place machines in physical settings where they can carry, inspect, build, or move.

A radiation-treatment diagram offers the clearest picture of information being assembled for a decision. It brings together patient scans, a dose prescription, clinical history, positioning and motion, prior radiation dose, target and organ contours, organ-dose limits, and beam parameters. The listed output is a radiation treatment plan. Rather than depicting a single input producing a result, the diagram frames planning as the coordination of several kinds of patient and treatment information.

A surgical display makes a related idea visible at the point of care. Over an endoscopic view of the brain, it labels critical structures and assigns confidence values: 96% for the sella, 81% for the carotid arteries, 92% for the clival recess, and 61% for the optic protuberances. The narration says that “the steadiest hands never work alone.” The image gives that line a specific setting: a surgical view accompanied by labeled structures and confidence estimates. The footage is credited to The Lancet Group, and the display names UCL Hawkes Institute.

Other interfaces apply the same visual logic to environments and movement. At a train station, a request to show crowd density on Platform 1 at 5:38 p.m. is followed by a forecast of high crowd density at that time. A vehicle display combines multiple camera feeds with a LiDAR bird’s-eye view. In a robotics setup, a digital twin of a robotic arm appears beside controls for synthetic data generation and task configuration. Across these scenes, information about people, places, or equipment is rendered in a form that can be inspected and used.

The keynote also places AI alongside physical work. A genome rendering labeled “EDEN Foundation Model” shows DNA inserted at a target site, with Basecamp Research named on screen. The narration describes tokens as deciphering nature’s code to reveal hidden cures. The image connects that ambition to a specific depiction of genome editing.

Medical delivery is shown not as an abstract network but as a sequence of transport tasks. A drone carries a medical payload over a city beneath an “Urgent Pathology” overlay listing 36 patient samples and an ETA of 1:37. A robotic dog carries a cooler up stairs; its overlay reads “Delivery to Lab,” also lists 36 patient samples, and gives an ETA of 0:24. NHS and Apian are named in the second scene. The narration’s phrase “deliver the care we need” sits alongside these examples of moving samples through different environments.

A different robotic dog enters a dark industrial facility, where an atmospheric scan reports hydrogen sulfide at 12 PPM. Here, the narration frames the machine as walking into danger to protect people. The sequence also shows a person using a brain-computer interface cap to control an exoskeleton hand, with a camera view identifying a dining table and displaying prompts to pick up a drink and move a bowl. Elsewhere, motion-tracking gloves control robotic hands that mirror their movements; an orange exoskeleton glove and a robotic prosthetic hand appear in separate scenes.

Taken together, the examples extend the idea of assistance beyond displaying information. Machines are shown carrying samples, entering hazardous spaces, and participating in tasks involving movement and grasping. The narration links these scenes to shared success and to putting “what matters within reach.”

The keynote’s claim of work on “a scale never seen before” is accompanied by scenes of both physical construction and computing infrastructure. A view from an excavator cab includes a tablet interface; another scene shows a snowy construction site in a mountainous region. The sequence then moves to a data center labeled JUPITER Germany and a server aisle labeled Bull France. The examples place the token-and-AI framing alongside equipment, worksites, and facilities that support large-scale computing and industrial activity.

The final scenes take that framing into orbit. Workers assemble satellite components under the EnduroSat name, followed by rocket-launch footage credited to Arianespace. An orbital satellite appears with two figures on screen:

The same overlay lists satellite power at 3.4 kW. The figures describe different properties—compute and power—and the imagery connects them to the satellite rather than treating them as interchangeable measures.

The closing promise—“Together, we take the next great leap into a bright new future, built for all mankind”—gathers the examples under a shared ambition. The final line, “And here is where it all begins,” returns to the opening proposition: tokens connect data with knowledge and action, whether the scene is a treatment plan, a delivery route, an industrial inspection, or computing in orbit.