Last month’s Nvidia GPU Technology Conference (GTC) set the tone with a three-minute opening film focused on a single concept: the token. The video portrayed tokens as the fundamental unit that transforms raw data into knowledge — linking scientific discovery, humanoid robots, healthcare breakthroughs and visions of a better future. However, the most striking moment came at the end, with the appearance of Starcloud, a start-up that designs and builds data centers for use in orbit.
Starcloud, backed by Nvidia through its Inception start-up program, launched its “Starcloud-1” satellite in November last year. It was equipped with an H100 graphics processing unit (GPU) with 80GB of memory to test in-orbit computing. Its bold proposition: Use the continuous solar energy of space and natural radiation-based heat dissipation to solve the severe power and cooling bottlenecks facing terrestrial artificial intelligence (AI) data centers.
The inclusion of Starcloud in the GTC introduction video, neatly packaged for a broad audience to intuitively grasp, was Nvidia’s way of signaling a strategic shift — that computing power is national power, and that computing itself is moving into space.
Nvidia founder and chief executive officer Jensen Huang (黃仁勳) later took the stage to deliver a two-hour keynote speech, the real highlights of which came in the final minutes.
One moment stood out: Huang sharing the stage with Olaf from the movie Frozen. On the surface, it was playful. In reality, it was deeply symbolic.
Olaf was no longer just an animated character, but a robot capable of interacting in the real world. The message was clear: AI is evolving beyond generating text, images and video on screens. It is entering the physical world, perceiving environments, understanding tasks and taking actions.
In short, AI is no longer just generative — it is becoming embodied.
The closing four-minute montage reinforced that shift. Humanoid robots played musical instruments. A digital avatar of Huang appeared alongside a quadruped robot dog. Scenes blended AI inference, AI factories, OpenClaw systems, agentic AI, open-source ecosystems and robotics into a single narrative thread.
What looked like performance art was a road map: AI is moving from generating answers to generating actions. Agentic AI would take over knowledge work. Robots would enter factories, transportation and logistics. Digital twins and machines would redefine the human–machine interface.
Those final minutes were not an ending; they were a synthesis of everything that came before, and a preview of what comes next.
Three concepts stood out from Huang’s keynote.
The first is tokens. Huang framed them as the fundamental unit of modern AI, effectively resetting the industry’s coordinates. Future competition would not simply hinge on model size, but on who can process tokens more efficiently and at lower cost, extending that capability into inference and AI factories.
Huang invoked Nvidia’s Compute Unified Device Architecture’s 20-year evolution as a “flywheel” of accelerated computing, underscoring the company’s ambition to dominate all layers of the AI stack, from chips and systems to software and applications.
The second is agentic AI. Platforms such as OpenClaw — and Nvidia’s NemoClaw agent that adds privacy and security controls to OpenClaw — signal a shift beyond chatbots toward autonomous digital workers, systems capable of planning, using tools and operating over long periods, and that would be subject to governance constraints.
The third is physical AI. From Cosmos and Isaac GR00T model families to robotaxis, industrial robots, edge infrastructure and orbital computing, Huang’s message was unmistakable: AI is transitioning from linguistic intelligence to physical intelligence.
For Taiwan, the shift carries particular significance.
Nvidia’s message at GTC was not just about incremental growth. The next wave of AI expansion would not be confined to areas where Taiwan already excels, such as semiconductors and AI server manufacturing. It would extend into sensors, edge computing, digital twins, industrial software, robotics integration and space-based data infrastructure.
Understanding that transition — from “generating content” to “driving the physical world” — is key to grasping the true meaning of Huang’s speech.
Notably, he announced Nvidia’s ambitions to expand into space. Systems such as “Nvidia Space-1 Vera Rubin” are envisioned to bring large-scale AI data center capabilities into orbit, extending accelerated computing beyond Earth.
For Taiwan, this is a strategic signal that cannot be ignored.
Maintaining leadership in terrestrial data centers is no longer enough. Taiwan must begin building capabilities in orbital thermal management, radiation-hardened certification and mission-grade system integration. By combining advanced cooling modules, radiation testing and space-qualified standards, Taiwan has the opportunity to move beyond being a component supplier and become a key player in the emerging infrastructure of space-based computing.
The future of AI would not just be written in code. It would be built — in factories, in cities and in orbit.
Liao Ming-hui is an assistant researcher at the Chung-Hua Institution for Economic Research.
Translated by Lin Lee-kai
In the event of a war with China, Taiwan has some surprisingly tough defenses that could make it as difficult to tackle as a porcupine: A shoreline dotted with swamps, rocks and concrete barriers; conscription for all adult men; highways and airports that are built to double as hardened combat facilities. This porcupine has a soft underbelly, though, and the war in Iran is exposing it: energy. About 39,000 ships dock at Taiwan’s ports each year, more than the 30,000 that transit the Strait of Hormuz. About one-fifth of their inbound tonnage is coal, oil, refined fuels and liquefied natural gas (LNG),
On Monday, the day before Chinese Nationalist Party (KMT) Chairwoman Cheng Li-wun (鄭麗文) departed on her visit to China, the party released a promotional video titled “Only with peace can we ‘lie flat’” to highlight its desire to have peace across the Taiwan Strait. However, its use of the expression “lie flat” (tang ping, 躺平) drew sarcastic comments, with critics saying it sounded as if the party was “bowing down” to the Chinese Communist Party (CCP). Amid the controversy over the opposition parties blocking proposed defense budgets, Cheng departed for China after receiving an invitation from the CCP, with a meeting with
Chinese Nationalist Party (KMT) Chairwoman Cheng Li-wun (鄭麗文) is leading a delegation to China through Sunday. She is expected to meet with Chinese President Xi Jinping (習近平) in Beijing tomorrow. That date coincides with the anniversary of the signing of the Taiwan Relations Act (TRA), which marked a cornerstone of Taiwan-US relations. Staging their meeting on this date makes it clear that the Chinese Communist Party (CCP) intends to challenge the US and demonstrate its “authority” over Taiwan. Since the US severed official diplomatic relations with Taiwan in 1979, it has relied on the TRA as a legal basis for all
To counter the CCP’s escalating threats, Taiwan must build a national consensus and demonstrate the capability and the will to fight. The Chinese Communist Party (CCP) often leans on a seductive mantra to soften its threats, such as “Chinese do not kill Chinese.” The slogan is designed to frame territorial conquest (annexation) as a domestic family matter. A look at the historical ledger reveals a different truth. For the CCP, being labeled “family” has never been a guarantee of safety; it has been the primary prerequisite for state-sanctioned slaughter. From the forced starvation of 150,000 civilians at the Siege of Changchun
RSS