Huang Renxun Expounds on DSX AI Factory Economics: "Compute Power is Income," 100GW of AI Compute Power to Come Online Globally by the End of this Decade

According to Omdia research, Huang Renxun systematically introduced NVIDIA DSX during his keynote at GTC Taipei 2026, positioning it as the third major product line of NVIDIA after RTX and DGX, as an end-to-end reference design and operation platform for AI factories. He stated that NVIDIA has evolved from a GPU company to a systems company and is now transforming again into an AI infrastructure company: "Customers don't want to buy a computer; they want to build an AI factory."

DSX consists of four main modules. DSX SIM, based on Omniverse digital twin technology, completed the layout planning, power and cooling simulation, and network validation for the entire factory before the first rack was installed; DSX OS is responsible for automatic infrastructure configuration, operational monitoring, and fault recovery, transforming the system into a multi-tenant, high-availability AI-ready computing power; DSX MAX LPS addresses the common 40% power over-provisioning issue in current AI factories by deploying more GPUs within the same power budget through dynamic power allocation between racks and peak current smoothing; DSX Flex dynamically adjusts factory power consumption when the grid needs relief, making the AI factory a "flexible grid asset." In addition, NVIDIA's ready-to-use 45°C high-temperature liquid cooling technology eliminates the need for traditional chiller plants, significantly reducing water and energy consumption.

Huang Renxun provided a clear economic rationale: "Every token can be profitable; computing power is income, and performance per watt is your revenue." The cost of a single 1GW-level AI factory has risen from $200-300 billion to $500-600 billion and will soon reach $800-1,000 billion. By the end of this decade, there will be 100GW of AI factories online worldwide, which he termed "the largest-scale infrastructure development in human history."

He also showcased several emerging AI cloud companies built on the NVIDIA stack: CoreWeave (valued at $500-700 billion and rapidly growing), Nebius, N Scale (with customers such as British Telecom and Google), Yoda from India, AI Singapore from Singapore, IndoSat from Indonesia, and GMI from Taiwan, emphasizing that "AI will land in every region, and every company will be AI-driven."