Hwang In-hoon explains DSX AI factory economics: computing power is revenue. By the end of this decade, the world will launch 100GW of AI computing power

Comparative news, according to monitoring, Huang Renxun systematically introduced Nvidia DSX (the third largest product line after RTX and DGX) during his speech at GTC Taipei 2026, positioning it as an end-to-end reference design and operation platform for AI factories. He said that Nvidia has evolved from a GPU company to a systems company and is now once again transforming into an AI infrastructure company: customers don't want to buy computers; they want to build AI factories. The DSX consists of four major modules. DSX SIM is based on Omniverse digital twins, and completed layout planning, power cooling simulation and network verification of the entire factory before the first rack is landed; DSX OS is responsible for automatic infrastructure configuration, operation monitoring and fault repair, transforming the system into multi-tenant, highly available AI-ready computing power; DSX MAX LPS solves the 40% power overallocation problem common in AI factories today, and deploys more GPUs within the same power budget through dynamic power allocation between racks and peak current smoothing; DSX Flex reads grid signals in real time and dynamically adjusts factory electricity consumption when the grid needs mitigation, making the AI factory a flexible energy asset for the power grid. Additionally, Nvidia's ready 45°C high-temperature liquid cooling technology can eliminate traditional chillers and drastically reduce water and energy consumption. Hwang In-hoon gave a clear economic account: every token can be profitable, computing power is revenue, and each watt of performance is your revenue. The cost of a single 1GW AI factory has climbed from $20 to $30 billion to $50 to $60 billion, and will soon reach $80 to $100 billion. By the end of this decade, a 100GW AI factory will be launched around the world, which he called the largest infrastructure construction in human history. He also showcased a number of emerging AI cloud companies built on Nvidia's full stack: CoreWeave (estimated at $50-70 billion and growing rapidly), Nebius, N Scale (customers include British Telecom and Google), India's Yoda, Singapore AI Singapore, Indonesia's IndoSat, and Taiwan's GMI, etc., stressing that AI will be implemented in every region, and every company will be driven by AI.