NVIDIA CEO Defends China Access, Unveils Enterprise AI
Taipei, Taiwan - NVIDIA Corp Chief Executive Jensen Huang defended his company's market access in China.
It said global artificial intelligence infrastructure development remains in early stages, as the chipmaker introduced new enterprise computing systems at its Taiwan technology conference.
Huang said NVIDIA's market share in China has declined to approximately 50 percent from 95 percent four years ago due to U.S. export restrictions.
The company has written off billions of dollars in H20 chip inventory following recent trade controls.
He characterized the export restrictions as an ineffective policy that has strengthened local Chinese competitors rather than limiting AI development.
The company unveiled enterprise AI computing systems during the NVIDIA GTC Taiwan event, including an RTX Pro server featuring eight graphics processing units connected through 800-gigabit-per-second NVLink technology.
The air-cooled system operates on x86 architecture and supports existing enterprise software, including Red Hat, VMware, and Nutanix platforms.
NVIDIA also introduced the DGX Spark personal AI development system, which Huang said provides individual developers with supercomputing capabilities equivalent to systems that required massive data centers five years ago.
According to company specifications, the desktop unit can run AI models with up to one trillion parameters and offers developers an approximately six-month return on investment.
Microsoft Corp announced that it would integrate its Windows ML platform with NVIDIA's RTX technology, extending AI capabilities to millions of existing personal computers equipped with NVIDIA graphics processors.
The collaboration enables Windows-based artificial intelligence development using NVIDIA's CUDA computing platform and Tensor processing cores.
The semiconductor company revealed NVLink Fusion technology, which opens its proprietary chip interconnect system to hardware partners, including Qualcomm Inc. and Fujitsu Ltd.
The technology allows different processor architectures to integrate with NVIDIA's networking infrastructure while maintaining compatibility with existing data center configurations.
Huang projected that reasoning-based AI models requiring multiple computational cycles will increase processing demands by 100 to 1,000 times compared to current single-response systems.
He said the transition from basic AI responses to models that analyze, plan, and research represents a fundamental shift in computational requirements.
The executive emphasized China's importance as a market representing approximately 50 percent of global AI researchers and potentially $50 billion in annual opportunity.
He cited recent developments, including the reasoning AI models of Chinese company DeepSeek, as evidence of continued regional innovation despite trade restrictions.
NVIDIA demonstrated advanced semiconductor packaging technology in its latest processors. Huang stated that the physical limitations of Moore's Law make multi-chip designs essential for future AI systems.
The company's processors utilize Taiwan Semiconductor Manufacturing Co. Ltd.'s Chip-on-Wafer-on-Substrate packaging technology, which Huang described as having no current viable alternatives for advanced AI applications.
The company outlined its Isaac Groot robotics platform for humanoid robot development.
Huang predicted that the sector would become a trillion-dollar industry due to technological convergence and broad commercial applications.
He compared the potential trajectory to smartphone industry development when multiple technologies reached simultaneous maturity.
Huang said AI infrastructure deployment will require gigawatt-scale power consumption equivalent to a nuclear power plant's output, noting that countries with excess energy capacity are positioning to export AI services rather than raw energy resources.
The executive projected a trillion-dollar investment in AI infrastructure development over the next decade.
The chipmaker is expanding office facilities in Shanghai to accommodate growing operations.
Huang clarified that NVIDIA has maintained Chinese operations for three decades and requires additional space for existing employees rather than establishing new strategic initiatives.