President Donald Trump signaled reluctance to share Nvidia’s Blackwell chips with China. Following a meeting in South Korea on Thursday, Trump told reporters aboard Air Force One that semiconductors had been discussed and that China would be “talking to Nvidia and others about taking chips,” but added, “We’re not talking about the Blackwell.”
Nvidia’s Blackwell architecture, announced in 2024 and rolling out through 2025, represents a significant advancement in GPU technology, particularly for AI and large-scale machine learning workloads. Named after mathematician David Blackwell, the architecture succeeds Hopper and introduces innovations such as the second-generation Transformer Engine, multi-die “superchip” configurations, and high-bandwidth interconnects.
READ: Trump signs executive order allowing TikTok US deal (
Flagship models like the B200 and GB200 are designed to accelerate LLM training and inference, with Nvidia claiming performance gains of up to 30× over previous GPUs in certain AI-specific workloads, though results vary by model size, task, and configuration. Blackwell also aims for improved energy efficiency, again depending on workload type. The architecture addresses the increasing demands of generative AI, enabling larger models and faster computations while bridging enterprise deployment and research environments. Its rollout in 2025 remains gradual, reflecting supply constraints and selective adoption among large-scale AI users.
Nvidia CEO Jensen Huang said on Thursday he was confident that President Trump and Chinese leader Xi Jinping had a good conversation during a much-anticipated meeting in South Korea earlier in the day.
“I have every confidence that the two presidents had a very good conversation. It doesn’t have to involve anything that I do,” said Huang.
The U.S. government strengthened export controls on advanced semiconductors, including GPUs, to limit China’s access to cutting-edge AI capabilities for commercial and military use.
The Bureau of Industry and Security (BIS) issued updated rules requiring broader licensing for high-performance chips destined for China, and emphasizing national security concerns. These measures specifically target processors capable of accelerating AI and machine learning workloads, effectively restricting the most advanced hardware while allowing limited, regulated exports. The controls reflect the U.S.’s strategic goal of maintaining technological leadership in AI and high-performance computing while managing geopolitical risks.
Amid these restrictions, Nvidia has publicly acknowledged the potential to bring its Blackwell-architecture GPUs to China under U.S. government approval. Huang indicated that any China-specific deployment would comply with export regulations, potentially involving performance-limited versions. This scenario illustrates the tension between commercial opportunities and regulatory constraints, highlighting how major tech firms must navigate U.S.-China geopolitical dynamics while supporting global AI innovation. It underscores the broader interplay of technology, trade policy, and national security in 2025.
For companies like Nvidia, navigating these dynamics requires balancing commercial opportunities with strict regulatory compliance, ensuring that technology deployment aligns with both government policy and international market considerations.