NVIDIA has introduced its Blackwell Ultra GPU architecture at Computex 2026, positioning the chip as a new option for large-scale artificial intelligence training. The company says the new design delivers up to 30% better performance per watt than the previous generation for AI workloads.

The launch extends NVIDIA’s push to maintain its lead in the rapidly expanding AI hardware market. Energy efficiency has become a key selling point as companies and cloud providers look for faster systems that can handle growing demand without sharply increasing power costs.

NVIDIA did not release every technical detail in the initial announcement, but the company framed Blackwell Ultra as an upgrade aimed at data centers and enterprise customers building and running advanced AI models. The debut comes as competition intensifies across the semiconductor industry over chips tailored for artificial intelligence.

The announcement at Computex underscores how central AI infrastructure has become to the technology sector. With firms racing to secure computing capacity, performance gains measured against power consumption are likely to remain a major factor in future chip sales.