Nvidia has unveiled Blackwell Ultra, a new AI chip architecture it says delivers up to 30% better training performance than the previous generation while using less power. The announcement was made at Computex, where chipmakers are racing to meet surging demand for artificial intelligence hardware.

According to the company, the updated design is aimed at speeding up model training for data centers and large-scale AI workloads. Lower energy use could also help operators reduce cooling and electricity costs as they expand capacity.

The launch adds to Nvidia’s push to maintain its lead in the AI chip market, which has become one of the most competitive segments in technology. Rival firms are also investing heavily in processors built for AI acceleration, but Nvidia remains the benchmark for many cloud and enterprise buyers.

The company did not disclose full pricing or broad availability details in the initial announcement, leaving customers and investors to watch for follow-up guidance on production timelines and deployment plans.