Nvidia used its Computex 2026 keynote to introduce Blackwell Ultra, a new AI processor the company says is built to significantly speed up inference workloads. Chief executive Jensen Huang described the chip as a major step forward for customers running large-scale AI systems, claiming performance gains of up to four times over the H200 generation.

The announcement underscores Nvidia’s effort to extend its lead in the fast-moving AI hardware market, where cloud providers, chipmakers and enterprise customers are racing to secure more efficient computing power. Inference, the phase where AI models generate outputs after training, has become a key battleground as demand rises for faster and cheaper deployment.

Reuters reported that Nvidia positioned Blackwell Ultra as part of its broader roadmap for next-generation AI infrastructure. While the company’s performance claims were prominent in the keynote, independent verification of the chip’s real-world gains will depend on third-party testing and customer deployments in the months ahead.

The launch also reflects continued investor and industry interest in Nvidia’s product cycle, with each new generation closely watched for signs of whether the company can maintain its dominance amid intensifying competition and surging demand for AI accelerators.

Computex 2026

Nvidia launches Blackwell Ultra AI chip, says it delivers up to 4x faster inference