• Nvidia unveiled NVLink Fusion at Computex 2025, a new chip-to-chip interconnect aimed at enhancing AI performance and supporting integration with third-party hardware.
• The technology enables more open, flexible AI systems, attracting partners like Marvell, MediaTek, Cadence, and Synopsys to support and adopt the ecosystem.
• Nvidia also announced a new Taipei headquarters and an AI supercomputer project in partnership with Foxconn and the Taiwanese government, powered by 10,000 Blackwell chips.

At the Computex 2025 tech convention in Taipei, Nvidia revealed NVLink Fusion, a major breakthrough in chip-to-chip interconnect technology. The technology is designed to boost the performance of artificial intelligence (AI) systems by making it easier and faster for several chips to communicate with each other. In contrast to its earlier versions that were used mainly in Nvidia's own hardware environment, NVLink Fusion is intended to be more open and adaptable so that it can be integrated with chips from other companies. This action represents Nvidia's strategic transition to a more cooperative stance in the hardware for AI market.

NVLink Fusion is an extension of NVLink-C2C, Nvidia's interconnect technology for chip-to-chip communication that provides high bandwidth and low latency. With this technology extended, NVLink Fusion facilitates the production of semi-custom systems in which Nvidia's GPUs can be paired with CPUs and AI accelerators from third parties. This kind of flexibility is specially useful for massive deployments of AI, where multiple hardware components must operate perfectly in tandem.

A number of leading technology firms have already shown interest in implementing NVLink Fusion. Some of the partners like Marvell Technology and MediaTek are anticipated to incorporate this technology into their chip architectures, making it easier to create higher-performance and higher-efficiency AI systems. Software companies like Cadence Design Systems and Synopsys are also working with Nvidia in order to assist the NVLink Fusion ecosystem, making the tools and frameworks available for mainstream adoption.

The release of NVLink Fusion also follows Nvidia's overall strategy for developing AI hardware. At the conference, CEO Jensen Huang emphasized future products, such as Blackwell Ultra chips available later this year, and future architecture like Rubin and Feynman due by 2028. These announcements reflect Nvidia's focus on improving AI technology and sustaining its dominance in the field.

Aside from the advancement in technology, Nvidia also revealed plans to build a new headquarters in northern Taipei, further entrenching its presence in the country. The move indicates the company's acknowledgment of Taiwan's pivotal position in the international electronics and semiconductor sectors. Furthermore, Nvidia is teaming up with local players such as Foxconn's Big Innovation Company and the government of Taiwan to construct an AI supercomputer fueled by 10,000 of its Blackwell chips. This initiative aims to bolster AI research and development, benefiting companies like Taiwan Semiconductor Manufacturing Company (TSMC).

Overall, the introduction of NVLink Fusion marks a turning point for Nvidia and the wider AI community. Through the promotion of interoperability and partnership-building among the technology community, Nvidia is enabling the creation of more adaptable and capable AI systems. With AI increasingly extending into every area of society, innovations such as NVLink Fusion will play a decisive part in shaping the computing future.

Read More News :

India Aims to become Leadership In 6G Technology

India Launches First Cluster of CCU Test beds