본문 바로가기
bar_progress

Text Size

Close

[Taiwan Chip News] "Nvidia Develops AI SSD... A Major Opportunity for Taiwan"

Nvidia and SK hynix Begin AI SSD Development
Strategy to Address AI Bottlenecks and Alleviate DRAM Supply Constraints
Phison’s Participation Presents “New Opportunities” for Taiwan’s Memory Industry

As the proliferation of artificial intelligence (AI) drives a surge in memory demand, Nvidia has partnered with SK hynix to embark on the development of a new AI solid-state drive (SSD). With Taiwan-based SSD controller company Phison joining the collaborative development, expectations are rising within Taiwan’s memory industry.


According to local media outlets such as Taiwan Economic Daily News, Phison will participate in Nvidia-led AI SSD development. Until now, Taiwan has not had a significant presence in AI memory segments such as Double Data Rate 5 (DDR5) and High Bandwidth Memory (HBM). However, with Phison joining this Nvidia-led AI SSD initiative, for the first time a Taiwanese company will be involved in the process of establishing Nvidia’s new standards.


[Taiwan Chip News] "Nvidia Develops AI SSD... A Major Opportunity for Taiwan" The E28 SSD controller from Phison, released in June this year. Phison

This new AI SSD is expected to deliver up to 10 times the performance of existing SSDs currently installed in AI servers, and is seen as Nvidia’s attempt to lead the establishment of new memory and storage standards. Phison will play a key role in building a “quasi-memory hierarchy” between memory and storage in AI servers, aiming to address future storage demands in space data centers as well. Industry insiders note that Phison already possesses solutions aligned with AI SSD architecture, and view this as an extension of the company’s longstanding achievements in the enterprise and high-performance storage markets.


The internal code name for the AI SSD project led by Nvidia and SK hynix is reportedly “Storage Next.” The goal is to boost performance by about 10 times compared to existing enterprise SSDs, targeting up to 100 million IOPS (input/output operations per second). Optimized for AI inference workloads, the SSD is expected to have a prototype unveiled as early as next year, and will be gradually adopted in Nvidia AI servers. Nvidia is also restructuring its AI computing architecture, including integrating Graphics Double Data Rate 7 (GDDR7) memory into its next-generation “Rubin” platform. The development of AI SSDs is seen as an effort to extend this strategy into the storage domain.


[Taiwan Chip News] "Nvidia Develops AI SSD... A Major Opportunity for Taiwan"

The semiconductor industry notes that, given the limited supply of HBM and DRAM-which are essential for AI computing-the latency and IOPS performance of conventional SSDs are insufficient to support real-time AI inference. Nvidia’s AI SSD development is seen as a strategy to resolve performance bottlenecks while also alleviating the burden of DRAM supply constraints.


Some analysts believe Nvidia’s vision targets not only terrestrial data centers but also, in the long term, highly reliable environments such as space data centers. Phison’s storage solutions have previously been validated in space environments through SpaceX missions, and this technical capability and reliability are cited as key reasons for its selection.


In the Taiwanese market, the high durability and reliability required by AI SSDs are expected to further strengthen Phison’s competitiveness in the next-generation AI storage market.


Taiwan Economic Daily News = Reporter Lee Mengshan / Translation = The Asia Business Daily


*This column is published as part of a strategic partnership between The Asia Business Daily and Taiwan Economic Daily News.


© The Asia Business Daily(www.asiae.co.kr). All rights reserved.

Special Coverage


Join us on social!

Top