Kioxia and NVIDIA Collaborate to Develop AI-Optimized SSDs
2025-09-12
In the current era of rapid development in artificial intelligence (AI) technology, the efficiency of data storage and processing has become one of the key factors constraining technological progress. Recently, Kioxia, a leading storage company, announced a collaboration with NVIDIA to jointly develop a new-generation solid-state drive (SSD) specifically designed for AI servers. This collaboration is expected to be commercialized around 2027.
Background and Objectives of the Collaboration
As the scale of AI models continues to expand, the demand for storage capacity and read-write speeds is also increasing. Traditional SSDs are gradually becoming a bottleneck in data transfer speeds, especially in AI servers that require rapid processing of large amounts of data. To address this issue, the collaboration between Kioxia and NVIDIA was initiated. The objective of this collaboration is to develop a new-generation AI-optimized SSD with a random read performance of 100 million IOPS (input/output operations per second), representing a 100-fold increase in read speed compared to traditional SSDs. This leap in performance will significantly accelerate the data processing speed of GPUs, thereby enhancing the efficiency of AI model training and inference.
Technical Highlights and Innovations
The new-generation SSD will adopt a GPU-direct architecture, bypassing the traditional CPU and system memory to directly exchange data with the GPU. This innovative architectural design dramatically reduces data transfer latency and significantly improves the efficiency of data flow between the GPU and the SSD. In this way, the new-generation SSD can provide near-real-time data access for GPUs, greatly reducing the training time of AI models and increasing inference response speed.
The new-generation SSD will also support the latest PCIe 7.0 standard. The adoption of this standard will provide a solid foundation for high-speed data transfer and further enhance the performance of the SSD. Compared to traditional PCIe standards, PCIe 7.0 can achieve higher data transfer rates, meeting the stringent storage performance requirements of AI servers.
NVIDIA has set a higher performance target for the new-generation SSD-reaching 200 million IOPS. To achieve this goal, Kioxia plans to use two SSDs working in tandem. This collaborative working mode not only enhances the overall read speed but also provides greater flexibility and reliability in data transfer.
Market Prospects and Industry Impact
With the continuous development of AI technology, the demand for high-performance storage solutions is growing rapidly. Kioxia estimates that by 2029, AI-related industries will account for nearly half of the demand for NAND flash memory. This trend indicates that high-performance storage technology will be a key support for the future development of AI. The collaboration between Kioxia and NVIDIA is aimed at meeting this growing market demand and providing more cost-effective and flexible storage solutions for the AI industry.
This collaboration not only marks a significant transformation in the storage architecture of AI servers but also heralds a new revolution in storage technology in the AI field. The introduction of the new-generation SSD will provide strong support for the development and popularization of generative AI technologies, driving the entire industry towards higher performance and greater efficiency. For enterprises and individuals pursuing high-performance computing, this technological advancement is undoubtedly an important milestone.
Semicon Integrated Circuit Distributor
Semicon is a company dedicated to the distribution of high-performance and high-reliability integrated circuit (IC) products, providing extensive services in fields such as industrial control, communication equipment, consumer electronics, and automotive electronics. Semicon has a rich ic inventory, covering various types of IC products such as analog, digital and storage, which can meet the needs of different customers. The following are several selected popular IC chips.
● 1N4148WT: A classic high-speed switching diode, suitable for rectification, protection and other functions in various circuits, featuring low reverse leakage current and fast recovery time.
● AD7177-2BRUZ: A 24-bit analog-to-digital converter with high precision and low noise, specifically designed for high-precision measurement applications, it can provide outstanding signal processing performance.
● TPSM265R1SILR: Integrated power module, featuring high-performance conversion and compact packaging, is suitable for equipment with limited space and high requirements for power efficiency.
● MPQ8626GD-Z: A high-performance power management chip that offers multiple power conversion functions and is suitable for the power management requirements of various electronic devices.
Email: info@semicone.com
Japan Grants Micron's Hiroshima DRAM Factory Up to ¥536 Billion in Subsidies
South Korea Aims for 20% SiC Semiconductor Self-Sufficiency by 2030
Related Article
Broadcom is finalizing a $60B+ debt deal for AI chip funding, with subordinated tranches pushing totals near $100B.
Broadcom Plans to Raise Up to $100 Billion in Debt for AI Chip Financing
STMicroelectronics enacts third 2026 price hike (Aug 23), citing rising supply‑chain costs. Auto MCUs up 15‑20%, lead times stretch to 52 weeks.
ST Implements Third Price Hike This Year, Effective August 23