MaiStorage aiDAPTIV+ & Enterprise SSD Solutions

MaiStorage (Maistorage Technology Sdn Bhd) was established in June 2024 in Puchong, Selangor, as the Malaysian subsidiary of Taiwan's Phison Electronics, a NAND controller supplier founded in 2000. It aims to build NAND storage technology capability in Malaysia, designing, making and delivering NAND controllers, ICs and storage modules for data centres, AI and automotive, starting with Phison's Pascari enterprise SSDs, aiDAPTIV+ and automotive products.

aiDAPTIV+ is a turnkey LLM fine-tuning solution that extends GPU memory with flash. An aiDAPTIVCache SSD with SLC NAND and 100 DWPD endurance holds model slices, while middleware slices the model, assigns slices to each GPU and swaps them as work finishes. The aiDAPTIVPro Suite GUI and a CLI cover model ingest, fine-tuning, validation and inference, and the solution works as a drop-in with PyTorch on PCs, workstations, servers and NVIDIA Jetson devices.

• aiDAPTIVCache SSD: SLC NAND drive with 100 DWPD endurance that extends GPU memory.
• Middleware: slices models and swaps slices between GPU and flash during training.
• aiDAPTIVPro Suite: GUI and CLI for ingest, fine-tuning, validation and inference.
• PyTorch drop-in: runs existing applications without code changes.
• Broad hardware support: PCs, notebooks, servers and NVIDIA Jetson edge devices from multiple OEMs.
• Enterprise SSDs: X, D, B and S series storage for data centres.

A Vietnamese bank or hospital can fine-tune a Vietnamese-language LLM on internal documents using workstations instead of a large GPU cluster. A university can equip AI training PCs for LLM courses. A robotics or edge AI company can run on-device training on NVIDIA Jetson systems.

Organisations can fine-tune larger LLMs on local hardware, with GPU memory extended by up to 320GB on PCs and up to 8TB on workstations and servers. Two 2TB aiDAPTIVCache drives support a 70B-parameter model. Offloading memory to flash reduces the number of costly GPUs required. Training stays behind the firewall, supporting data sovereignty.

Enterprises and public agencies want to fine-tune LLMs on private data without sending it to the cloud, but GPU memory is scarce and expensive. Malaysia and Vietnam are both promoting sovereign AI and local language models, which raises demand for affordable on-premises AI training hardware.

Established in June 2024 in Puchong, Selangor, Malaysia, as a subsidiary of Phison Electronics (Taiwan, founded 2000); Phison announced an RM1 billion investment in AI storage in Malaysia in July 2024. aiDAPTIV+ is Phison technology sold and supported by MaiStorage in Malaysia.