Phison is delivering this through a proprietary software layer and a new hyper-converged infrastructure (HCI) platform called the “AI Data Platform,” which is designed to let enterprises deploy on-premises AI with data sovereignty at a lower cost than cloud or GPU-heavy alternatives .

Pascari aiDAPTIV AI20EH: The Award-Winning AI PC Solution

The flagship product for this new vision is the Pascari aiDAPTIV AI20EH SSD, which won a Computex 2026 Best Choice Award in the “AI PC Solution” category on May 21, 2026 .

How aiDAPTIV Works

the aiDAPTIV technology uses NAND flash as a dedicated AI key-value (KV) cache tier. By adding its middleware, the system pools together GPU VRAM, system DRAM, and NAND flash into one dynamic memory resource. This removes the traditional local memory wall that prevents large models from running on standard hardware .

Claimed Performance Gains

According to Phison’s internal testing, adding the aiDAPTIV AI20EH drive and middleware to an AI PC delivers dramatic improvements :

• Up to 102x improvement in AI inference performance
• 67% reduction in DRAM usage
• Up to 53% lower total cost of local AI deployment

Real-World Demo

Phison demonstrated the technology’s impact live at the show. A 26-billion-parameter AI model, which would normally require at least 32 GB of DRAM, ran smoothly on a system with only 16 GB of DRAM by offloading memory to the aiDAPTIV flash tier .

Intel Partnership Expands Reach

In a separate announcement also made at the show, Phison and Intel revealed a collaboration to bring aiDAPTIV to Intel Core Ultra Series 3 platforms. The partnership is designed to enable more demanding local AI workloads—specifically Mixture-of-Experts (MoE) models and agentic AI applications—on Intel-powered AI PCs .

Next-Gen Raw Speed: The PCIe Gen6 X3 SSD Controller

Beyond the immediate AI memory pitch, Phison demonstrated its ongoing work on raw storage performance with the new PCIe 6.0 X3 SSD controller.

• Architecture: 16-channel, now finalized enough for reference SSD designs
• Sequential Read Speed: Up to 28 GB/s
• Random Read/Write IOPS: Up to 6.8 million
• Max Drive Capacity: Supports up to 2 petabytes per SSD

Tom’s Hardware reported that Phison live-demoed the X3 controller at its Computex booth. Customer sampling is expected in December, with volume shipments planned for mid-2027 .

This speed puts Phison in a tight race with competitors. Silicon Motion’s MonTitan SM8466 controller also hits 28 GB/s, while FADU’s Sierra FC6161 controller pushes slightly ahead at 28.5 GB/s and 6.9 million IOPS .

Enterprise and Sovereign AI: The AISO Consortium

Phison’s ambitions extend well beyond individual SSDs and into multi-vendor enterprise infrastructure. The company is a co-founding member of the AISO (AI for Sovereignty) consortium, a powerful new alliance alongside software integrator TPIsoftware and hardware maker Elitegroup Computer Systems (ECS) .

AISO’s mission is to provide fully integrated, on-premises “all-in-one” AI appliances for enterprises and governments that require data sovereignty and physical control. At Computex, the three co-founders jointly showed these AISO appliances, which combine Phison’s aiDAPTIV technology with TPIsoftware’s OrientAI control center and ECS hardware, delivered as a ready-to-deploy unit .

The AISO consortium also reflects Phison’s recognition that winning enterprise AI contracts requires more than a great chip. Launched in March 2026, AISO is backed by 12 member companies, including Gogolook, Acer Synergy Tech, and Taiwan’s Industrial Technology Research Institute (ITRI), with major early customers Cathay Financial Holdings and Delta Electronics already signed on .

A Broader Industry in Motion

Phison’s Computex announcements did not happen in isolation. They are part of a clear industry-wide realignment where traditional storage companies are racing to become essential to AI infrastructure. The thesis across all of them is the same: as AI inference moves from massive cloud datacenters to edge devices and on-premises servers, flash storage can no longer be a passive data archive—it must become a high-performance memory and compute tier .

The competitive field shown at and around Computex 2026 included:

• Silicon Motion: MonTitan SM8466 PCIe Gen6 controller at 28 GB/s and 7 million IOPS
• FADU: Sierra FC6161 controller reaching 28.5 GB/s reads and 6.9 million IOPS
• Tenam: T-CREATE MASTER Ai I6E PCIe Gen6 SSD in EDSFF E1.S form, targeting HPC and AI training, also at 28 GB/s

Phison’s distinguishing play is not just speed; it’s the aggressive software integration through aiDAPTIV and the AISO consortium that tries to make flash not just faster storage, but an active participant in AI computation itself .

Phison is delivering this through a proprietary software layer and a new hyper-converged infrastructure (HCI) platform called the “AI Data Platform,” which is designed to let enterprises deploy on-premises AI with data sovereignty at a lower cost than cloud or GPU-heavy alternatives .

Pascari aiDAPTIV AI20EH: The Award-Winning AI PC Solution

The flagship product for this new vision is the Pascari aiDAPTIV AI20EH SSD, which won a Computex 2026 Best Choice Award in the “AI PC Solution” category on May 21, 2026 .

How aiDAPTIV Works

the aiDAPTIV technology uses NAND flash as a dedicated AI key-value (KV) cache tier. By adding its middleware, the system pools together GPU VRAM, system DRAM, and NAND flash into one dynamic memory resource. This removes the traditional local memory wall that prevents large models from running on standard hardware .

Claimed Performance Gains

According to Phison’s internal testing, adding the aiDAPTIV AI20EH drive and middleware to an AI PC delivers dramatic improvements :

• Up to 102x improvement in AI inference performance
• 67% reduction in DRAM usage
• Up to 53% lower total cost of local AI deployment

Real-World Demo

Phison demonstrated the technology’s impact live at the show. A 26-billion-parameter AI model, which would normally require at least 32 GB of DRAM, ran smoothly on a system with only 16 GB of DRAM by offloading memory to the aiDAPTIV flash tier .

Intel Partnership Expands Reach

In a separate announcement also made at the show, Phison and Intel revealed a collaboration to bring aiDAPTIV to Intel Core Ultra Series 3 platforms. The partnership is designed to enable more demanding local AI workloads—specifically Mixture-of-Experts (MoE) models and agentic AI applications—on Intel-powered AI PCs .

Next-Gen Raw Speed: The PCIe Gen6 X3 SSD Controller

Beyond the immediate AI memory pitch, Phison demonstrated its ongoing work on raw storage performance with the new PCIe 6.0 X3 SSD controller.

• Architecture: 16-channel, now finalized enough for reference SSD designs
• Sequential Read Speed: Up to 28 GB/s
• Random Read/Write IOPS: Up to 6.8 million
• Max Drive Capacity: Supports up to 2 petabytes per SSD

Tom’s Hardware reported that Phison live-demoed the X3 controller at its Computex booth. Customer sampling is expected in December, with volume shipments planned for mid-2027 .

This speed puts Phison in a tight race with competitors. Silicon Motion’s MonTitan SM8466 controller also hits 28 GB/s, while FADU’s Sierra FC6161 controller pushes slightly ahead at 28.5 GB/s and 6.9 million IOPS .

Enterprise and Sovereign AI: The AISO Consortium

Phison’s ambitions extend well beyond individual SSDs and into multi-vendor enterprise infrastructure. The company is a co-founding member of the AISO (AI for Sovereignty) consortium, a powerful new alliance alongside software integrator TPIsoftware and hardware maker Elitegroup Computer Systems (ECS) .

AISO’s mission is to provide fully integrated, on-premises “all-in-one” AI appliances for enterprises and governments that require data sovereignty and physical control. At Computex, the three co-founders jointly showed these AISO appliances, which combine Phison’s aiDAPTIV technology with TPIsoftware’s OrientAI control center and ECS hardware, delivered as a ready-to-deploy unit .

The AISO consortium also reflects Phison’s recognition that winning enterprise AI contracts requires more than a great chip. Launched in March 2026, AISO is backed by 12 member companies, including Gogolook, Acer Synergy Tech, and Taiwan’s Industrial Technology Research Institute (ITRI), with major early customers Cathay Financial Holdings and Delta Electronics already signed on .

A Broader Industry in Motion

Phison’s Computex announcements did not happen in isolation. They are part of a clear industry-wide realignment where traditional storage companies are racing to become essential to AI infrastructure. The thesis across all of them is the same: as AI inference moves from massive cloud datacenters to edge devices and on-premises servers, flash storage can no longer be a passive data archive—it must become a high-performance memory and compute tier .

The competitive field shown at and around Computex 2026 included:

• Silicon Motion: MonTitan SM8466 PCIe Gen6 controller at 28 GB/s and 7 million IOPS
• FADU: Sierra FC6161 controller reaching 28.5 GB/s reads and 6.9 million IOPS
• Tenam: T-CREATE MASTER Ai I6E PCIe Gen6 SSD in EDSFF E1.S form, targeting HPC and AI training, also at 28 GB/s

Phison’s distinguishing play is not just speed; it’s the aggressive software integration through aiDAPTIV and the AISO consortium that tries to make flash not just faster storage, but an active participant in AI computation itself .