Fabless SSD controller and power-management IC specialist FADU is recalibrating its technology bets. While demand for AI data center SSDs is driving rapid revenue growth, the company is cooling its investment in CXL switching and openly questioning whether high-bandwidth flash (HBF) will ever become a meaningful market.
Rather than chasing every emerging interconnect trend, FADU is concentrating on areas where it believes near-term deployment, power efficiency, and system integration will matter most.
📈 Strong Growth Driven by AI SSD Demand #
FADU has reported a sharp acceleration in financial performance, fueled by hyperscaler and server OEM demand for AI-focused storage.
- Revenue growth:
From KRW 10.1 billion in Q3 2024 to KRW 25.6 billion in Q3 2025 — roughly 2.5× year-over-year growth - Drivers:
- Expanding AI data center deployments
- Multiple large-scale customer wins over four consecutive months
- Outlook:
Management expects revenue to rise substantially again next year as AI infrastructure spending continues.
This momentum reinforces FADU’s core strength: high-performance, power-efficient SSD controller technology optimized for modern data center workloads.
⚙️ PCIe Gen6 Controllers and Integrated Power Management #
At a recent earnings briefing, co-founder and CTO Eyee Hyun Nam outlined FADU’s near-term technology roadmap.
- PCIe Gen6 SSD controller:
Currently under development, targeting next-generation AI servers - DRAM PMIC modules:
Planned for launch next year, designed to deliver higher power efficiency than competing solutions - Key differentiation:
FADU is pursuing integrated SSD controller + PMIC designs, while many competitors continue to ship these as separate devices
This tighter integration reflects a growing industry focus on power density and thermal efficiency, particularly in AI data centers where every watt matters.
🔗 Cooling Enthusiasm for CXL Switching #
Like Panmnesia, FADU has explored CXL 3.0 memory-sharing switches, developed through its EEUM subsidiary. However, the company is now scaling back active investment in this area.
The reasoning is pragmatic:
- NVLink dominance:
NVIDIA’s NVLink-based memory pooling inside DGX/HGX pods continues to strengthen, reducing near-term demand for external CXL memory fabrics. - Slow market adoption:
The broader CXL memory-sharing ecosystem is developing more slowly than initially expected.
FADU’s current plan is to:
- Continue CXL switch development only up to FPGA-based proof-of-concept
- Freeze the project afterward until (and unless) a large, sustainable CXL memory-sharing market emerges
This signals a shift from speculative infrastructure bets toward technologies with clearer customer pull.
💾 Skepticism Around High-Bandwidth Flash (HBF) #
CEO and co-founder Jihyo Lee also addressed the growing industry discussion around high-bandwidth flash, promoted by Sandisk and SK hynix as a potential complement to GPU HBM.
FADU outlined three major obstacles that could limit HBF adoption:
-
Thermal mismatch
GPUs can operate at temperatures that exceed what NAND flash can reliably tolerate, complicating co-packaged designs. -
Endurance limitations
TLC and QLC NAND have far lower write endurance than DRAM, potentially shortening the lifespan of HBF stacks under AI workloads. -
Ecosystem fragmentation
Limited compatibility across NAND types makes it difficult for controller vendors to address a large, unified market.
Crucially, NVIDIA — the largest HBM consumer and the most important potential HBF customer — has not publicly endorsed HBF, adding further uncertainty to the technology’s prospects.
🧭 Strategic Takeaway #
FADU’s repositioning highlights a broader reality in AI infrastructure:
- Not every promising interconnect or memory concept will scale into a mass market
- Power efficiency, integration, and deployability increasingly matter more than raw bandwidth alone
By focusing on PCIe Gen6 SSD controllers, integrated PMICs, and proven AI data center demand, FADU is betting that incremental but deployable gains will outperform more speculative architectural shifts — at least in the near to medium term.
For now, CXL switching and high-bandwidth flash remain technologies to watch, not technologies FADU is willing to heavily fund.