Samsung Electro-Mechanics Mass-Produces FC-BGA Substrates for Qualcomm AI200, Entering Server Market for the First Time
Alina Collins
Samsung Electro-Mechanics (三星电机) has begun mass-producing FC-BGA packaging substrates for Qualcomm's first data-center AI accelerator, the AI200, extending their supply relationship from smartphone chips into server-grade semiconductors for the first time — and pulling Korean substrate makers into the AI-server supply-chain race.
What is FC-BGA, and why do AI accelerators need it?
FC-BGA — flip-chip ball grid array, a packaging method that mounts the chip face-down onto the substrate via solder bumps rather than wire bonds — delivers better electrical performance and heat dissipation than traditional packaging.
This means → the higher a chip's compute power and heat output, the more it depends on high-end packaging like FC-BGA to keep things running.
The AI200's FC-BGA sits in the low-to-mid teens in layer count, while top-tier training accelerators exceed 20 layers. In plain terms = the AI200 runs inference (executing pre-trained models), so its substrate specs are less extreme than those of training chips.
Where does Qualcomm's AI200 fit — and how does it differ from Nvidia?
The AI200 focuses on AI inference, not training. It pairs Qualcomm's in-house Oryon CPU with a Hexagon NPU and uses low-power LPDDR5 memory — no reliance on HBM.
This means → Qualcomm is betting on a "low-power inference" lane, differentiating from Nvidia's HBM-driven, high-compute training chips.
Qualcomm unveiled the AI200 and its successor AI250 in October 2025, with commercial availability planned for 2026 and 2027 respectively.
Why did Samsung Electro-Mechanics win the first-mover slot?
Samsung Electro-Mechanics has supplied packaging substrates for Qualcomm's mobile and PC chips for years — the existing relationship was the most direct entry ticket.
This marks the first time their partnership has extended into data-center chips. Initial volumes are limited, consistent with first-batch production norms.
In its earnings disclosures, Samsung Electro-Mechanics already cites AI servers as a growth driver: rising MLCC shipments + FC-BGA demand for AI accelerators and server CPUs have driven year-on-year revenue and operating-profit gains.
Can LG Innotek catch up?
LG Innotek is also targeting Qualcomm's AI200 FC-BGA supply chain, with a mass-production target of 2027 and ongoing development with customers.
Phase 1 of its Vietnam plant represents KRW 1 trillion (~$650 million) in investment, focused on RF-SiP and FC-CSP substrates; a Phase 2 dedicated to FC-BGA capacity is still under negotiation with customers.
In plain terms = the AI200's relatively modest technical bar gives a latecomer like LG Innotek a viable entry window — but Samsung Electro-Mechanics is already shipping, and that time gap is itself a competitive moat.
What does this mean for Korea's packaging industry?
Qualcomm's data-center push is pulling Korean substrate makers out of the smartphone supply chain and into the server supply chain.
This reflects a broader opening: AI inference chips demand moderate packaging specs, unlocking a door that top-tier training chips had kept shut for second-tier Japanese and Korean substrate firms.
Whether Samsung Electro-Mechanics can convert its first-mover position into stable, scaled orders is the key test of how deep this supply relationship can go.
Content is for reference only, not financial advice.