Skip to content
hw.dev
hw.dev/signal/qualcomm-samsung-2nm-snapdragon-foundry
SignalTrendForce

Qualcomm Eyes Samsung SF2 2nm for Snapdragon 8 Elite 2 as TSMC Dual-Source Strategy Takes Shape

Qualcomm CEO Cristiano Amon visited Samsung and SK Hynix in Korea to explore manufacturing the Snapdragon 8 Elite 2 on Samsung's SF2 2nm process -- a return to Samsung foundry after a five-year absence driven by dual-sourcing strategy.

Thesis connection
decision making

Qualcomm using Samsung SF2 as a second foundry source for its flagship AP signals that 2nm node capacity is no longer a TSMC exclusive -- which changes the leverage equation for any team allocating advanced node tape-out slots.

#semiconductor#supply-chain
Read Original

Qualcomm CEO Cristiano Amon traveled to Seoul on April 21 for meetings with Samsung Foundry and SK Hynix. The foundry discussion centers on the Snapdragon 8 Elite 2 -- Qualcomm's next application processor -- potentially taping out on Samsung's SF2 2nm process. The deal is not confirmed, but Amon had already disclosed at CES 2026 that the chip design is finished and that Qualcomm has been in discussions with Samsung about manufacturing. The memory conversations covered LPDDR6X samples already delivered by Samsung, with mass production expected in the second half of 2027, plus HBM and SOCAMM configurations for Qualcomm's data center AI200 and AI250 chips.

The foundry angle matters more than the mobile chip story. Qualcomm has manufactured exclusively at TSMC for five years. Its motivation for returning to Samsung is not that TSMC has performance problems -- it is TSMC pricing leverage and allocation priority. TSMC's N2 node is heavily subscribed by Apple, NVIDIA, and now hyperscaler ASICs. Qualcomm having Samsung SF2 as an alternative forces TSMC to negotiate differently. For the mobile AP market this is incremental. For the broader foundry ecosystem, a major fabless customer actively qualifying a second-source at 2nm is a structural shift in who has leverage over tape-out slot allocation.

The data center component of the Seoul meetings is the piece that tends to get buried in the mobile chip coverage. Qualcomm is pursuing server-class memory contracts alongside the AP deal. SOCAMM -- a standardized memory module for servers -- and HBM supply from SK Hynix positions Qualcomm's AI200 and AI250 server chips as viable DRAM-adjacent alternatives to GPU memory configurations. These chips are not competing with Nvidia on training workloads. They are targeting inference on the power-constrained edge of the data center, where Qualcomm's architectural efficiency from the mobile market translates directly. That conversation is happening now because the memory supply chain for those chips needs to be locked in well ahead of the 2027 production ramp.