Samsung Foundry Loss Seen Narrowing 42% as HBM4 Base-Die Orders Fill 4nm Lines
HBM4 volumes are doing double duty at Samsung, generating logic-foundry revenue through the 4nm base die, lifting utilisation enough to support higher contract prices and cut projected 2026 foundry losses by more than 40%.
TrendForce reported on September 30, citing Korean media and Kiwoom Securities estimates, that the combined operating loss of Samsung Electronics' foundry and System LSI businesses is projected to narrow 41.8% year on year in 2026, from KRW 6.74 trillion to about KRW 3.92 trillion, with the quarterly loss shrinking to roughly KRW 777 billion in the third quarter. The improvement is structural rather than financial engineering: because HBM4 base dies are produced on Samsung's own foundry logic process, every HBM4 stack shipped also books foundry revenue. Samsung began HBM4 volume shipments in February using 1c DRAM paired with a 4nm base die, then shipped HBM4E samples to customers in May, and now expects HBM4 revenue to more than triple quarter on quarter in the third quarter and account for well over 60% of total HBM revenue in the second half.
That demand is changing how Samsung prices its own capacity. TrendForce reported, citing iNews24, that strong demand for HBM4 base dies and advanced-node orders has reduced the foundry's need to accept low-priced contracts purely to raise utilisation, and that Samsung raised prices on new 4nm orders in July, with some advanced-node prices reportedly up 10% to 15%. For a business that has posted quarterly losses between KRW 700 billion and KRW 1.2 trillion this year, the shift from filling lines at any price to turning away marginal work is a meaningful change in negotiating position, and it matters for every fabless customer competing for the same limited advanced-node slots.
The company is still working to broaden its base. After exceeding KRW 2 trillion of quarterly losses in the first half of last year, Samsung has leaned on internal memory-related demand, but Korean reporting notes that captive orders alone cannot fully load its foundry lines. Samsung signed a memorandum of understanding with Broadcom in July covering more than $200 billion of cooperation through 2030, including HBM4 and HBM4E supply for Broadcom-designed custom AI accelerators alongside 2nm-and-below logic and advanced packaging, and it secured 2nm projects from a major cloud service provider plus AI and high-performance computing customers in the second quarter, with customer product design already underway. Whether those external orders arrive fast enough is the clearest test of whether the loss-reduction path holds into 2027.