Samsung Electronics warns AI chip shortage to worsen by 2028

Samsung’s latest earnings call confirms what many in AI already feel on the ground, the real bottleneck is shifting from model design to memory supply. A 250-fold jump in chip profits off the back of AI demand tells you that high bandwidth memory is becoming the new oil for frontier models. When the world’s largest memory producer says that shortages in 2027 could be worse than today, it signals that the constraint on AGI scale over the next few years will be access to advanced DRAM capacity as much as GPU compute itself.

For the race to AGI, this deepens the moat for firms that have already locked in multi‑year supply with Samsung and SK Hynix. Hyperscalers and top labs that secured HBM pipelines early can keep training ever larger and more agentic systems while latecomers struggle to get on the allocation list at any price. It also increases the strategic leverage of Korean memory makers in the global AI stack, tilting some bargaining power away from US GPU vendors. Over the medium term, sustained supernormal profits will encourage aggressive capex, but the two to three year lag on new fabs means the industry is likely to run hot and constrained just as agentic AI deployments are accelerating.