The AI boom has overloaded TSMC's production lines, putting even major clients in a difficult situation. Meta*, Qualcomm, and AMD have stopped viewing Samsung Foundry as a backup option; the Korean manufacturer is becoming a real alternative for bringing products to market in 2026.
The principle is simple: whoever books the capacity first gets priority. Apple and NVIDIA have almost completely reserved the lines for the 3nm process and the future 2nm. For the rest, this means either extreme prices (2nm wafers cost over $30,000) or no guarantee of supply. In this situation, Samsung remains the only player capable of taking part of the orders.
Meta* is actively developing MTIA AI accelerators and is considering Samsung's 2nm SF2 (GAA) process to scale infrastructure without relying on NVIDIA. AMD is in talks to produce EPYC Venice and Instinct MI450 at Samsung's facilities. Intel is also active with the 18A and 14A processes in the US, but Samsung's advantage is experience with GAA.
Weakening TSMC's monopoly gives major clients more freedom and a chance to avoid sharp price increases. 2026 is increasingly being called the year of large-scale diversification—or the "great departure from Taiwan."
***
* — Meta is recognized as an extremist and terrorist organization and is prohibited in the Russian Federation
