A recent report suggests that AMD is preparing to partner with Samsung for its advanced 2nm chip production, a significant shift in the AI semiconductor landscape.
This move isn't happening in a vacuum; it's a direct response to the immense pressure on the global chip supply chain. For years, TSMC has been the undisputed leader for cutting-edge chips, but its capacity is now stretched to the limit. Reports indicate that its advanced production lines are fully booked through 2028, with a single customer, Apple, reportedly securing over 50% of the initial 2nm supply. This leaves companies like AMD in a precarious position, facing potential production delays and limited supply for their next-generation AI CPUs and GPUs. To mitigate this risk, finding a reliable second source has become a strategic necessity.
Samsung has emerged as the most credible alternative for several key reasons. First, the company offers a powerful 'total solution' that bundles logic chips with high-demand HBM4 memory. AMD and Samsung already signed a memorandum of understanding (MoU) in March making Samsung the primary HBM4 supplier for AMD's upcoming AI accelerators. In a market where HBM is a major bottleneck, sourcing both critical components from one partner is a compelling advantage.
Second, Samsung's manufacturing technology is catching up. While its 2nm process yield, estimated around 55%, still trails TSMC's, it's approaching the threshold considered viable for mass production. For AMD's chiplet-based designs, which use smaller, interconnected tiles, this level of yield becomes manageable, as it allows for binning less-than-perfect components for different product tiers. Finally, with advanced packaging like CoWoS also in short supply, Samsung’s alternative packaging solutions provide another layer of supply chain security.
This diversification strategy is further supported by government policy and a broader industry trend. The U.S. CHIPS Act actively encourages building resilient, geographically diverse supply chains, providing funding to both Samsung and TSMC for their U.S. facilities. Moreover, with even TSMC's most loyal customer, Apple, reportedly exploring other foundries, the idea of multi-sourcing is no longer a backup plan but a mainstream strategy.
In conclusion, AMD's move is best understood as a calculated hedge. TSMC will likely remain its primary partner for performance-critical chips, but bringing Samsung into the fold provides a crucial buffer against supply constraints, gives AMD more negotiating power, and secures a vital supply of HBM memory. It signals a new era where the absolute dominance of a single foundry is giving way to a more balanced and resilient ecosystem.
- Foundry: A company that manufactures semiconductor chips designed by other companies (known as 'fabless' companies).
- Yield: In chip manufacturing, this refers to the percentage of functional, non-defective chips produced from a single silicon wafer. Higher yield means lower cost and more efficient production.
- HBM (High Bandwidth Memory): A specialized type of high-performance memory stacked vertically to provide much faster data speeds than traditional memory, essential for AI and graphics processing.