NextFin News - Samsung Electronics reported a 3.9% decline in annual foundry revenue to $12.6 billion on Monday, a contraction that stands in stark contrast to the explosive growth seen across the broader semiconductor manufacturing sector. While the top 10 global foundries saw combined revenues surge 26.3% to $169.5 billion last year, Samsung’s market share slipped to 7.2%, down from 9.4% a year prior. The divergence highlights a widening gap between the South Korean giant and Taiwan Semiconductor Manufacturing Co. (TSMC), which now commands a dominant 69.9% of the market following a 36.1% jump in its own sales.
The revenue dip reflects persistent structural hurdles within Samsung’s logic chip division, primarily centered on manufacturing yields that have historically lagged behind its Taiwanese rival. Industry data suggests Samsung’s 3-nanometer (nm) yields have hovered near 50%, a figure that makes it difficult to secure high-volume orders from margin-conscious clients like Apple or Nvidia, who typically demand yields exceeding 90% to ensure cost-efficiency. Consequently, while TSMC’s 3nm and 2nm order books are reportedly "tapped out" by a roster of tech titans, Samsung has been forced to pivot its strategy toward a more integrated, "one-stop-shop" model to lure customers back to its fabs.
This strategic shift relies on Samsung’s undisputed strength in memory. Unlike TSMC, which is a pure-play foundry that does not design or sell its own branded chips, Samsung is a vertically integrated behemoth. The company is now leveraging its latest High Bandwidth Memory (HBM4) as a bargaining chip to win foundry contracts. Last week, Samsung finalized an agreement to supply HBM4 to AMD, a deal that reportedly includes a framework for AMD to explore using Samsung’s foundry for its next generation of AI accelerators. By bundling scarce memory components with manufacturing capacity, Samsung is attempting to create a value proposition that TSMC—which must partner with external memory makers like SK Hynix—cannot easily replicate.
The stakes for this "memory-first" foundry push are visible in Taylor, Texas, where Samsung is pouring $16.5 billion into a new semiconductor compound. The facility is slated to begin mass-producing AI chips for Tesla in the second half of 2027, a critical win that Samsung hopes will validate its 2nm Gate-All-Around (GAA) transistor architecture. While TSMC is only now transitioning to GAA for its 2nm node, Samsung has been utilizing the technology since its 3nm generation. This early adoption was a gamble that initially hurt yields, but Samsung executives are betting that the "learning pain" will translate into a smoother, more efficient ramp-up for 2nm production compared to TSMC’s upcoming transition.
Capital expenditure remains the primary engine of this pursuit. U.S. President Trump’s administration has continued to emphasize domestic chip manufacturing, providing a political tailwind for Samsung’s Texas expansion. The company announced it will increase its total facilities and research investment by 21.7% this year to $73.5 billion. This aggressive spending is aimed at a specific target: a 20% market share by 2027. Achieving this would require Samsung to not only fix its yield issues but also successfully convince the industry that its integrated HBM-plus-Foundry model is the superior architecture for the AI era.
The immediate financial picture remains challenging. Despite the revenue decline, there are flickers of a turnaround; utilization rates at Samsung’s foundries reportedly climbed back to 80% in the first quarter of 2026, a one-year high. However, the path to profitability is still projected to be a multi-year journey. As OpenAI and other major AI developers begin placing orders for HBM4, Samsung’s ability to convert these memory relationships into long-term manufacturing partnerships will determine if it can finally move out of TSMC’s shadow or if it will remain a specialist player in a market increasingly dominated by a single Taiwanese titan.
Explore more exclusive insights at nextfin.ai.
