TSMC vs Samsung vs Intel Foundry: The Three-Way Competition
The foundry business is a three-horse race, and it's not particularly close. TSMC dominates with about 60% market share by revenue, Samsung Foundry holds roughly 12%, and Intel Foundry Services (IFS) is an ambitious newcomer burning through billions to establish itself. Here's where each stands, what their strategies are, and how the competition is likely to play out.
TSMC: The Undisputed Leader
TSMC's dominance is built on three pillars: technology leadership, manufacturing excellence, and customer trust. They manufacture chips for Apple, NVIDIA, AMD, Qualcomm, Broadcom, MediaTek — essentially every major fabless chip designer except Samsung's own product divisions.
At the leading edge, TSMC's track record is unmatched. Their N5 (5nm) node ramped to high yield faster than anyone expected. N3 (3nm) followed the same trajectory. When TSMC says a node will deliver certain performance at a certain yield level by a certain date, customers have learned to believe them. That reliability is worth more than any spec sheet advantage.
TSMC's N2 (2nm) is on track for 2025 volume production, using GAA nanosheet transistors. They're also developing N2P (with backside power delivery) and A14 (1.4nm equivalent) for 2027-2028. Their technology roadmap extends visibility three to four generations ahead, giving customers confidence to design chips targeting process nodes that won't be ready for years.
Revenue numbers tell the story: TSMC generated roughly $75 billion in 2024 revenue, with advanced nodes (7nm and below) representing about 65% of that. Their gross margins hover around 55% — extraordinary for a manufacturing business and reflective of their pricing power.
Samsung Foundry: The Challenger's Dilemma
Samsung's foundry division faces a structural challenge: it competes for customers while Samsung's own semiconductor division (which designs Exynos, memory chips, and display drivers) is also a customer. Other chip designers worry that Samsung might prioritize its own products over external customers during capacity constraints, or that proprietary design information might leak between Samsung's divisions.
See also: Persistent Memory: Intel Optane Legacy, CXL-Attached PM, and.
This conflict of interest, whether real or perceived, has cost Samsung major customers. Qualcomm famously split its orders between Samsung and TSMC after the Snapdragon 888 (Samsung 5nm) had higher power consumption than expected. The perception of quality issues at Samsung's advanced nodes — partially deserved, partially unfair — created a negative feedback loop: customers leaving meant less revenue for R&D, which made it harder to close the technology gap.
Samsung's technical position: their 3nm GAA process (the first in the industry) demonstrated that GAA manufacturing is possible, but early yields were disappointing. Their 2nm node, building on GAA lessons learned, needs to perform well to retain remaining customers and attract new ones.
Samsung has a few genuine advantages: competitive pricing (they'll undercut TSMC to win business), advanced packaging capabilities, and co-located DRAM production that enables tight memory-logic integration. For customers willing to accept some technology risk in exchange for lower cost and packaging flexibility, Samsung can be attractive.
Intel Foundry Services: The $100 Billion Bet
Intel's foundry ambitions are the biggest gamble in semiconductor history. CEO Pat Gelsinger committed Intel to becoming a world-class contract manufacturer — a business Intel had never been in — while simultaneously trying to fix Intel's own product competitiveness issues.
The IFS strategy rests on Intel 18A (roughly 1.8nm), which introduces both RibbonFET (GAA nanosheets) and PowerVia (backside power delivery). If 18A is competitive with TSMC N2 and Samsung 2nm, IFS becomes a credible alternative foundry. If it's not, Intel has spent tens of billions on fabs that won't generate adequate returns.
Related reading: HBM4 Memory Technology: Architecture, Bandwidth, and the AI .
Early signs are cautiously positive. Microsoft has committed to using Intel 18A for custom chips. The US government's CHIPS Act subsidies provide financial support for Intel's domestic fab construction. Intel's new fabs in Ohio, Arizona, and Germany are under construction.
But the challenges are immense. Intel has no track record manufacturing other companies' designs. Their EDA tool ecosystem is optimized for Intel's own design methodology, not for the diverse requirements of external customers. Customer support, IP libraries, and design enablement — all the things that TSMC excels at — need to be built from scratch.
Technology Comparison at the 2nm Generation
All three foundries are transitioning to GAA nanosheet transistors at their respective 2nm nodes. The high-level specifications are similar, but the details matter:
- TSMC N2: Conservative GAA implementation focused on yield and reliability. No backside power delivery in initial version (planned for N2P). Likely the highest yield and most predictable performance.
- Samsung SF2: Second-generation GAA building on 3nm experience. Targeting competitive PPA with aggressive pricing. Needs to prove yield improvement over SF3.
- Intel 18A: GAA plus backside power delivery from the start. Technically the most ambitious approach, but also the riskiest since both technologies are new for Intel.
The Mature Node Business
The foundry conversation usually focuses on leading-edge nodes, but the majority of chips by volume are manufactured on older processes. 28nm, 40nm, 65nm, and even 180nm nodes still serve massive markets: automotive, industrial, RF, power management, and consumer electronics.
TSMC dominates mature nodes too, but companies like GlobalFoundries, UMC, and SMIC are significant players. China's SMIC has expanded aggressively in mature node capacity, and Chinese government subsidies make their pricing very competitive. This has raised concerns about potential overcapacity and price dumping in the mature node market.
For a related perspective, see AI Chip 2026: NVIDIA B200 vs AMD MI455X vs Gaudi 3.
For foundry customers, dual-sourcing across mature node foundries is relatively straightforward compared to leading-edge nodes. The processes are well-established, IP libraries are widely available, and switching costs are lower. This limits pricing power for any single foundry at mature nodes.
Looking Forward
The most likely scenario for the next five years: TSMC maintains technology leadership and the majority of advanced node revenue. Samsung stabilizes its foundry business at a distant second place, competing on price and packaging. Intel's IFS either establishes itself as a credible third option (if 18A succeeds) or scales back its foundry ambitions (if it doesn't).
The wild card is geopolitics. If tensions over Taiwan escalate, the value of non-Taiwan foundry capacity increases dramatically regardless of the technology gap. This is explicitly why the US, EU, and Japan are subsidizing domestic semiconductor manufacturing. TSMC's technology leadership means little if access to their fabs is disrupted.
I'd argue the foundry market is healthiest with three strong competitors. TSMC as a near-monopoly on advanced manufacturing creates a systemic risk that benefits no one — including TSMC, which faces increasing pressure to build fabs outside Taiwan. Competition from Samsung and Intel pushes all three to innovate faster and price more competitively. Whether Samsung and Intel can actually deliver competitive technology is the trillion-dollar question.