Intel shipped its first 18A chips to customers in late 2025, and TSMC's N2 node entered high-volume manufacturing around the same window, according to industry reporting tracked by TechInsights. Samsung's SF2 node, by contrast, has yet to move from risk production into volume, leaving the Korean foundry a full step behind its two rivals as of mid-2026.
That gap matters because 2nm is the node where the AI boom meets physics. TSMC, Intel, and Samsung are all chasing the same pool of AI accelerator, mobile, and high-performance computing customers, and the company that delivers working silicon first gets to set the price.
This story matters right now because the balance of power in advanced manufacturing is being decided this quarter, not next year. Intel's Panther Lake and Clearwater Forest parts are already sampling on 18A, TSMC's N2 is absorbing Apple and AMD volume, and Samsung is still courting customers with promises rather than silicon. Whichever foundry stumbles in yield this year will pay for it in 2027 orders.
Intel's 18A Bet
Intel's 18A node uses RibbonFET gate-all-around transistors and PowerVia backside power delivery, two features the company has spent years developing. Early yield reports from Intel suggest the node is producing workable die, but the company has not published a yield figure, and analysts at TechInsights have noted that 18A's density lead over TSMC N2 is smaller than Intel's marketing claims.
Intel has bet its foundry business on 18A, and the stakes are existential. The company's foundry unit has lost money for years, and 18A is the node meant to prove that Intel can manufacture for external customers at the leading edge. If 18A yields hold, Intel can win business from customers who want US-based manufacturing. If they do not, the foundry strategy collapses into a cost center.
PowerVia backside power delivery is the part of 18A that competitors cannot easily copy. By routing power from the back of the wafer, Intel claims it frees front-side metal layers for signal routing, improving density. Whether that translates into a real performance advantage depends on how customers design their chips, and on whether Intel can keep yields high enough to make the node economically viable.
TSMC's Proven Execution
TSMC, meanwhile, has kept its N2 schedule on track. The node entered high-volume manufacturing in late 2025 with a ramp through the first quarter of 2026, and Apple is widely expected to be the anchor customer for its first 2nm iPhone chips. AMD has also been linked to N2 for future EPYC and Radeon parts.
TSMC's strength is execution, not invention. The company rarely leads on transistor architecture, but it consistently delivers working silicon at scale, and that predictability is what keeps Apple, Nvidia, and AMD coming back. N2 is no exception: the node uses nanosheet transistors, a more conservative evolution than Intel's RibbonFET, but one that TSMC has refined through multiple generations of learning.
The risk for TSMC is capacity. Demand for advanced nodes outstrips supply, and customers have complained about long lead times and tight allocation. TSMC is building new fabs in Arizona and Japan, but those facilities will not reach leading-edge production for years. In the meantime, N2 capacity in Taiwan is the bottleneck for the entire AI industry.
Samsung's Catch-Up Problem
Samsung's position is the weakest of the three. The company's SF2 node remains in risk production, and its 3nm predecessor nodes suffered yield problems that pushed major customers toward TSMC. Samsung has said it is targeting volume production in 2026, but analysts have heard that before.
Samsung is not standing still. The company has invested in advanced packaging and is chasing HBM4 business alongside SK hynix and Micron. But until SF2 produces silicon in volume, those packaging gains will not matter for leading-edge logic customers. Samsung's foundry unit needs a win on SF2 to stay relevant in the leading-edge race.
The company's broader strategy is to compete on price and capacity rather than on transistor innovation. Samsung has offered aggressive pricing to win customers, and its fabs in Korea and Texas give it geographic diversification that TSMC lacks. But price only matters if the yields are high enough to make the economics work, and that is the question Samsung has not yet answered.
The Packaging Bottleneck
The packaging layer adds another wrinkle. Even if the transistor node works, advanced packaging like TSMC's CoWoS and Intel's Foveros is the real bottleneck for AI chips. SK hynix emphasized memory-logic integration at TSMC's 2026 symposium, a sign that HBM and logic are being co-designed rather than treated as separate parts.
For AI accelerators, the package matters as much as the die. A leading-edge chip that cannot be connected to enough memory is useless, and the industry is short of both CoWoS capacity and HBM supply. That means the foundry race is really two races: one for transistors and one for packaging, and the winner has to win both.
Geopolitics and the Customer Calculus
The US export control regime hangs over all of this. Restrictions on advanced chipmaking equipment to China have tightened in recent years, and both TSMC and Intel have faced pressure to keep certain processes out of Chinese fabs. Samsung's foundry business, which has a fab in China, has had to navigate those rules carefully.
For customers, the choice comes down to risk. TSMC offers proven execution but limited capacity. Intel offers aggressive pricing and US-based manufacturing, which matters to customers worried about geopolitics. Samsung offers a backup option, but only if it can prove its yields.
The financial stakes are enormous. A single advanced fab now costs more than $20 billion to build, and yield on a new node can swing a company's quarterly results by hundreds of millions of dollars. If AI capital spending slows, the math behind those fabs gets harder, and the foundry race could turn into a capacity glut.
Analysts expect 2026 to be another record year for semiconductor sales, driven by AI demand. That demand is what funds the 2nm race in the first place. The next twelve months will show whether the current ordering holds, or whether one of the three stumbles in a way that reshuffles the entire foundry market.
Read more about the industry outlook at Deloitte's 2026 semiconductor outlook.
See also our semiconductors coverage for related reporting.