Table of Contents
Samsung’s 2029 target represents a two-year delay from original plans
In sum – what we know:
- A two-year delay – Samsung has resumed 1.4nm development with mass production now targeted for 2029, versus the 2027 date it originally promised.
- 2nm comes first – Samsung redirected resources toward stabilizing 2nm and 4nm, a pivot that helped land a large Tesla foundry order.
- Equipment moving forward – Samsung has shared 1.4nm process plans with tool vendors like Applied Materials and Lam Research and installed ASML’s High-NA EUV system for the node.
Samsung Foundry has officially resumed commercialization and R&D work on its 1.4nm (SF1.4) process node, with mass production now formally targeted for 2029. That’s a notable restart after a period in which the node looked like it might be scaled back, mothballed, or scrapped altogether. The move confirms Samsung hasn’t given up on the leading edge, even if it’s arriving a few years later than it once promised.
The timing is worth putting in context. Back at Samsung Foundry Forum 2022, the company laid out an aggressive roadmap that had 2nm entering mass production in 2025 and 1.4nm following in 2027. The new 2029 target represents a two-year slip on that original plan, and will be quickly followed up by the more advanced SF1.4+ process node.
TSMC, of course, is planning its own 1.4nm equivalent, sometimes referred to as A14, for mass production around 2028, and Intel has signaled its “14A” process for mass production in 2029.
Strategic pivot to 2nm and 4nm
When Samsung first postponed 1.4nm, leaks and reports through 2024 and early 2025 suggested Samsung might abandon high-volume 1.4nm manufacturing entirely, and the company did postpone construction of a 1.4nm trial line that had been slated for Pyeongtaek. That fueled a fairly grim narrative about a foundry business in crisis, one that couldn’t afford to chase the next node while its current ones were still shaky.
There was substance behind the worry. Samsung’s 3nm introduction was hampered by well-documented yield problems, and those issues cost it credibility against TSMC among customers who want the most advanced nodes. So rather than push 1.4nm out the door on schedule, Samsung redirected its resources toward stabilizing 2nm and 4nm, improving yields, and lifting fab utilization rates. A company spokesperson framed the shift as strategic, emphasizing the focus on “stabilising existing processes such as 2nm and 4nm and improving yields” over rushing to the next node.
Samsung secured a $16.5 billion foundry order from Tesla for its “AI6” chip, reportedly slated to be built on Samsung’s improved 2nm process — a deal that’s being read by some as a sign of improving confidence. The AI6 is planned for cars, AI servers, and humanoid robots, which is about as broad an AI and edge-compute mandate as a single chip can carry. More to the point, landing a customer of that profile is being read across the industry as evidence that Samsung’s 2nm yields have recovered enough to attract serious business. That’s the kind of validation Samsung couldn’t buy after the 3nm episode.
On the timeline, 2nm (the SF2 family) is expected to reach mass production in Korea by the end of 2025 and at the Taylor, Texas fab in late 2026 or early 2027, subject to execution and demand. Samsung’s 2nm roadmap also includes SF2P and SF2P+ variants, with SF2P+ reportedly targeted for around 2027. In other words, Samsung is putting its weight behind proving reliability at 2nm before it commits fully to 1.4nm. Given how the 3nm story went, that’s a defensible order of operations.
Implications for AI chips
For the next several years, any AI silicon Samsung builds will run on 2nm and advanced 4nm, not 1.4nm. That’s simply where the capacity and the proven yields are. The Tesla AI6 deal already positions 2nm as an AI platform, and Samsung is tuning specific nodes for the job rather than waiting on a single future process. On the 4nm side, SF4U is being optimized explicitly for AI accelerators, automotive SoCs, and mobile APs, while high-performance SF2P and SF2P+ cover the more demanding end.
SF1.4 is the long-term bet. Once it enters mass production around 2029, it promises higher transistor density, lower power, and better performance-per-watt, all of which matter enormously for large AI accelerators and AI-centric SoCs. Paired with High-NA EUV and customized equipment, it’s likely to be marketed squarely at AI and HPC customers, the cloud vendors and GPU designers chasing every last bit of efficiency. That’s the ambition, anyway.
The problem is that node parity alone won’t win Samsung the business. TSMC currently dominates AI GPU and accelerator production, building for Nvidia, AMD, and the rest at 3nm and increasingly 2nm. Analysts warn that a delayed SF1.4 risks major accelerator customers locking into TSMC’s earlier 1.4nm rollout, which arrives roughly a year ahead. Once a customer commits its next-generation designs to a foundry, prying them loose is hard. So Samsung’s AI success will likely hinge on pricing, capacity, ecosystem, and integrated turnkey memory and packaging, rather than simply matching TSMC on transistors. Samsung has reportedly promised partners it will strengthen turnkey services using its own memory, which is a sensible angle given it makes both the logic and the DRAM.
Equipment and High-NA EUV integration
The clearest sign that the resumption is more than talk is what’s happening on the equipment side. Samsung has reportedly shared detailed SF1.4 process plans with major tool vendors, including Applied Materials and Lam Research, to coordinate development and process integration. That’s the sort of collaborative, node-specific engineering you’d expect from a company that’s serious about a schedule, not one quietly winding a project down.
Samsung is also said to be pursuing customized versions of standard manufacturing tools tailored to 1.4nm’s particular requirements, rather than adopting off-the-shelf equipment. And it has installed ASML’s High-NA EUV lithography system at its NRD-K R&D complex, with industry reporting suggesting the tool will enter production starting with the SF1.4 node.
That last point matters more than it might sound. High-NA EUV delivers the sub-2nm patterning density and precision that extreme transistor requirements demand, which is precisely the kind of workload AI accelerators and HPC chips generate. Getting the tool in place and integrated now, well ahead of 2029, is how you avoid another 3nm-style yield scramble later. Whether Samsung executes on that is the open question, but the pieces are being put in position.
