Samsung 2nm Yield Nears 60% — But TSMC Still Owns the Table

Samsung has a yield number. The competition has a moat.

Samsung Electronics disclosed this week that its 2-nanometer foundry process has climbed to roughly 60 percent yield — a concrete data point that moves the conversation forward from hopeful speculation. The company is now concentrating engineering resources on tightening power-per-performance metrics rather than chasing volume throughput, a signal that it knows yield alone does not win design wins.

The math behind the yield improvement is stark. Samsung’s SF2 process — its first GAA (gate-all-around) node — sat in the mid-30s when production troubles surfaced last year. It ticked to the low 50s by January 2026, reached roughly 55 percent by August, and now sits near 60 percent. On paper, that trajectory looks like recovery. In practice, the gap between Samsung and TSMC at the leading edge remains the defining story of the foundry market, and a single yield number does not close it.

Who benefits, who does not

The immediate beneficiaries of higher yield are Samsung’s internal economics and any prospective customer willing to test a small tape-out. Better yield means fewer wasted wafers, lower cost per good die, and a stronger argument when negotiating with a fabless team that has already been burned by process volatility. Samsung is leveraging this window by targeting Tesla as a first external 2nm customer, reportedly committing production volume for next year. That relationship, if it holds, would give Samsung a credible reference design and a beachhead for subsequent pitches to larger clients.

The biggerlosers are the alternative foundry candidates other than TSMC. Intel’s 18A process, which supplies Panther Lake chips from its Arizona Fab 52, has delivered solid results inside Intel’s own product line — the company claims a 56 percent share of the AI PC segment. But Intel has failed to convert those process improvements into external foundry orders at scale. Industry observers note that Intel is already looking ahead to its 14A node (mass production targeted after 2027) as the real bet for external customer recruitment, suggesting that the 18A window has passed without a significant foundry win.

TSMC, meanwhile, faces no pressure from either Samsung’s yield gains or Intel’s process roadmap. Its 2nm wafer price sits at roughly $30,000 — a 50 percent premium over 3nm — and the company maintains a strict no-discount policy. Capacity for 2nm is fully booked through 2028, according to industry assessments. Apple, AMD, and NVIDIA remain anchor customers, and TSMC’s Q2 market share stood at 72.5 percent versus Samsung’s 5.9 percent. The gap is not narrowing; it is calcifying.

The power-efficiency gamble that could change everything

The yield headline conceals a more consequential effort: Samsung’s push to improve power efficiency, which is where foundry competitions are actually won at the leading edge. Exynos 2600 — the first chip manufactured on SF2 and used in some Galaxy S26 models — introduced a copper heat block stacked above the memory layer, reducing thermal resistance by an estimated 30 percent. Independent overseas testing, however, found the chip consuming more power than Qualcomm’s latest Snapdragon parts for identical workloads, drawing criticism that the GAA architecture itself may carry inherent efficiency penalties at this stage.

Samsung is aware of the problem and has SF2P — a second-generation 2nm variant — in development with a 2026 mass-production target. The company projects SF2P will deliver 12 percent more performance, 25 percent lower power consumption, and an 8 percent reduction in chip area compared to SF2. Those figures, if realized, would address the power-efficiency criticism head-on and could become Samsung’s strongest argument when courting AI-ASIC designers who track watts-per-FLOP obsessively.

But projections are not tape-outs. SF2P has yet to run a full reliability validation cycle, and the company has not confirmed whether the power figures hold under real-world HBM-integrated packages — the very configurations that major AI chip buyers require. This is the hidden bottleneck: TSMC already ships 2nm designs with HBM stacked directly on the interposer; Samsung is still proving that its 2nm node can deliver efficiency in those same configurations without thermal throttling or yield collapse.

Why 60 percent matters less than the number suggests

A 60 percent yield on a first-pass GAA node is respectable — Samsung’s own Exynos 2600 is now running in a flagship phone at a reasonable rate. But foundry customers do not buy yield in isolation. They buy the combination of yield, performance stability across their target operating conditions, thermal behavior under sustained AI workloads, and commercial terms that reflect the supplier’s desperation or confidence.

TSMC is confident. Its customers have no credible alternative for the most demanding chips, and the company’s pricing discipline reflects that reality. Samsung is still desperate for external validation. The Tesla order, even at modest volume, would be a significant credibility event — it proves a non-mobile customer trusts Samsung’s 2nm for a high-profile application. But one anchor client does not restructure a market where 72.5 percent of revenue flows to a single foundry.

The financial dimension matters, too. Samsung’s foundry segment remains in the red, and its path to profitability is widely seen as tied to utilization rates at its Taylor, Texas, facility — a plant that has struggled to achieve the capacity factor needed to cover fixed costs. Higher yield reduces per-die waste, but it does not generate demand. Demand comes from design wins, and design wins come from customers who trust a foundry not to delay tape-outs, not to ship chips that overheat in server racks, and not to renegotiate prices mid-project because the node is under financial stress.

What happens next

The near-term trajectory is clear. Samsung will ship SF2P in volume later this year or early next, and the power-efficiency claims will face real-world scrutiny from first external customers. If SF2P delivers, Samsung gains leverage in negotiations with fabless AI and automotive chip teams that are actively seeking a second-source foundry to reduce TSMC dependency. If SF2P underperforms, the 60-percent yield headline becomes a footnote — an impressive recovery on paper that did not translate into competitive advantage.

TSMC will not be idle. Its 2nm capacity lock-up through 2028 means it can afford to wait for Samsung to stumble. Even a strong SF2P rollout will not immediately disrupt TSMC’s customer base, because switching foundries at the 2nm node carries enormous non-recurring engineering costs and re-qualification risk that most fabless companies are unwilling to take lightly.

The structural takeaway is that Samsung has closed a gap — yield is no longer a 30-percent crisis but a 60-percent operational reality. But the market gap, measured in revenue share, customer commitments, and process maturity, remains wider than a single metric can bridge. Until Samsung secures multiple large-design wins on SF2P and proves it at volume, the foundry hierarchy will look the same: TSMC at the top, Samsung fighting for relevance below.