The deadlock is the story. Qualcomm has asked Samsung for a price cut on 2nm, Samsung has refused, and the two sides have left a flagship AP program potentially sitting until 2027.
That is the whole negotiating room in one sentence. Qualcomm wants a discount. Samsung no longer believes it has to buy orders.
Last week we traced how Samsung's 4nm capacity was being pulled toward HBM base dies, tightening the supply of conventional logic wafers. This week the same fight has moved up a node, to the 2nm line, and the number both parties are actually staring at is a yield figure neither controls.
The talk of a "dual source" is the choreography around it. Reports have Qualcomm evaluating splitting Snapdragon 8 Elite Gen 6 production between Samsung and TSMC, with discussions covering specifications, pricing, and process optimization. Exact production split: undecided. Expected role for Samsung: backup supplier, not equal partner, at least initially.
Which is not a supply plan. It is a bargaining instrument.
Start with Samsung's side of the table, because it explains the refusal. The company's SF2 and SF2P yields have climbed from roughly 20% in the second half of 2025 to 50–60% now. Some reports put it near 55%.
That is a genuine recovery, and the reason Samsung is not discounting. Earlier in this cycle, it accepted customer-set cost conditions to win Tesla's AI5 — the classic price-for-volume trade. Not this time.
But 50–60% is not the threshold. Qualcomm's requirement for high-volume flagship contracts is approximately 70% yield, a bar set because a flagship Snapdragon ships in the tens of millions and a yield curve that wobbles mid-ramp is a supply crisis, not a cost problem.
Samsung knows this history personally. Its 3nm GAA process started with yields around 20%, customers walked, and the 3nm business lost $385 million. Samsung's 4nm built the Snapdragon 8 Gen 1, whose heat and power problems pushed Qualcomm toward TSMC and kept it there.
Now the comparison: TSMC's N2 is running at roughly 70% yield, and the company is aiming to push N2P toward 80% within the year. The gap between Samsung and TSMC is no longer a cliff. It is a step.
The problem is that a step is exactly the distance between a qualified second source and a press release.
Here is where the supply side makes Qualcomm's position worse, not better. TSMC plans to lift 2nm capacity to 60,000 wafers per month in the fourth quarter — and Apple has reportedly locked more than half of that for the iPhone 18 family and M6 chips.
Everyone else on the customer list is competing for the remainder. AMD, MediaTek, Broadcom, Intel, Qualcomm.
TSMC's own expansion targets, outlined to suppliers for the first time, are real but modest against that demand. 2nm goes from 90,000 wafers per month at the end of 2026 to 110,000 by mid-2027 — 22%. 3nm goes from just over 180,000 to 210,000 — over 16%, and roughly 70% higher than the 120,000–130,000 wafers per month of late 2025.
Set those steps against order visibility that TSMC says now extends into 2029 and 2030. A 22% capacity increase is not a relief valve against that kind of backlog. TSMC is adding three 3nm fabs across Taiwan, Arizona, and Japan, converting 5nm equipment in Taiwan, and running a 2026 capital budget of $60–64 billion, with 70–80% of it going to advanced nodes.
That is enormous spending producing an incremental supply response. The squeeze is structural, not temporary, and Qualcomm is feeling it in exactly the quarter it needs to book.
So the logic of the dual-source talk becomes legible. Samsung at 55% cannot manufacture Snapdragon volume. Samsung at 55% can still shift the price TSMC quotes.
That is what Qualcomm is buying with this negotiation. Not wafers — leverage. And Samsung, knowing this, will not discount for orders it can't run at scale and doesn't need to chase while its own 2nm line is being fought over.
Both sides are, in effect, negotiating over a yield number rather than a contract.
Both sides are, in effect, negotiating over a yield number rather than a contract.
The question is when that number moves. Qualcomm's stated threshold is 70%, and the industry's more important metric is not the peak but the hold — whether a yield line stays above 70% through a high-volume ramp, not whether it touches it in a pilot run.
Until that happens, the second source is a rumor with a spreadsheet attached. TSMC's 2nm wafer pricing, reportedly around $30,000 apiece, is what gives Samsung its opening. It is also what keeps Qualcomm at the table without a commitment.
Worth naming the bloc here. Qualcomm, Nvidia, AMD, Apple, and the large cloud silicon teams all buy from the same two or three leading-edge lines, and their individual negotiating positions are weaker than they look because they are all queuing for the same capacity.
Apple's lock on more than half of initial 2nm output makes that queue longer for everyone else. Nvidia and AMD's 3nm orders are reportedly booked into 2027. A credible Samsung would give that whole group a second door.
A 55%-yield Samsung does not. It gives each of them a talking point.
The consequence lands in 2027. Wafer negotiations for that year will be settled on Samsung's yield number, not on the announcement of a design win, and not on the existence of a memorandum of understanding.
When Samsung holds 70% through a ramp, the second source arrives quietly — because it becomes boring, because it books capacity without headlines, because procurement teams stop needing to mention it in meetings to get a better quote.
Until then, the Nvidia/AMD/Qualcomm/Apple bloc keeps negotiating against a single supplier, and Qualcomm's dual-source discussions keep serving their actual function: telling TSMC there is somewhere else to go.