TSMC, Samsung lift foundry prices 15% as AI demand shifts power

Wafer foundry pricing is undergoing its most structural shift in a decade as AI chip demand keeps advanced nodes at full capacity.

TSMC and Samsung Electronics have raised prices on advanced chipmaking processes by as much as 15%, the clearest sign yet that AI chip demand has flipped wafer foundry pricing power from customers to suppliers. TSMC notified Nvidia, Apple, and AMD of 5% to 10% increases on its 3nm, 5nm, and 7nm nodes — processes that account for more than 70% of its foundry revenue. Samsung followed with roughly 15% hikes for new clients on 4nm and 5nm processes, plus select automotive-grade 8nm nodes.

"The pricing dynamic has fundamentally changed because capacity utilization at leading-edge nodes hasn't dropped below 95% in over a year," Donghai Securities analysts wrote in a July 13 report. "This is no longer about process maturity discounts — it's about who controls the bottleneck."

The price increases break with decades of industry convention where foundries gradually lowered per-wafer costs as process yields improved and production scaled. TSMC's 3nm node, which entered volume production in late 2023, would typically be entering a phase of price moderation by now. Instead, the opposite is happening: the 3nm price hike comes as the node is still ramping, reflecting demand that outstrips available supply. Samsung's 15% increase on 4nm and 5nm — nodes that are now several years old — is even more unusual, suggesting the company is using tight capacity to normalize pricing that had fallen during its earlier yield struggles.

The root cause is AI chip demand that shows no signs of slowing. Nvidia's H100 and B200 data center GPUs, AMD's MI300X series, and Apple's M-series processors all compete for the same limited pool of TSMC's advanced capacity. Samsung, meanwhile, is diverting some of its 4nm and 5nm capacity to produce HBM (high-bandwidth memory) logic dies for AI accelerators, further squeezing supply for traditional foundry customers. The industry is also absorbing massive capital expenditure for 2nm development — a node expected to cost more than $30 billion in R&D and tooling across both foundries — costs that must be recovered through higher wafer prices.

Who Wins, Who Loses

The pricing shift creates a clear divergence in fortunes. TSMC and Samsung benefit directly: every percentage point of price increase flows largely to operating profit given that wafer costs are fixed in the short term. TSMC reported gross margins above 53% in its most recent quarter, and the new increases could push that figure toward 55% or higher. Samsung's foundry business, which has lagged TSMC in both yield and profitability, stands to narrow the gap as its price increases outpace TSMC's.

For fabless chip companies — Nvidia, AMD, Apple, Qualcomm — the higher wafer costs will compress gross margins unless they pass them to end customers. Nvidia's data center gross margins, which have hovered around 78%, could face 100 to 200 basis points of pressure per percentage point of wafer cost increase, according to Bernstein estimates. Apple, which designs its own A-series and M-series chips but relies entirely on TSMC for production, faces a similar dynamic across its iPhone and Mac lines.

Smaller fabless firms without the pricing power of Nvidia or Apple are most exposed. Companies like AMD, which competes directly with Nvidia in AI GPUs, may find it harder to absorb the increases while maintaining competitive pricing. MediaTek and Qualcomm, which use TSMC's 4nm and 3nm nodes for smartphone processors, face margin compression in a market where end-device pricing is already under pressure.

A Structural Shift in Pricing Power

The broader implication is that wafer foundry has permanently moved from a buyer's market — where customers could demand lower prices as nodes matured — to a supplier-dominated market where foundries set terms based on demand and investment needs. This mirrors what happened in the memory chip industry a decade ago, when Samsung and SK Hynix consolidated production and shifted from cyclical pricing to more disciplined capacity management.

For investors, the key question is whether this pricing power is sustainable. The answer depends on how much new capacity comes online and when. TSMC is building new fabs in Arizona, Japan, and Germany, but those facilities won't produce advanced-node wafers at scale until 2028 at the earliest. Samsung's foundry expansion in Taylor, Texas, has faced delays. In the meantime, AI chip demand continues to grow: Nvidia alone is expected to ship more than 3 million AI GPUs this year, each requiring multiple advanced-node wafers.

Nvidia shares trade at about 35 times forward earnings, a premium that reflects its dominant position in AI computing. TSMC trades at roughly 20 times forward earnings, a discount that may narrow if the pricing increases sustain margin expansion. Samsung's foundry business, valued as part of a conglomerate with memory and consumer electronics, is harder to isolate but stands to benefit disproportionately given its smaller base.

This article is for informational purposes only and does not constitute investment advice.