TSMC’s 2027 price rise could drive up costs for everything

Wafer prices could rise up to 10% from January 2027, weighted toward the mature nodes inside almost every device.

TSMC has spent two decades trying not to behave like a memory company. Where DRAM and NAND prices lurch with each cycle, doubling in a shortage and collapsing in a glut, the world's largest contract chipmaker has moved in small steps, signalled early and rarely reversed. Customers can budget around it. That predictability is an asset, and it is the frame for what the company is now doing.

Nikkei Asia reported on Tuesday that TSMC plans to raise prices from the start of 2027, citing multiple sources. Base increases run between 5% and 10% depending on the customer and the product. Negotiations opened in June and concluded this month. The stated reasons are unremarkable and entirely believable: materials cost more, tools cost more, and building fabs in Arizona and Kumamoto costs a great deal more than building them in Hsinchu.

What makes the move worth reading closely is that it comes from a position of unusual strength. TSMC held 72.3% of the global foundry market in the first quarter of this year, according to TrendForce, against 6.5% for Samsung, its nearest competitor. Last week it reported second-quarter net profit of T$706.6 billion, roughly US$22 billion, up 77% and a record, on a gross margin of 67.7% that was also a record. This is not a company raising prices because it is struggling.

Its costs are genuinely climbing all the same. TSMC lifted its capital spending guidance for the year to between US$60 billion and US$64 billion, and the fabs it is building in the United States and Japan are structurally more expensive to run than the ones at home. A price increase announced 18 months ahead of the invoice is how the company has always handled that: enough warning for customers to absorb it, and no repeat of the sudden repricing the memory industry has spent the past year inflicting on everyone.

Reuters put the report to TSMC, which as a rule does not discuss pricing. A spokesperson said only that the company's approach is "strategic, not opportunistic". Chief executive C.C. Wei made much the same point to analysts last week. "We don't suddenly increase our price," he said.

The headline number, though, matters less than which chips it covers.

Not the AI chips

Most coverage of a TSMC price move goes straight to the leading edge, because that is where the AI accelerators are made. For most buyers of computing equipment, that part of the story is remote. Almost nobody outside a handful of chip designers negotiates 2nm wafer pricing.

The detail that matters most is the one about mature nodes. Nikkei reported that 12nm, 16nm and 28nm production faces increases of up to 10%, the steeper end of the range.

Those nodes are the plumbing. They produce power management chips, microcontrollers, storage and network controllers, display drivers and the analogue parts that surround a processor and make a finished product work. A server contains dozens of them. So does a switch, a laptop, an access point, a point-of-sale terminal and a car. They are cheap individually and unavoidable collectively, which is why an increase there reaches more product categories than a leading-edge hike ever does.

It also breaks a pattern. For two years mature-node pricing went one direction, pushed down by aggressive Chinese capacity in the 28nm-to-90nm range. That has turned. TrendForce reported at the end of June that foundry prices rose 5% to 15% on average between the first and second quarters of 2026, with further increases of 5% to 10% appearing on capacity-constrained nodes into the third quarter and suppliers aiming for a broader round in 2027.

Because TSMC sets the reference point, the effect runs wider than its own customer list. The same firm has noted that as TSMC trims mature-node capacity and reallocates orders, smaller foundries gain room to lift their own quotes. Tellingly, it also observed that customers were already negotiating to delay the increases scheduled for the second half of this year.

Everyone is pushing the bill downstream

TSMC is not the origin of this. It sits in the middle of it. Last week ASML, which has no real competitor in advanced lithography, signalled that it sees room to charge more. Chief financial officer Roger Dassen told analysts that continued productivity gains on the company's low-NA EUV systems give it "a pretty strong runway for potential price improvements going forward".

The Information reported that ASML has raised the subject directly with TSMC, that TSMC is resisting, and that some Chinese customers have accepted a 10% increase on older DUV tools.

The timing means the two are not cause and effect. ASML's 2027 output is close to sold out and much of 2028 is booked, so any increase mostly touches tools delivered from late 2028 onward.

TSMC is not raising prices next year to cover equipment it has not paid for yet. Both are symptoms of the same thing: a cost base rising at every stage of chip production, with each stage passing on what its bargaining position allows. The largest foundry in the world is refusing a supplier's increase in one meeting and handing its own customers an increase in the next.

That is leverage, and it is a fair guide to how the rest of the chain will behave.

Why a small number bites

Taken alone, 5% to 10% on a wafer is a minor input. A wafer yields thousands of chips, and the silicon is a modest share of what a finished server or notebook costs. The problem is what it follows. Memory has been the far more violent story this year, with DRAM and NAND contract prices climbing at rates foundry pricing has never approached, and PC and smartphone makers already passing that through to retail.

TrendForce expects conventional DRAM contract prices to rise another 13% to 18% in the third quarter and NAND 10% to 15%, and calls that a slowdown, since the second quarter saw jumps of around 60%. So the foundry increase arrives at a point where hardware makers have spent a year absorbing cost and have very little room left to swallow more. It does not need to be large to show up on a price list.

The practical consequence sits in paperwork rather than procurement. Anything signed now that delivers or renews in 2027, whether a staged rollout tendered this year, a multi-year support agreement or a fixed-price service contract, is being costed against a component base that is scheduled to move. Long quote validity windows are generous in a stable market and expensive in this one. Refreshes drifting toward next year without a strong reason are worth pulling forward.

One caveat is worth keeping in view. TSMC has not confirmed the report, base prices are an opening position rather than a settled outcome, and large customers have talked foundries out of announced increases before. Morgan Stanley had already pencilled in a 5% to 10% rise for 2027 before Nikkei's story ran, which suggests the market treats this as expected rather than surprising.

Whether it lands at 5% or 10%, nothing in this chain is getting cheaper.