TSMC Posts Record $16.35 Billion August Revenue, Preps Up to 15% Price Hikes

Taiwan Semiconductor Manufacturing Co. posted revenue of NT$514.81 billion ($16.35 billion) for August, up 53.3 percent from a year earlier and 10.1 percent from July, the company said in its monthly filing on Sept. 10. The same scarcity that produced that number is the reason TSMC can raise prices on the customers who have nowhere else to go.

August was TSMC's fourth straight monthly revenue record, CNBC reported, citing the company's own filing and market-share data from the research firm TrendForce. January-through-August revenue reached NT$3,386.87 billion, up 39.3 percent from the same period in 2025. TSMC's second-quarter profit rose more than 77 percent year over year, and the company has guided third-quarter revenue to between $44.6 billion and $45.8 billion.

Three Nanometers, Zero Slack

The number that matters here is not the revenue figure but the capacity utilization behind it. TrendForce data cited by CNBC put TSMC's second-quarter foundry market share at 72.5 percent, with its 5-, 4- and 3-nanometer lines fully booked. Samsung Foundry held 5.9 percent of the market and China's SMIC held 5.4 percent, neither offering AI customers a comparable node to move volume to.

1. TSMC72.5 %
2. Samsung Foundry5.9 %
3. SMIC5.4 %

Source: TrendForce data cited by CNBC, accessed 2026-09-11

A foundry at full utilization with no qualified second source is a foundry that sets price, not one that takes it. That is the mechanism, not a metaphor: when TSMC's advanced nodes are booked solid, the constraint shifts from TSMC's willingness to sell wafers to Nvidia and AMD's willingness to pay for them, and the willingness to pay is not really a choice when the alternative is missing a product cycle.

The Bill Coming Due

TrendForce reported on May 27 that TSMC is expected to raise 3-nanometer pricing by up to 15 percent in the second half of 2026, with a further 5 to 10 percent increase penciled in for 2027. The research firm tied the increases to 3-nanometer capacity expanding from roughly 130,000 wafers a month to 160,000-175,000, driven by AI server demand rather than the smartphone chips that used to fill the node. Nvidia, AMD, Google and AWS were named as the customers pulling that capacity forward; Broadcom and Marvell were named as the ones using it for custom AI accelerators instead of general-purpose GPUs. The same demand has already shown up downstream: Oracle told investors it tripled its GPU deliveries to 300,000 units, and EuroHPC's $450 million AMD-powered LUMI-AI deal locked in accelerator supply years ahead of delivery for the same reason: nobody wants to be last in line when the fab that makes the chip is already full. A wafer price increase at the source of that chain does not stay contained to TSMC's ledger; it moves through Nvidia and AMD's cost of goods sold, then into the price Oracle and EuroHPC pay per accelerator, arriving last at whoever is renting GPU time by the hour.

Metric	Value	Period
August revenue	NT$514.81B ($16.35B)	Aug. 2026
Revenue growth	+10.1% month over month	vs. July 2026
Revenue growth	+53.3% year over year	vs. Aug. 2025
3nm price increase	up to 15%	2H 2026 (TrendForce)

A 15 percent wafer price increase does not translate one-to-one into a 15 percent cost increase for a finished GPU, because the wafer is one line item among packaging, high-bandwidth memory and the board itself. But it is a line item with no substitute, which is a different kind of cost than the ones a buyer can shop around.

What TSMC Says It Will Do With the Money

TSMC's capital spending is set to hit the high end of a $52 billion-to-$56 billion range for 2026, and Chief Executive C.C. Wei said on the company's first-quarter earnings call that spending over the next three years would be "significantly higher than the past three years," the South China Morning Post reported on April 16. That capex is going partly toward extending the current nodes and partly toward the next one: TSMC and ASML this week announced they will bring ASML's High-NA lithography tools into large-scale production starting in 2030, CNBC reported, a multi-year runway that assumes AI demand for leading-edge silicon keeps growing long enough to justify it.

What Would Change This Read

The case that this is a genuine margin transfer from chip designers to TSMC rests on advanced-node capacity staying the binding constraint. Two things could loosen it. One is hyperscaler in-house silicon: Google and AWS are already named among the customers pulling 3-nanometer capacity, and both are also the companies with the most incentive to shift workloads onto their own accelerators once those chips are mature enough to reduce Nvidia dependence. The other is TSMC's own guidance; a third-quarter revenue range of $44.6 billion to $45.8 billion implies growth close to what the company has already delivered, not an acceleration, and a forecast that flat would be the first sign that the 2H 2026 price increase is being absorbed through volume rather than passed through as margin pressure. Neither company has said publicly how it plans to treat the increase in its own pricing, which is the number worth watching in the next earnings cycle rather than this one.