TSMC to Raise Chip Prices by Up to 10% in 2027 as AI Demand Heats Up — Premium for Extra HPC Capacity Could Push Some Bills 25% Higher
TSMC, the world's biggest contract chipmaker, plans to hike its chipmaking prices by up to 10% across the board starting in 2027, according to a Nikkei Asia report that hit the wires Wednesday. The move comes as insatiable AI demand, rising tool costs, and a $52 billion-plus annual capex bill squeeze the foundry's margins.
Nvidia, AMD, Apple, and Qualcomm all rely on TSMC for their most advanced chips, and the price adjustment will hit every corner of the semiconductor industry. Customers that need extra high-performance computing capacity beyond their original volume commitments face an even steeper surcharge — another 10% to 15% on top of the baseline increase. For more on how this impacts the broader chip ecosystem, check our Semiconductors coverage.
The Taiwan-based giant told customers in June that baseline prices for advanced nodes — anything 7nm-class and below — would climb 5% to 10% depending on the specific node and the client's volume. For clients who need extra HPC wafer capacity beyond what they originally reserved, there's a steeper penalty: an additional 10% to 15% premium on top of the standard increase. That means a big AI-chip customer like Nvidia or AMD could see its per-wafer bill jump by as much as 25% for overflow orders.
The Price Ripple Across the Semiconductor World
The hike is not limited to bleeding-edge nodes. TSMC also told customers it plans to raise prices on mature technologies — its 12nm, 16nm, 28nm-class nodes and other legacy processes — by up to 10%, though some nodes will see smaller adjustments. Since advanced technologies accounted for roughly 77% of TSMC's Q2 2026 revenue and mature nodes made up the remaining 23%, the price increase effectively touches every part of its business.
The timing is strategic. According to Nikkei, TSMC discussed the pricing changes with clients starting in June and wrapped up negotiations in July. Rather than hitting customers with an immediate increase, the company opted to implement the new pricing from the beginning of 2027, giving clients like Apple, AMD, Nvidia, Qualcomm, and MediaTek time to adjust their own product pricing and supply chain planning.
This wont be a one-off. Vanguard International Semiconductor has already raised prices, and United Microelectronics Corp (UMC) started implementing its own increases in July. Memory makers have also jacked up prices in a big way — SK Hynix, Samsung, and Micron are all riding the HBM and DDR5 wave to record margins, something that hasn't escaped TSMC's attention.
"I am really jealous about memory companies' 86% gross margin," TSMC's CEO C.C. Wei told analysts during the company's Q2 earnings call. "86% [margin at memory makers] — 68% [margin at TSMC], I will be happy about that."
What This Means for the AI Chip Supply Chain
TSMC is the sole manufacturer of the most advanced AI accelerators on the planet. Nvidia's Blackwell and next-gen Rubin GPUs, AMD's MI300 and MI400 series, and most of Apple's A-series and M-series processors all come off TSMC's N3, N4, and N5 lines. A price increase at this level cascades through the entire electronics industry — from hyperscale data-center builds to consumer smartphones.
Wei stressed the company has no plans to gouge its customers. "So we do not suddenly increase our price by which I like to have 4x or 5x," he said. "You cannot survive for that kind of price increase for your customer. So we earn our value, and we make sure that our profit, our gross margin, is enough for our long-term sustaining expansion, that is to the benefit of my customers and TSMC also, that is our philosophy."
Still, the price increase comes at a delicate moment. Wall Street has been jittery about AI capex spending all year — the Philadelphia Semiconductor Index wiped out $1.3 trillion in a single week back in early July as investors questioned whether the AI buildout would deliver promised returns. Higher chip prices could reignite those fears, especially if hyperscalers like Microsoft, Amazon, Google, and Meta pass their increased infrastructure costs back to shareholders.
On the other hand, TSMC's pricing power reflects a simple reality: it has no real competition at the leading edge. Samsung Foundry has struggled with yield on its 3nm GAA process, and Intel Foundry is still years away from being a credible alternative for most TSMC customers. Despite Intel's 18A process touting backside power delivery and gate-all-around transistors, no major external customer has publicly committed to volume production on it yet.
TSMC's Billion-Dollar Bet on Expansion
TSMC isn't just asking customers to pay more — it's spending more too. The company raised its 2026 capital spending forecast to between $52 billion and $56 billion, up nearly 40% from the previous year, as it builds out new fabs in Arizona, Japan, and Germany while upgrading its Taiwan-based GigaFabs for 2nm and 1.4nm production.
The foundry earlier this month reported a record second quarter, with revenue surging more than 40% year-over-year driven by AI chip demand. June revenue alone hit a record high, up 68% compared to the same month in 2025. The company also unveiled its technology roadmap through 2029, announcing plans for A12 and A13 process nodes alongside its N2P and N2U offerings.
TSMC's decision to push a chunk of its price increase to 2027 rather than implementing it immediately suggests the foundry is balancing near-term goodwill against the reality of rising costs. EUV toolmaker ASML has been pushing its own price increases — Low-NA EUV machines reportedly get more expensive under ASML's new value-based pricing model — and TSMC needs to pass those costs along to keep its 53%-ish gross margin target intact.
For the broader semiconductor ecosystem, the message is clear: everything is getting more expensive. Chip packaging substrate prices are up. Silicon wafer costs are climbing. Engineering talent is scarce and expensive. Even the construction of new fabs has gotten pricier as cleanroom buildout costs rise across the board. The years of steadily declining cost per transistor — the semiconductor industry's secular superpower — appear to be over at the leading edge.
The timing of TSMC's announcement is also notable because it comes just as Taiwan's government is weighing tighter export controls on advanced chipmaking technology. Taiwan's N-2 rule, which restricts offshore deployment to nodes two generations behind domestic leading-edge production, could limit how much of TSMC's newest capacity ends up abroad — meaning the fabs in Arizona and Kumamoto may not get access to TSMC's latest 1.4nm or 1.2nm processes when those come online later this decade. Our Robotics and Drones coverage explores how chip supply constraints affect autonomous systems, and Cloud and Edge Computing covers the data-center implications.
TSMC's customers, especially the hyperscalers and top-tier chip designers, have the margins to absorb these increases for now. But smaller fabless chip companies that rely on TSMC's mature nodes — automotive MCU makers, IoT chip designers, display driver manufacturers — will feel the pinch more acutely. Their products run on razor-thin margins, and a 10% wafer price hike could compress profitability to the point where consolidation becomes inevitable.
Some analysts predict TSMC's price hike could accelerate the shift toward chiplet-based designs, where customers mix and match dies from different process nodes to optimize cost. If premium 3nm wafers get 10% more expensive, it makes more sense to keep less performance-critical blocks on cheaper N6 or N12e silicon and integrate them via advanced packaging like CoWoS or SoIC.
TSMC's own CoWoS roadmap, detailed at its recent technology symposium, calls for more than 14-reticle packages capable of 48x the compute power of current-generation designs by 2029. Those packages will use staggering amounts of HBM5e memory — 24 stacks per GPU — which itself is getting more expensive as memory makers dramatically raise prices. The cost of an AI accelerator from wafer to finished package could double within three years if all these trends hold.
For now, the industry is watching to see whether TSMC's customers push back on the pricing or accept it as the cost of doing business with the world's only high-volume leading-edge foundry. Given the lack of alternatives and the unrelenting demand for AI compute, the latter outcome seems far more likely. As one industry executive put it recently: "You don't walk away from the only game in town."
Keywords: TSMC, Taiwan Semiconductor Manufacturing Company, chip price increase, semiconductor foundry, AI demand, C.C. Wei, advanced nodes, N3, N2, wafer pricing, HPC, CoWoS, chip industry supply chain, ASML, capex, semiconductor shortage
Creator: Tech Desk