AI Demand Spills Over to Mature Nodes as TSMC Leads Foundry Pricing Power Reshaping
Miles Bennett
AI chip demand has spread from GPUs to power management, networking, and sensor chips, tightening both 8-inch legacy and 12-inch advanced capacity simultaneously. UMC, PSMC, and VIS are raising prices in a rare collective move — foundry pricing power is being structurally rewritten.
How did AI demand spill from cutting-edge chips into legacy nodes?
For three years, AI spending concentrated on sub-5nm advanced nodes. By mid-2026, demand has spread to PMICs (power management ICs), MCUs, MOSFETs (power switching devices), sensors, and optical-communication components.
This means → every large AI data center needs not just GPUs but a large supporting cast of "unglamorous" legacy-node chips — the high end is pulling the low end, and the entire chain is tightening.
In plain terms = the GPU is the engine, but an engine cannot run without fuel lines, wiring, and bolts. Those "bolts" are legacy-node chips — and they are now in short supply too.
What gives TSMC the confidence to keep raising prices?
TSMC plans another round of price increases in H2 2026 and 2027 for sub-5nm process and CoWoS advanced packaging, extending a pricing upcycle that began during the pandemic.
CoWoS — a packaging technology that tiles multiple chips together — and HBM (high-bandwidth memory) remain in persistent shortage, forming the core bottleneck of this supply crunch.
This reflects TSMC's largest customer, Nvidia, accelerating mass production of the Vera Rubin platform. The stronger downstream demand grows, the greater TSMC's pricing leverage becomes.
Why are second-tier foundries also daring to raise prices?
UMC notified customers of a price increase effective July 2026 — a rare proactive hike for a second-tier fab. PSMC issued a logic-line price increase in Q1; the revenue uplift is expected to show fully in Q2.
The backdrop: TSMC is gradually trimming 28–90nm capacity on its 12-inch lines to redirect capital toward advanced nodes. The customer orders freed up naturally flow to UMC, VIS, and PSMC.
This means → TSMC is "yielding" legacy-node share, and second-tier fabs are catching those orders. The supply-demand balance has flipped — real pricing power is landing in second-tier hands for the first time.
Why is VIS the most direct beneficiary?
VIS (Vanguard International Semiconductor) chairman Fang Lüe says the 8-inch foundry market will remain undersupplied throughout 2026, with visibility of three to five months and strong H2 demand.
Q2 utilization has recovered to nearly 90%; AI-related revenue is approaching double-digit share and still climbing.
VIS has also added a silicon interposer line — a "bridge" layer connecting multiple chips — at its Singapore VSMC fab, signing a long-term contract with TSMC to enter the advanced-packaging supply chain.
What is "silicon inflation," and how does it differ from pandemic-era shortages?
Since H2 2025, AI-driven structural supply gaps have triggered "silicon inflation" — lead times for fiberglass cloth, copper-clad laminates (CCL), ABF substrates, packaging materials, and ultra-pure chemicals have all stretched significantly.
In plain terms = the pandemic shortage was "trucks stuck on the road." This time the problem is "not enough factories" — the constraint is capacity itself, not logistics, so it lasts longer and cuts deeper.
HBM production is crowding out DRAM lines, pushing DDR5 and DDR4 memory prices sharply higher. Cost pressure is propagating from upstream suppliers all the way to end products.
Can this pricing cycle last? What is the key variable?
PSMC ended ten consecutive quarters of losses in Q1 2026, returning its core business to profit. Future earnings growth will come mainly from higher average selling prices (ASP), not capacity expansion alone.
UMC is advancing specialty processes and advanced packaging in parallel; its 12nm project with Intel is expected to reach volume production in H2 2027. A thin-film lithium niobate (TFLN) platform — a key material for next-generation high-speed optical communications — is also under construction.
This means → whether the pricing upcycle holds depends on one core variable: can the pace of AI infrastructure investment sustain the structural improvement in legacy-node demand? That is the litmus test for whether second-tier foundries' newfound pricing power is here to stay.
Content is for reference only, not financial advice.