TSMC Weighs 20% CoWoS Price Hike as AI Chip Packaging Tightens

| 2 Min Read
TSMC is reportedly weighing higher CoWoS prices as complex AI chips, longer test cycles and delayed equipment deliveries strain packaging capacity.

Advanced packaging is becoming a more expensive constraint on AI chip production. TSMC is reportedly considering raising CoWoS prices by as much as 20%, according to Trendforce, citing supply chain sources and local media. TSMC hasn’t confirmed the report, so the proposed increase remains unverified. If it goes ahead, it would put a higher price on a production step that’s already struggling to keep pace with demand. That’s a sign that scarce capacity can command a premium even while manufacturers spend heavily to expand it.

More compute means a harder packaging job

CoWoS places logic dies and high-bandwidth memory (HBM) in one package, giving AI accelerators the bandwidth their workloads demand. That integration also makes packaging part of the product’s performance equation: the logic and memory need to work together closely, and the package has to connect them at scale. But each generation packs in more compute and memory, increasing the complexity of assembly and the demands on production. Nvidia’s Blackwell combines two large GPU dies with eight HBM3E stacks. Rubin and Rubin Ultra are expected to push integration further, adding pressure to packaging and the materials supply chain that supports it.

Testing compounds the capacity problem. Blackwell reportedly takes three to four times longer to test than the previous-generation Hopper, while Rubin could require up to seven times as long. Those are reported estimates, but they point to a practical issue: capacity depends on how quickly products move through each stage, not simply on how many tools are installed. Longer cycles tie up equipment and slow throughput just as demand rises. Server CPUs and custom ASICs also need advanced packaging, so accelerators aren’t the only products competing for limited capacity. A price increase would therefore reflect a crowded production queue, not just the cost of assembling one class of AI chip.

Equipment access is taking priority

Lead times for chipmaking tools have reportedly lengthened from three to six months to as long as eight. TSMC is said to have placed this year’s equipment orders at current prices, skipping its usual negotiations for discounts or added features. That’s a revealing trade-off: when delivery timing is uncertain, a lower purchase price matters less if the equipment arrives too late to support planned output. Securing a production slot can be the more valuable outcome.

The company’s equipment needs have surged alongside its production plans. A September report said projected requirements had nearly doubled in six months. By July, expected demand had reached 1.9 times TSMC’s original estimate, according to deputy co-COO Cliff Hou. “Compared with the pattern 30 years ago, we have never seen the demand increase that much, that frequency of changes,” Hou said. His comment describes more than a large order book: frequent changes make it harder to plan purchases and capacity around a stable forecast. TSMC’s willingness to accept current equipment prices suggests the cost of waiting has become a central part of its calculations.

New supply takes time to arrive

TSMC is widening its packaging supply chain, but new capacity won’t arrive immediately. Earlier this month, GlobalFoundries agreed to make silicon interposers for TSMC’s CoWoS in the U.S. under a five-year, $2 billion contract. Production is expected to ramp in the first half of 2028. The agreement gives TSMC another source for an important component, but interposers are only one part of the packaging process. Capacity across the other steps still has to keep pace. This deal alone won’t resolve near-term constraints.

The reported packaging increase may be part of a broader rise in chip costs. TSMC is also said to be considering 2027 wafer price increases of 5% to 10% for advanced process technologies. Customers using 2nm and 3nm-class nodes could see the largest increases; orders beyond existing allocations may carry extra premiums of up to 25%. These are separate reported pricing changes, and neither should be treated as confirmed company policy. Taken together, though, they suggest that customers may face pressure at multiple stages: making the silicon, packaging it, and securing capacity beyond current allocations.

What the bottleneck means for AI chip supply

Industry estimates cited by Trendforce put TSMC’s monthly capacity above 106,000 wafers in 2026, with a target of 227,000 in 2027. Those figures point to a major production expansion, but wafer output alone won’t determine how many finished AI chips reach customers. A wafer can be produced without resolving the separate limits imposed by assembly, materials, and testing. This is more significant than it looks: a factory can increase its wafer capacity and still see finished-chip output held back downstream.

For customers, that distinction matters because more compute demand doesn’t translate directly into more available accelerators. If you’re working in this space, the practical question is whether packaging and test capacity will grow quickly enough to match the new silicon supply. And yet, higher reported prices don’t prove that supply will remain tight indefinitely; they show that TSMC may be pricing a current constraint while trying to expand the systems around it. Here’s the thing: the cost of scaling AI hardware increasingly depends on how well those production stages move together. For now, the pinch point. If the reported increases materialize, more of that cost could flow through the supply chain.

Source: Newton Gitonga · www.tomshardware.com

Comments

Please sign in to comment.