AI Server Demand Surges: Traditional Packaging Capacity May Face Over 20% Shortfall by 2027
Nashnova编辑部
Multiple OSAT firms are warning clients that legacy wire-bonding capacity could fall more than 20% short of demand by 2027 — the bottleneck is not the chips themselves but the packaging equipment that wraps them.
AI servers need far more than GPUs — why is legacy packaging running short?
An AI server requires CPUs, network chips, power-management ICs (PMICs), controllers, and dozens of peripheral components alongside GPUs and HBM.
Most of these chips do not need advanced packaging like CoWoS. They still rely on wire bonding — a mature process that connects chips to substrates with fine metal wires.
This means → every additional AI server deployed multiplies demand for legacy packaging. The pressure point is not the most expensive chip but the high-volume supporting components around it.
Why is the gap only surfacing now?
Legacy packaging demand was flat for years. OSATs shifted investment toward advanced packaging and kept purchases of traditional wire-bonding equipment conservative.
AI has now reactivated demand for mature-node chips. OSATs want to ramp utilisation on legacy lines but find equipment additions cannot keep pace.
Memory, telecom, and industrial-control markets are also recovering, consuming mature packaging capacity in parallel — advanced and legacy packaging demand are rising at the same time, creating a double squeeze.
Where exactly is the equipment bottleneck?
The wire-bonding equipment market is highly concentrated. Kulicke & Soffa (K&S) and ASMPT together control the vast majority of global supply.
In plain terms = OSATs have accumulated years of process parameters and production know-how on their existing K&S and ASMPT machines. Switching suppliers means re-qualifying processes, re-certifying with customers, and retraining engineers — plus retooling maintenance, spare parts, consumables, and automation systems.
This reflects a strong vendor lock-in in legacy packaging — there is virtually no short-term alternative, and expansion pace depends entirely on how fast K&S and ASMPT can ship.
What is K&S itself doing?
K&S announced that Raj Talluri, former senior vice president of Micron's mobile business unit, will become president and CEO on September 1, 2026.
In recent years K&S has extended its business from traditional wire bonding into thermo-compression bonding (TCB), fan-out packaging, and chiplet-based advanced packaging.
This means → K&S's own strategic gravity is shifting toward advanced packaging. Releasing more legacy wire-bonding capacity is not its top priority, which could tighten supply further.
What does this mean for chip designers?
For IC design houses, the resource battle has expanded from wafer capacity and advanced packaging to legacy packaging.
Whether a company can lock in sufficient wire-bonding capacity ahead of 2027 is becoming a key variable in its ability to deliver on schedule.
Put simply = the queue used to form at TSMC's wafer fabs and CoWoS lines. Now even the most "commodity" packaging step requires booking well in advance.
市场有风险，内容仅供研究参考，不构成投资建议。