Packaging, not silicon, is the choke point strangling AI GPU supply, and TSMC is reportedly building a workaround with substrate maker Kinsus called quasi-EMIB. The name nods at Intel's bridge trick, where tiny silicon links sit inside the substrate and let neighboring chiplets talk fast without a giant interposer underneath.
CoWoS carries the AI accelerators that NVIDIA, AMD, and the rest depend on, and orders keep outrunning the lines. Wafers can be sitting ready while packaging slots stay full.
Fewer steps, quicker turnaround, cheaper output, that is the pitch. Neither company has coughed up bridge dimensions, interconnect density, process details, or customer names, and quasi-EMIB reads like an internal label rather than something shipping with a logo.
Whether it backs up CoWoS on lighter products or eventually goes after the same top-shelf accelerators, nobody knows. Kinsus builds ABF and BGA substrates, which fits neatly if the plan involves burying interconnect inside organic material.
More packaging capacity means more AI GPUs escaping into the wild, though gamers should expect any trickle-down to arrive slowly.
CoWoS carries the AI accelerators that NVIDIA, AMD, and the rest depend on, and orders keep outrunning the lines. Wafers can be sitting ready while packaging slots stay full.
Fewer steps, quicker turnaround, cheaper output, that is the pitch. Neither company has coughed up bridge dimensions, interconnect density, process details, or customer names, and quasi-EMIB reads like an internal label rather than something shipping with a logo.
Whether it backs up CoWoS on lighter products or eventually goes after the same top-shelf accelerators, nobody knows. Kinsus builds ABF and BGA substrates, which fits neatly if the plan involves burying interconnect inside organic material.
More packaging capacity means more AI GPUs escaping into the wild, though gamers should expect any trickle-down to arrive slowly.
