TSMC accelerates CPO deployment upgrade, ABF substrate supply and demand may face renewed tension

The surging demand for high-speed transmission in AI data centers is rapidly pushing the semiconductor supply chain competition to a new battlefield.

According to Taiwan's Economic Daily News on Monday, TSMC recently revealed that its "COUPE on Substrate" solution, which combines the Compact Universal Photonics Engine (COUPE), is expected to enter mass production in the second half of 2026.

The industry interprets this as a key sign that AI chip manufacturing is moving from "advanced process + advanced packaging" to a new stage of "co-packaged optics + system integration," with the focus of competition further extending to the ABF substrate and co-packaged optics integration segments.

As Nvidia's next-generation Vera Rubin platform continues to enhance AI GPU, HBM, and high-speed network interconnection integration, Nvidia's next supply chain bottleneck is likely to be ABF substrates—by means such as framework agreements, advance payments, strategic partnerships, or even equity investments—to avoid repeating the previous supply crunches with CoWoS and HBM. Market expectations are that, with the simultaneous surge in demand for AI GPUs, ASICs, and CPOs, high-end ABF substrates may once again face a supply shortage next year.

TSMC's CPO Extends to Substrate Level, Marking a Technological Upgrade

TSMC's move to extend the COUPE solution to the substrate level signifies a major evolution in the technological roadmap for AI chip packaging. Supply chain sources note that as optical components get closer to the compute chips, the shorter data transmission distance can simultaneously reduce signal loss, latency, and power consumption — a crucial factor as AI servers shift from single-machine performance competition to full-rack and large cluster efficiency races.

Nvidia recently also announced a deepened partnership with Corning, expanding its optical component supply chain for AI data centers to secure future high-speed optical interconnect capacity. Analysts believe these moves point in the same direction: as cloud service providers (CSPs) continue to pursue higher rack efficiency and lower power consumption, critical links such as wafer foundry, HBM, optical fiber, optical modules, and ABF substrates may all become strategic resources for large AI companies to secure capacity in advance.

Structural Surge in ABF Substrate Demand, Consumption Several Times That of Traditional CPUs

Supply chain sources revealed that compared to traditional CPU substrates, the required area and number of layers for AI GPU and ASIC substrates have increased dramatically, boosting ABF material consumption by 5 to 10 times. With continual rises in demand for AI GPUs, ASICs, and high-end networking chips, the supply-demand structure for high-end ABF substrates could remain tight for the long run.

Media analysis suggests that Nvidia may preemptively lock in high-end substrate capacity through long-term agreements, advance payments, or even strategic collaborations to mitigate supply chain risks. This expectation is consistent with Nvidia's prior strategies to secure capacity in CoWoS packaging and the HBM memory sector.

Suppliers of high-end ABF substrates such as Kinsus, Unimicron, and Nanya PCB are expected to benefit from these demand trends. Among them, Kinsus—which has already entered Nvidia's Vera Rubin platform supply chain—may see especially pronounced benefits. The industry anticipates that as CPO gradually becomes the mainstream architecture for future AI servers, the strategic importance of high-end ABF substrates will continue to rise, and the capacity and technical deployment of relevant manufacturers may become key bargaining chips in supply chain negotiations.