Late at night, flipping through the Q1 earnings report of AMC—a nearly anonymous materials supplier in TSMC’s supply chain—I almost laughed out loud. Revenue up 170% year-over-year, all thanks to the gritty, unglamorous work inside AI advanced packaging. This isn’t just the semiconductor industry anymore; it’s a meticulously orchestrated hunt.
AMD’s CEO recently met TSMC’s CC Wei in Taiwan. On the surface, it’s about capacity coordination. In reality, Lisa Su is hedging her bets for the MI300 series. NVIDIA dominates with CoWoS, but TSM’s line is booked solid through 2026 and beyond. Su can’t afford to put all her eggs in one basket—hence the quiet backing of Taiwan’s EFB (Embedded Fan-out Bridge) packaging ecosystem. It’s a shrewd move: avoids direct confrontation with TSMC while nurturing an alternative. But can EFB truly meet the bandwidth and thermal demands of next-gen AI chips? Or is it merely a stopgap?
Don’t fixate on fabs. The real storm is brewing in packaging. BOE and Corning’s joint push into glass substrates isn’t coincidental. Organic substrates are hitting physical limits under high-frequency, high-density AI workloads. Glass, with its near-silicon coefficient of thermal expansion, superior flatness, and scalability to large panel formats, suddenly turns decades of display manufacturing expertise into a lifeline for chip packaging. Lam Research’s new PLP (Panel Level Packaging) center in Salzburg may seem bold—but it’s actually a calculated bet on this shift. Still, don’t get carried away: glass is brittle, yields are stubborn, and the equipment ecosystem is virtually nonexistent. Mass adoption? Not before 2027 at the earliest.
Samsung’s quiet maneuvers in Taiwan are even more telling. While pushing HBM4, it’s also in talks with local OSATs to bundle memory and logic into an integrated “Foundry + Memory” offering. Everyone knows who the target is: TSMC. By monopolizing CoWoS, TSMC controls the golden interface of AI chips. Samsung hopes vertical integration can crack that dominance. But will customers really hand both logic and memory to a single vendor—especially one that’s also their competitor?
Nanya Technology’s plan to ramp a new fab by 2027 is straightforward: the AI memory crunch is real. Yet DRAM isn’t logic—it’s capital-intensive, slow to scale, and perilously vulnerable to demand swings. Remember 2000? When the dot-com bubble burst, DRAM makers bled out in a sea of overcapacity. History doesn’t repeat, but it often rhymes.
Then there’s InPsytech, a little-known startup that just hired a former Altera and Intel executive as senior strategy adviser. Global expansion, they say. But in today’s fractured geopolitical climate, any “global” narrative feels naive—unless they’ve cracked a packaging technology that sidesteps U.S. export controls. Otherwise, it’s just storytelling with a different accent.
Twenty years ago, we debated 90nm vs. 65nm. Ten years ago, FinFET versus FD-SOI. Today, the battlefield has shifted below the die—to the package. With Moore’s Law slowing, performance gains now hinge less on transistor scaling and more on system-level integration. Whoever controls advanced packaging defines the architecture of AI itself.
TSMC knows this all too well. That’s why it guards CoWoS like a dragon hoarding gold—even at the cost of margin compression. But monopolies aren’t eternal. Intel once ruled the PC era with its IDM might, only to be toppled by the TSMC-ARM alliance in mobile. Could TSMC now face its own “innovator’s dilemma” in the packaging revolution?
Glass substrates, panel-level packaging, EFB, silicon photonics—the technical paths are proliferating. But the winner won’t necessarily be the most advanced; it’ll be the one that best balances cost, yield, and supply chain resilience. Chinese panel makers bring scale, U.S. firms lead in design, Taiwan excels in manufacturing, and Korea holds memory. This packaging war is, at its core, a global supply chain reshuffle.
So stop obsessing over ASML’s EUV backlog. The real inflection point may lie in a sheet of glass thinner than a human hair. While everyone scrambles for wafer capacity, the smart money is already building the next packaging ecosystem. The question isn’t who will win—it’s who will become the next forgotten “AMC,” and who the next disrupted “Intel.”