AMD

AMD needs little introduction to anyone following AI chips: the Santa Clara semiconductor company that spent the 2010s clawing back CPU market share from Intel under CEO Lisa Su, and has spent the 2020s building out its Instinct line of AI accelerators (the MI300, MI350, and MI400 series) as the most credible alternative to Nvidia's GPUs. Every generation of that accelerator line has pushed further into the same physical wall every chip designer profiled in the "AI demand" stage of this chain is running into: chiplets and memory stacks are getting too big and too hot for the packaging materials underneath them to keep up.

Where AMD stands out in this specific value chain is that it's not just evaluating glass substrates quietly like most of its peers — it's the first company in the industry to put public, specific dates on a glass substrate roadmap.

AMD's roadmap calls for pilot production of glass substrates in 2026, with actual product application following in 2028 — described in industry coverage as the first confirmed glass substrate roadmap anywhere in the semiconductor industry. The plan is to apply glass first to the chip interposer, the layer that sits between compute dies and memory stacks in an advanced package, rather than replacing the entire substrate core on day one. That's the more conservative of the two glass adoption paths available (glass interposer vs. full glass core substrate), and it mirrors the approach Intel and Samsung Electro-Mechanics are also taking as a first step. Industry outlet The Economy was the first to report this as the semiconductor industry's first confirmed, dated glass substrate roadmap from any single company.

Separately, and more concretely in the near term, Korea's ETNews and trade outlet Techbrew have both reported that AMD is in active discussions with Absolics — SKC's glass substrate subsidiary — about an actual supply relationship, with the two sides said to be approaching a "pre-qualification" stage where basic performance and quality of Absolics' glass substrate samples get validated. That's a real, near-term signal distinct from AMD's own longer-range roadmap: it's evidence that AMD is actively sourcing, not just planning.

The technical driver is warpage. As AMD's Instinct accelerators have grown into larger, more complex packages — more compute chiplets, more HBM stacks, wider interconnects — the organic substrates underneath them expand and contract at a different rate than the silicon dies sitting on top when heated during manufacturing and operation. Above a certain package size, that mismatch causes warping severe enough to hurt manufacturing yield and long-term reliability. Glass has a coefficient of thermal expansion much closer to silicon's, which is the whole reason the industry is chasing it — not a cosmetic upgrade, but a fix for a real physical ceiling on how large and powerful a single chip package can get.

AMD isn't chasing this alone or first for its own sake; it's chasing it because Nvidia, Intel, and its own foundry partners are all bumping into the same wall at roughly the same time, and whoever qualifies a working glass supply chain earliest gets to build bigger, more competitive AI packages sooner.

AMD sits in stage 5, chip designers & AI demand — the demand side of this chain, alongside Amazon and (through its own packaging ambitions) Intel. It doesn't make or process glass; it's one of the customers whose chip designs are the reason the rest of this value chain exists at all.

AMD is also profiled in The AI Infrastructure Value Chain — stage 01, AI chip design.

AMD's roadmap matters to the rest of this value chain disproportionately, because it functions as a demand signal every upstream company can point to. When SKC discusses Absolics' path to high-volume manufacturing, when PhilOptics or YC Chem talk about "global customers," AMD is one of the few names analysts can actually attach to that demand with real specificity, precisely because AMD published dates. That's rare in an industry where most customer relationships are covered by NDAs.

It's also a reminder that glass substrate adoption isn't a single event but a staggered rollout across the industry: AMD's 2026 pilot / 2028 product timeline sits roughly in the same window as Absolics' and Samsung Electro-Mechanics' own manufacturing targets, and slightly ahead of Samsung Electronics' 2028 foundry-side rollout. If any one of these timelines slips, it tends to ripple through the read on all the others, since they're all racing toward the same underlying customer base.