StatementDog | AI & Tech Supply Chains

If you follow $LITE and $COHR, the AI optics buildout also runs through 17 Taiwan-listed companies. We mapped them across five layers and rated how directly each one benefits. Supply is tightest in laser materials today, while CPO parts are mostly a 2027 and 2028 revenue story. Top tier: LandMark (3081 TT) VPEC (2455 TT) EZconn (6442 TT). LandMark and VPEC grow the InP and GaAs epitaxial wafers that laser and photodetector chips are built on. LandMark has a four-year agreement with a US customer for data center CW lasers covering 2027 to 2030. WIN Semiconductors (3105 TT) and IntelliEPI (4971 TT) sit in the same layer, with less of their revenue tied to AI optics. Modules and laser packaging are where 2026 demand shows up first, because pluggable 800G and 1.6T modules remain the volume product. LuxNet (4979 TT) began volume production of 800G products this year and expects 1.6T NPO mass production in the second half of 2027. Elite Advanced Laser (3450 TT) mounts laser chips onto submounts for module makers. Fiber passives are benefiting from denser data center cabling now. EZconn makes high fiber-count trunk cables and MPO cords. High fiber-count products passed half its sales in 1H26, when gross margin reached 59.5%. FOCI (3363 TT) is developing fiber array units for CPO and has told local media it expects shipments to pick up clearly in 2027. Optical module boards need mSAP HDI from 800G up. Zhen Ding (4958 TT) and Compeq (2313 TT) both make them. A US broker expects the mSAP market for 1.6T and faster modules to grow from $0.8B in 2026 to $2.9B in 2028. Accton (2345 TT) builds AI data center switches and also sells optical transceivers. How much it benefits depends on AI networking’s share of its revenue. What we’re watching closely: -How fast 1.6T replaces 800G in 2027 -Whether new InP substrate and CW laser capacity eases pricing -CPO switch volumes, which decide when FAU and CPO parts turn into revenue

After Jensen Huang’s departure from Nvidia, will AI hardware upgrades slow down? Google and AMD were already pushing suppliers in different directions. Google’s TPU 8i Max puts two of its existing TPU 8i chips in one package. Google says that doubles both compute and high-bandwidth memory bandwidth versus the 8i, within the same rack footprint. AMD’s Helios shows where Taiwan fits. Server makers Wiwynn, Wistron and Inventec are helping build systems around AMD’s MI450X GPUs. Another Taiwanese supplier, AIC, is working on the mechanical design of the rack and the compute trays that slide into it. For investors following Taiwan’s server supply chain, that last detail matters. AMD’s design decisions are already shaping what Taiwanese engineers build, down to the trays that hold the hardware.

[Taiwan’s Weekly Standout #4] Taiwan Glass (1802 TT) reported a 56.5% jump in October revenue to NT$5.57 billion, citing stronger glass-fabric demand from AI and other smart technologies. Its shares rose 47.2% between the Nov. 6 and Nov. 13 closes. The company makes glass yarn and fabric used in electronic materials. Two product families connect it to AI hardware. Low-DK glass fabric helps reduce signal loss in high-speed circuit boards. Low-CTE glass fabric limits thermal expansion in chip-package substrates, helping reduce warping as the chip and substrate expand by different amounts when heated. We estimates that October revenue from high-end glass fabric rose about 14% to 17% from September. The two product families are ramping at different speeds. Low-CTE shipments began in small volumes in late 2025 and have increased substantially since early 2026. Low-DK2 is expected to contribute more revenue than Low-CTE in 2027, with Low-CTE narrowing the gap as new capacity comes online in the second half of 2027 and 2028. On Nov. 2, Chairman Lin Po-feng said high-end glass-fabric supply was still falling short of demand. Taiwan Glass is targeting 2027 completion and production for its third expansion phase and has launched a fourth, according to Economic Daily News. Our read is that Low-DK2 shipments for high-speed circuit boards matter more to near-term growth, while Low-CTE for package substrates offers a longer ramp.

Taiwanese here. A lot of Nvidia's server, rack, cooling and power suppliers in Taiwan started as old-economy family businesses. Foxconn began in 1974 making plastic knobs for TVs. That world still runs on ties built over years of dinners and drinks, and many of the strongest ones sit with Jensen personally. Getting that many supply chain bosses around one table isn't easy, and Jensen had the pull to do it. I think it does make a difference when the CEO across the table isn't familiar with Taiwanese culture. When a problem spans several suppliers, someone still has to get every decision-maker in the room, and that person is now gone. Local analysts don't see every supplier as equally exposed. $TSM, $ASX and MediaTek (2454TT) were called irreplaceable, since any AI chip still needs them, whether for wafers, packaging or design. The worry is the next tier down, where several Taiwanese vendors can do the same job and relationships help decide who gets the order. One Taipei analyst called it a risk of orders getting "poached." I don't think this supply chain can be replaced anytime soon. Any lost Nvidia orders would likely shift to other customers, and the same suppliers already build for $AMD and custom ASIC programs.

Jensen Huang has resigned as CEO of Nvidia

Google’s TPU v9 could use fewer PCB layers than Amazon’s Trainium 4 while requiring more HDI build-up steps. We think total layer count alone would understate the manufacturing work involved. Taiwan sell-side research points to a 3-28-3 HDI design for Trainium 4 in Q3 2027 and a 6-12-6 design for TPU v9 in Q4 2027. That works out to 34 total layers for Trainium 4 and 24 for TPU v9. But TPU v9 would have six HDI build-up layers on each side of its inner stack, versus three for Trainium 4. Those outer layers are added through sequential lamination, with laser-drilled microvias connecting them. More build-up layers add processing steps and place greater demands on alignment and interconnect reliability. For AI PCB suppliers, the two designs would create different production requirements. High-layer-count capability and the ability to reliably manufacture deeper HDI structures both matter as cloud providers develop their next custom chips. $GOOG $AMZN

Morgan Stanley expects niche-memory pricing power to weaken into 2028. Its revised model puts DDR4 supply 1% above demand that year. The bank has shifted about 5GW of expected US data-center additions from 2028 to 2029, roughly 20% of its previous estimate for US additions in 2028. That pushes some expected enterprise demand further out. Consumer demand is also weakening. Morgan Stanley points to PC procurement brought forward into the first three quarters of 2026, followed by a deeper correction in Q4. Earlier orders may have overstated underlying consumption. Niche-memory suppliers benefited from having fewer fixed-price long-term agreements because they could raise prices faster as shortages worsened. Those arrangements also provide less protection when customers start resisting further increases. The bank still models shortages in 2028 for SLC NAND, MLC NAND and NOR, at 3%, 5% and 1%, respectively. Its concern is that smaller shortages will make additional price increases harder to sustain. This call covers DDR3/4, NOR and SLC/MLC NAND, excluding mainstream DDR5 and TLC/QLC NAND. For Winbond (2344 TT), Morgan Stanley raised its 2026 EPS forecast by 8%, while cutting 2027 and 2028 estimates by 9% and 22%.

Double-sided cold plates are already shipping at AI cooling supplier Asia Vital Components (3017 TT). The company says they command higher prices and gross margins than single-sided designs. We see power-delivery upgrades translating into higher-value cooling products. Vertical power delivery places power components beneath the processor, shortening the path electricity travels. Those components also generate heat, creating a need to cool the underside of the board as well as the processor above it. AVC says it is working with power-supply manufacturers on these designs. Pricing and margins vary by customer configuration, while some GPU customers have already adopted double-sided cooling. For cooling suppliers, that creates a source of growth beyond server shipment volumes. A change in how the same system receives power can increase the cooling hardware it needs and the value of the supplier’s delivery.

Inventec (2356 TT) expects to ship more complete custom-chip servers in 2027. We see a larger revenue opportunity per shipment, alongside the margin pressure that can come with delivering more of the hardware. The server manufacturer said at its November 10 briefing that its custom-chip programs span barebones systems and complete servers, with more complete-server shipments expected next year. Its NVIDIA Vera Rubin work extends to rack and multi-rack cluster integration. Those roles involve combining systems, networking and other infrastructure into larger working configurations. The assembly scope matters when interpreting growth. Delivering a more complete system brings more components into the sale, so revenue can rise through both shipment growth and the value of hardware included. Inventec said a higher share of complete-rack shipments lifted Q3 revenue but weighed on gross margin. $NVDA $AMD $GOOGL

ASIC-related server revenue at cooling supplier Sunon rose about 48% sequentially in Q3. We noticed a stronger component-sales signal, while complete-rack deliveries still require separate checks. Sunon (2421 TT) makes cooling fans and other thermal-management products. KGI reports that its server revenue rose from NT$2.33bn in Q2 to NT$3.08bn in Q3, with ASIC AI servers increasing from 51% to 57% of that revenue. ASIC-related server revenue is approximately NT$1.19bn and NT$1.76bn. Separately, at chassis and rack supplier Chenbro (8210 TT), Morgan Stanley says qualification of an ASIC customer’s new product continues to weigh on Q4 business momentum. These are separate supplier readouts. Sunon’s sales show revenue being realized in cooling components; Chenbro’s qualification process highlights another condition that must be met before systems can ship. We would track both when assessing the pace of an ASIC server ramp.