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

I think in this whole Nvidia situation, there's an important second-order effect that could play out over the next six months. Everyone’s been wondering how Chinese AI models caught up so fast. Besides China’s dense AI talent and its efficiency in model distillation and algorithms, getting advanced GPUs through gray channels for training is probably part of the reason too. But from here on, Chinese companies are bound to face much stricter scrutiny when buying GPUs. On top of that, the US is getting ready to restrict Chinese companies from renting advanced GPUs through overseas data centers. China could gradually fall behind in the next generation of frontier models. That’s bullish for US frontier model makers, including the CSPs that control distribution, and for AI’s long-term development. Chinese models aren’t just closing the gap on tech. Their rock-bottom token prices are squeezing US model makers’ margins. Once China’s catch-up slows, US model makers can hold their tech lead, and their monetization and profitability could improve at the same time.

Thank you @benitoz for taking the time doing the interview with us. Explore the remarkable 3D rack models at BEP Research, and read his research on Substack.

I’ve been hearing similar things, but still need to check with senior executives in the supply chain to confirm. China hawks will regain the upper hand in Washington after the midterms. Restrictions on China’s access to technology and tighter supply chain controls will likely keep expanding, bringing U.S. policy closer to what we saw toward the end of Biden’s term.

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.

Statementdog’s Weekly Take On the Market: Jensen’s Exit, TSMC’s AI Bets and the Cost of Keeping AI Growing Companies mentioned: $NVDA $TSM $CRWV

I think TSMC’s planned investments in OpenAI and Anthropic could give it a clearer view of how much compute its customers ultimately need. Its board approved investments of up to $500 million in each company. One potential benefit is a more direct line to the AI developers whose demand feeds through cloud providers and chip designers into wafer orders. There’s this thing called the bullwhip effect in supply chains. When capacity is tight, companies can order extra to secure supply. If deployment plans later slip, each layer may cut those buffers. A modest change in end demand can then become a much larger drop in orders upstream. Thus, closer ties to the model developers could help $TSM compare their compute plans with the orders coming from chip customers, giving it another input when deciding how much capacity to build.

We expect HDD price increases to moderate, with higher capacity per drive still supporting revenue growth at Seagate and Western Digital into the first half of 2027. HDD makers have historically relied on shipping more storage capacity to offset falling prices per terabyte. AI-driven demand for data-center storage has tightened supply enough for both shipment capacity and prices to rise. Higher-density recording allows them to ship more terabytes without a comparable increase in the number of drives. That gives revenue a source of growth even as price increases become smaller. Our outlook assumes storage demand remains strong enough to absorb that additional capacity. Under those conditions, modest price increases could continue to add to revenue growth. $STX $WDC

[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.

$ASX opened the first phase of its advanced packaging site today. For investors tracking packaging supply in 2027 and 2028, one part of the expansion is already operating. The second phase broke ground today. ASE converted an existing factory for the first phase. It expects total investment across both phases to reach about NT$100 billion and estimates the site could generate NT$80 billion in annual revenue once both phases are running at full capacity. That puts the spending timeline in perspective. ASE has already invested in the first production lines, while construction spending on the second phase still lies ahead.

Nvidia CEO Jensen Huang stepped down, and Gigabyte was the first Taiwanese AI server maker to respond publicly. Nvidia’s ties with Gigabyte (2376 TT) run well beyond the executive suite, the Taiwanese AI server maker said today when asked how Nvidia’s leadership change might affect their partnership. We estimate that 55% to 65% of Gigabyte’s 2026 revenue is tied to NVIDIA, the highest share among the 15 Taiwanese suppliers we reviewed. Gigabyte’s co-president expects the partnership to continue through the leadership change. Gigabyte’s margins also challenge a common assumption that building AI servers squeezes assemblers’ margins. Its company-wide gross margin rose from 9.6% in Q2 to 14.0% in Q3, and its operating margin reached 8.3%. money.udn.com/money/story/56…

Amazon’s Trainium 3 helped lift Taiwanese rail maker King Slide (2059 TT) ’s gross margin to 88.2% in Q3. CLSA says higher-margin rail kits for those servers accounted for a larger share of King Slide’s sales. Where can King Slide hold its pricing? Even within Trainium 3, it depends on which rack Amazon builds. CLSA sees room for new suppliers to compete for the air-cooled racks as early as 2027, putting pressure on rail prices. For the liquid-cooled racks, King Slide is expected to keep its lead. $AMZN

TSMC’s Arizona subsidiary earned NT$54 billion in the first nine months of 2026, including NT$17.9 billion in Q3. Arizona turned a profit in each of the three quarters. For $TSM shareholders, a profitable Arizona subsidiary can still mean a lower group gross margin. Overseas fabs cost more to run. TSMC expects their ramp-up to trim about 2 percentage points from its gross margin in 2026, widening to 3 to 4 points over the next few years as more US fabs come online. Government support has also grown. TSMC subsidiaries in the US, Germany, Japan and China received NT$167.6 billion in grants in the first nine months, up from NT$71.9 billion a year earlier. Those grants help cover equipment, construction and some operating costs across all four markets.

We went to this year’s materials forum at National Cheng Kung University in Taiwan, and here are some takeaways: Changing one material in an AI chip package can force another supplier to rework a design and rerun tests. Several speakers told us that packaging specs are changing so quickly that suppliers need to share design data earlier to leave enough time for testing. Yang Wei Wen, president of IV Technologies, a Taiwanese maker of CMP pads, gave us an example. New packaging materials change what it takes to polish a surface flat. His team already develops processes with customers, and he wants equipment makers involved from the start. Co-packaged optics shows why the timing matters. Companies are still exploring several ways to integrate optics into a chip package. Which design they choose changes what their partners need to test. We will release interview videos with some suppliers, so stay tuned.

AMD and NVIDIA are taking different approaches to cooling their next AI chips. You can see both at Jentech (3653 TT), the Taiwanese supplier that makes the metal lids that draw heat away from the chips. Based on our supply chain checks, Jentech’s two-piece lid for AMD’s MI450 is entering volume production this quarter. For NVIDIA’s Vera Rubin, Jentech has shipped small batches of a removable lid for testing. Management sees volume use as early as Q1 2027. The assembly includes a stiffening frame, a removable lid and screws, which makes machining and testing for flatness more demanding. To meet customer demand for larger lids, Jentech has moved up the planned production start at its new plant from Q3 to Q1 2027. robinhood:0x86923f96303d656e4aa86d9d42d1e57ad2023fdc $NVDA

Amazon’s custom AI chip shipments could reach roughly 3 million in 2027, up about 30% year over year, according to Daiwa. For Accton (2345TT), the Taiwanese supplier that makes accelerator cards for Amazon’s Trainium 3, the shift to liquid cooling could also lift revenue per board. Daiwa estimates the liquid-cooled motherboard carries four chipsets versus two on the air-cooled version. It expects the higher count to offset a modest decline in price per chipset. On the networking side, Accton told analysts that a major social media customer secured more switch chips from Marvell and more NVIDIA Spectrum chips. It was still negotiating with Broadcom for a larger allocation. That customer overtook Accton’s e-commerce customer by revenue contribution in Q3 2026, according to UBS’s account of the briefing. $AMZN $MRVL $AVGO $NVDA

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.

This is a national security crisis for Taiwan.

Jensen Huang has resigned as CEO of Nvidia