Tracking the companies and technologies behind the AI buildout. Research informed by company interviews, site visits, analysts, and industry experts.

We cover the companies behind the AI buildout, from chips and memory to the equipment and materials that make them possible. Based in Taiwan, we draw on company interviews, site visits, public filings and industry research to explain what’s changing across these supply chains—and what it means for the businesses involved. A few places to start: • Phison’s move into systems and software. CEO K.S. Pua spoke with us about AI’s memory constraints and why a company known for storage is expanding into a broader business. Read the interview: nitter.siddz.com/statementdog/status/20… • What it takes to expand ABF substrate production. Goldman Sachs reports that Eternal Precision’s equipment delivery slots are booked into 2031. We look at how equipment availability could shape suppliers’ expansion plans. Read more: nitter.siddz.com/statementdog/status/21… • Why faster optical modules could compete with AI chips for capacity. Drawing on DIGITIMES reporting, we examine the substrate requirements behind the shift to 3.2T optical modules. Read more: nitter.siddz.com/statementdog/status/21… We follow these developments beyond the initial announcement: when capacity comes online, whether products pass customer qualification, and how supply affects pricing. Follow @statementdog for company interviews and ongoing analysis of AI hardware and semiconductor supply chains.
Only three PCB makers currently meet the requirements for 3.2T optical module boards, according to an industry source cited by DIGITIMES: Zhen Ding Tech (4958 TT), Unimicron (3037 TT) and Shennan Circuits (002916 CH). The modules are expected to begin shipping in 2028. Their boards require line widths and spacing of 10 to 15 microns, which could push production from mSAP to the more complex SAP process. Suppliers also need FC-BGA substrate capacity. More suppliers, including Compeq (2313 TT), can handle mSAP, which supports line widths and spacing of 12 to 25 microns. Boards for 1.6T modules already require mSAP and M8 laminates, while M9 is planned for 3.2T boards. The ramp-up is expected to overlap with the IC substrate shortage. Module makers that don’t reserve capacity early could end up competing with AI and HPC chipmakers for production slots. digitimes.com.tw/tech/dt/n/s…
1
1
38
3,847
NVIDIA’s latest preliminary specs list Rubin’s GPU-to-GPU bandwidth at 3TB/s, down from 3.6TB/s. Workloads that spend more time waiting for data from other GPUs could be more affected, but the bandwidth change alone doesn’t tell us the overall performance impact.
NVIDIA has revised Vera Rubin NVL72’s NVLink bandwidth from 3.6 TB/s to 3.0 TB/s per Rubin GPU package, a ~17% reduction. Vera Rubin utilizes bidirectional SerDes, with a theoretical full-rate bandwidth of 448 Gbps per physical connection. However, crosstalk, insertion loss, and return loss create significant signal integrity challenges at these speeds. (1/2)🧵
2
900
ASE is putting half of its extra $1 billion in planned 2026 investment into facilities. Suitable factories are still hard to find, making space another constraint on how quickly packaging and testing capacity can grow.
ASE, the chip packaging & testing giant 3rd quarter highlights, media report: -Raised capex to US$11.5 billion from $10.5 billion; 2027 to be even higher -Forecast 4th quarter revenue +6%-8% from the 3rd -Raised 2026 advanced packaging revenue forecast to US$4 billion from $3.7B revenue, sees figure doubling in 2027. -CPO revenue to top US$500 million in 2027 $ASX #semiconductors money.udn.com/money/story/56…
1
4
673
Chroma, a Taiwanese test-equipment maker, says it has received purchase orders for Q1 2027 to support volume production of co-packaged optics (CPO). The company expects the current pilot phase to wrap up by the end of 2026, and customers have provided their 2027 demand forecasts. For Nvidia's Spectrum-X, Chroma expects 2027 shipments of Insertion 3 optical-engine test equipment to be at least four to five times the pilot-run level. Chroma said Marvell plans to enter the pilot stage in 2027, with volume production targeted for Q4 of that year. Q3 revenue rose 33% from the previous quarter and 180% year over year to NT$17.93 billion. Operating profit increased 37% from Q2 and 288% from a year earlier to NT$7.17 billion. Gross margin slipped to 57.5% from 60.5% in Q2, but operating margin improved to 40.0% from 38.7% as revenue grew faster than operating expenses. Net profit attributable to shareholders was NT$5.65 billion. Basic EPS rose 11% year over year to NT$13.34. The year-earlier quarter included a NT$3.19 billion gain from residential property sales. On a parent-company basis, semiconductor and photonics testing revenue reached NT$6.06 billion, up 38% from Q2 and 189% year over year, accounting for 42% of parent-company sales. Test instruments and automated test systems generated NT$7.43 billion, up 21% from Q2. Operating cash flow for the first nine months was NT$6.21 billion, compared with NT$14.96 billion in consolidated net income. Inventory rose 74% from the end of 2025 to NT$13.81 billion.
26
1,447
Good point on the PC spillover. For $AMD, we’d watch the next Client outlook for signs of memory costs hitting PC demand. In Q2 2026, Client revenue grew 23% YoY. AMD attributed Gaming’s 31% decline to lower semi-custom sales.
Some quick thoughts on Intel, Sandisk...and AMD. You can make sense of Intel's earnings miss via Sandisk's updated revenue mix. Sandisk: Around 88% of Sandisk's Q/Q revenue gain came from Data Center. However Ex-DC revenue, which includes edge devices like PCs, grew by only 4% despite price increases. Follow that through to Intel & Lip-Bu Tan's comments: "Turning to our client business, in recent weeks, our customers began signaling a PC slowdown based on the memory prices and availability." While the market had already somewhat priced-in a weak PC market for Intel earlier in the year, I don't think anyone expected it to be *this* quick. I therefore naturally wonder whether this is a negative (but smaller) read on AMD, whose gaming business was already down 31% Y/Y in Q2 due to memory constraints. I ultimately think that's why AMD's stock is currently down 7% after-hours.
2
710
Elite Material (2383 TT), a maker of copper-clad laminates, expects to supply half the core laminate used in ABF substrates for MediaTek’s TPU v8 program for Google ($GOOGL), according to Goldman Sachs. The company said on its Oct. 29 earnings call that tight supplies from Japanese material makers were prompting customers to accelerate qualification of a second source. Elite Material expects to begin shipments in 2027. The company expects gross margins of at least 50% on the substrate material, compared with more than 65% for the market leader, according to Goldman. Elite Material is also working with several customers on qualification projects for high-end ABF substrate materials. Goldman says shipments from those projects are expected to begin in Q4 2027. The company plans to add capacity ahead of customer demand. That capacity could also be used to make M9, M10 and M11 circuit-board laminates if substrate demand does not fully absorb it, according to Goldman.
8
773
Sandisk says some customers are coming back for more NAND than they originally expected to need. Turning those requests into firm commitments would give it a stronger basis for deciding how much supply to add.
Sandisk $SNDK says its pipeline of long-term NAND supply agreements now exceeds available supply through the end of the decade. About half of FY27 bits and two-thirds of FY28 bits are already committed, with more deals expected.
11
1,174
Hanmi Semiconductor (042700 KS), which makes bonding equipment for HBM production, reported preliminary Q3 revenue of 181.0 billion won, down 27.9% from Q2. Operating profit fell 30.3% to 90.8 billion won. Revenue was up 8.9% from a year earlier, while operating profit rose 33.9%. Operating margin widened to 50.2% from 40.8%. Hanmi said some equipment revenue had been deferred to Q4 and expects a stronger fourth quarter. The company also expects 2026 to deliver a third straight year of record revenue and a new annual high in operating profit. thelec.kr/news/articleView.h…
4
623
StatementDog | AI & Tech Supply Chains retweeted
$INTC Intel Deep Dive: The Server Story Is Real. The Guide not Pretty. Intel beat Q3 and still fell 21% after hours. Data center revenue grew 78%, the best in 15 years. DCAI now earns more than the PC business. 18A costs are down 40% since Q1. Q4 came in $1.8B below the Street. A tragic accident at Fab 28 in Israel cost $300M. PCs are in freefall because AI is eating the world’s memory supply. And 2027 won’t be a free cash flow year. I break down the numbers, the bull and bear case, what the stock prices in, and the five catalysts I’m watching before I buy.
Made with AI
10
3
50
5,540
Plastics used in battery backup units (BBUs) can take months to qualify. Materials supplier Nytex ran roughly 30–40 of its own battery-pack tests in one year during development, excluding tests done with customers. In EP572, general manager Henry Chen described repeated rounds of failure analysis, formulation changes, and testing. When third-party labs were fully booked, another test slot could be two or three weeks away. For a new supplier, he put qualification at six to nine months or longer. The work includes battery fire-propagation testing under UL 9540A, alongside separate evaluations of the material’s long-term thermal performance.
9
747
Intel ($INTC) reported higher Q3 adjusted earnings on strong data center demand, but PC weakness and the Fab 28 shutdown are weighing on Q4. Q3 2026: • Revenue: $16.51B, up 21% YoY • Adjusted operating income: $3.23B; operating margin: 19.6% • Adjusted gross margin: 44.1%, up 4.1 percentage points YoY • Adjusted EPS: $0.52, up 126% YoY DCAI revenue rose 78% YoY to $7.32B, while client revenue fell 5% YoY. Q3 results: intc.com/news-events/press-r… For Q4, Intel forecasts $15–16B in revenue, a 38% adjusted gross margin and $0.25 in adjusted EPS. An October accident led to an extended shutdown at Fab 28 in Israel. Intel estimates a $300M Q4 revenue impact, almost entirely from 10nm server products, plus a two-percentage-point hit to gross margin. Both impacts are already included in Q4 guidance. Despite the disruption, Intel still expects DCAI revenue to grow at a mid-teens percentage rate sequentially—underscoring that server supply remains a constraint even as PC demand weakens. Earnings-call remarks: d1io3yog0oux5.cloudfront.net…
1
1
1
662
Delta (2308 TT) may need to keep inventory buffers above historical levels even after delayed orders ship, as it adjusts to uneven AI deployment schedules. Component shortages, customer schedule changes and shifting specs drove inventory buildup, with Q3 provisions equal to 1.7% of revenue. We think that clearing delayed shipments may offer only partial relief. Keeping more material on hand to handle unpredictable schedules could leave Delta with a lasting need for more working capital. Microsoft $MSFT, meanwhile, shifted some third-party capacity deliveries into the second half of FY27. For suppliers, a pause in customer spending can still mean holding inventory while waiting to ship.
22
1,229
Samsung plans to spend about ₩84.4 trillion on capital projects in 2026 and expects AI-driven memory demand to exceed industry supply in 2027. Q3 revenue rose 14% from the previous quarter to ₩195.3 trillion, while operating profit increased 20% to ₩107.4 trillion. Its semiconductor division, Device Solutions (DS), reported revenue of ₩147.1 trillion and operating profit of ₩106.5 trillion. Samsung said strong server demand pushed memory prices higher across applications. Its memory business posted record quarterly revenue and operating profit as it prioritized AI products amid capacity constraints. HBM4 sales expanded substantially, and demand for server SSDs remained strong. For Q4, Samsung expects new CPU and GPU platforms to support memory demand. It plans to increase sales of HBM, server DRAM and SSDs, while expanding supplies of high-speed server DDR5 and QLC SSDs. Capital spending reached ₩29 trillion in Q3, up ₩12.2 trillion from Q2. DS accounted for ₩27.2 trillion of that spending. The full-year DS budget is ₩76.8 trillion. That covers both memory and foundry investment, including new fabs and other infrastructure to support AI and high-performance computing. Samsung also expects AI infrastructure investment to reach a record in 2027. It plans to pursue multi-year customer agreements to build a more stable sales base.
4
644
Powertech’s $400 million Singapore investment has won Taiwan’s approval, according to local media. Using larger rectangular panels could lower packaging costs for big AI chips, provided the process can deliver high yields at scale.
Taiwan’s Powertech is reportedly partnering with Broadcom to build a $400 million panel-level packaging (PLP) fab in Singapore. money.udn.com/money/story/56…
7
813
Larger carriers for advanced chip packaging could lift plastics demand even at unchanged prices per kilogram. In EP572, Henry Chen, general manager of materials supplier Nytex, pointed to the greater amount of material used in each carrier. Larger designs still need formulation changes to maintain strength, dimensional stability, and electrostatic control. Nytex is working with carrier supplier Gudeng on materials for advanced packaging, including CoWoS applications. Those projects are still in verification, so the opportunity for higher material volumes depends on them reaching volume production.
11
887
Google Cloud revenue more than doubled in Q3, while Search advertising continued to grow. Equity investment losses pulled Alphabet’s earnings lower, and higher capital spending left free cash flow negative for a second straight quarter. Alphabet reported revenue of $127.6 billion, up 25% from a year earlier, and operating income of $40.6 billion, up 30%. Net income available to common shareholders fell 25% to $26.2 billion, while diluted EPS declined 26% to $2.12. Google Cloud revenue grew 103% to $30.7 billion, accelerating from 82% growth in Q2. Segment operating income reached $11.1 billion, with its operating margin rising to 36.2% from 23.7% a year earlier. Based on the reported figures, Cloud accounted for about 62% of Alphabet’s year-over-year revenue increase. Alphabet cited strong demand for enterprise AI infrastructure and solutions. TPU system sales also contributed meaningfully to growth, so Cloud’s revenue includes hardware sales alongside cloud services. Search and other revenue rose 12% to $63.4 billion, slowing from 17% growth in Q2. YouTube ad revenue increased 9% to $11.2 billion. Management said the year-over-year growth rate in Search queries was its highest in five years, with AI also driving commercial queries. How effectively Google turns that additional usage into advertising revenue remains an open question. The decline in earnings largely reflected equity investment losses of $8.7 billion, compared with gains of $10.7 billion a year earlier. Alphabet said this quarter’s losses reduced EPS by $0.55. These investment results include both realized and unrealized gains and losses. Group operating margin was 31.8%, down from 34% in Q2. Operating cash flow reached $42.3 billion, while capital spending jumped 114% year over year to $51.4 billion. Free cash flow fell to negative $9.1 billion from negative $5.9 billion in Q2, though the trailing 12-month total remained positive at $19.7 billion. Alphabet also raised $12.4 billion through its at-the-market common-stock program and $28.7 billion from senior unsecured notes, both net of issuance costs. It repurchased no shares during the quarter. When it announced the ATM program in Q2, the company said the proceeds were primarily intended to cover tax obligations tied to employee equity grants. Alphabet-level activities recorded an operating loss of $7.7 billion, primarily reflecting shared AI research and development. Cloud’s 36.2% segment margin excludes those shared costs.
1
14
1,136
Microsoft’s cloud growth accelerated in the September quarter, but a sharp rise in capital spending squeezed free cash flow. Revenue rose 19% year over year to $92.3 billion, while operating income increased 21% to $45.9 billion. Microsoft Cloud revenue grew 30% to $63.9 billion, led by 48% growth in Azure. Cash spending on property and equipment more than doubled to $39.4 billion. Subtracting that from $58.0 billion in operating cash flow leaves about $18.6 billion in free cash flow, down 27% from a year earlier. The cash-spending figure excludes finance leases. The 32% increase in reported EPS also needs context. OpenAI investment losses weighed on last year’s earnings, while gains helped this quarter. Excluding those effects in both periods, EPS rose 18% to $4.89, compared with reported EPS of $4.92. Commercial remaining performance obligations climbed 73% to $677 billion, representing contracted revenue still to be recognized. Consumer businesses were weaker: Windows OEM and Devices revenue fell 22%, while Xbox content and services declined 2%. Microsoft also changed its segment structure and reporting definitions this quarter, so historical comparisons should use recast figures. microsoft.com/en-us/Investor…
2
10
908
Converting existing capacity can help UMC respond sooner than waiting for new fabs. The longer-term payoff depends on how quickly its expansion turns into usable capacity.
UMC approves an additional $3.2 billion as capacity tightens $UMC’s Q3 utilization reached 92%, with several 12-inch fabs fully loaded. Growing advanced packaging and silicon photonics demand is prompting the company to repurpose existing capacity. The additional budget supports expansion in Singapore and Tainan, while 2026 capex guidance remains $2 billion. This is bullish for UMC’s medium-term growth and semiconductor equipment demand. Capacity reallocation shows new applications are competing for manufacturing resources today; expansion creates potential equipment demand beyond the current spending cycle. $AMAT $LRCX $KLAC money.udn.com/money/story/57…
9
728
Elite Material, a Taiwanese maker of copper-clad laminates (CCL) used in AI server and networking boards, plans to more than double monthly capacity by the end of 2030. The company says its planned expansion would still fall short of demand covered by signed long-term supply agreements, according to Goldman Sachs’ Oct. 28 report. Monthly capacity would rise from 5.85 million sheets today to 15.25 million by the end of 2030. Interim year-end targets are 8.70 million sheets a month in 2027, 10.45 million in 2028 and 11.95 million in 2029. Customers are initiating the agreements, which cover volumes at least three years into the future, Goldman reports. Nearer term, an additional 1.2 million sheets of monthly capacity at Kunshan could come online in stages from December 2026 through May 2027, ahead of the original schedule. Goldman expects that faster ramp to offset a delay at Guanyin, where 300,000 sheets of planned monthly capacity have been pushed back from Q4 2026 to the second half of 2027.
14
1,165