Listed PCB makers' accounts and Thailand's build-out: a 38-company study

Key points

• The 38 listed bare-board makers in the sample sold US$50.9 billion in 2025, up 22.0% year on year in US$ on a like-for-like basis, and spent 15.2% of sales on plant and equipment.^[1][2][3]
• At company level, annual sales are roughly equal to the previous ten years' capital spending: the median ratio is 1.07, with the middle 80% of observations between 0.71 and 2.28. The ratio describes whole companies, not individual plants.
• Within the same company, one additional turn of plant (one more unit of annual sales per unit of net plant) was associated with 5.5 percentage points higher gross margin.
• Capital spending rises in the same year as sales growth and does not predict additional growth in later years. Machine-learning models on the accounts forecast next-year growth only marginally better than a no-change baseline.
• Across all 38 makers, installed plant could support 26.5% more sales at each maker's historical turnover (22.3% for the 28 makers with a plant in Thailand), and construction in progress reached 15.2% of the Chinese makers' net plant in 2025.
• Thai PCB exports tracked the Taiwanese makers' sales until 2025. In 2026 Thai exports outgrew them: +39.4% from January to May against +24.7%, and +78.8% in July against +26.0%.^[7][8][2]

Context

Of the 38 makers in the sample, 28 own or are building a plant in Thailand, according to the MPBxChange register of plants. Their public filings show what they spend on plant, the sales that spending supports at company level and the behaviour of margins as utilisation changes, which are the reference figures for buyers assessing a new plant and for plants planning capacity. This report treats the filings as a single panel and tests their predictive power.

Data

The data set covers all published annual figures in the open sources for 43 listed companies, 42 of which enter the analysis:

• 26 companies listed in Shanghai and Shenzhen: income statements, balance sheets, cash flow statements and product segments from their annual reports.^[1] Avary Holding is consolidated in Zhen Ding Technology (listed in Taipei) and is not counted separately.
• 16 companies listed in Taipei, 13 board makers and 3 laminate makers: quarterly statements and monthly revenue as filed on the Market Observation Post System.^[2]
• KCE Electronics, the listed Thai maker, from its One Reports filed with Thailand's Securities and Exchange Commission.^[3][4][5]

Levels are converted to US$ at annual average exchange rates; ratios are computed in local currency.^[6] A company is classified as a board maker in a year when boards are at least 60% of its main business, which includes Dongshan Precision (boards were 63.9% of its 2025 business) and excludes the laminate makers.^[1] The panel holds 507 company-years from 2012 to 2025, with 38 board makers in 2025.

Industry aggregates, 2016 to 2025

Note: growth compares the same companies in both years (year on year, in US$) and excludes the effect of new listings. Aggregate sales also rise with the number of listed makers.

Capital spending peaked at 16.9% of sales in 2021, eased to 12.8% in 2024 and rose to 15.2% in 2025, when the makers spent US$7.8 billion. Gross margin ranged between 16.9% and 24.0% through the cycle.

Sales per unit of investment

An announced investment covers land, buildings and machinery. Two ratios from the accounts relate that cost to the output it supports.

Sales per unit of ten years' capital spending. Plant in use approximates a company's spending on plant over the last ten years, so this ratio relates sales to plant cost. It requires ten years of cash flow statements, available for 30 of the 38 makers.

1. Sales: Shenzhen Kinwong Electronic, CNY 15.31 billion in 2025.^[1]
2. Cash paid for plant and equipment, 2016 to 2025: CNY 0.34 billion in 2016 rising to CNY 2.77 billion in 2025, CNY 14.60 billion in total.^[1]
3. Ratio: 15.31 / 14.60 = 1.05, close to the sample median.

Sales per unit of net plant. Net property, plant and equipment is plant at cost less depreciation, plus construction in progress. Because older plant is written down, this ratio is higher.

1. Revenue from operations: KCE Electronics, THB 13,074.69 million in 2025.^[3]
2. Net plant: THB 7,645.37 million at end-2024 and THB 7,267.23 million at end-2025; average THB 7,456.30 million.^[3]
3. Ratio: 13,074.69 / 7,456.30 = 1.75.

Distribution across the panel:

The ten-year spending ratio is the relevant benchmark for an investment announcement. At the median, a company's annual sales are about 1.07 times its previous ten years of capital spending, or THB 10.7 billion for each THB 10 billion spent, with a range of THB 7.1 billion to THB 22.8 billion between the low and high cases. The ratio is company-wide: it mixes plants of every age and level of utilisation and is not the measured output of one plant running full. On MPBxChange's reading, the spread reflects product mix as well as utilisation, as makers of high-layer server boards earn more sales per baht of equipment than makers of simpler boards. The net-plant ratio is an upper bound, as it is inflated by older, depreciated plant.

Utilisation and gross margin

Plant costs are largely fixed (depreciation, power, staff), so margin is expected to rise with output from the same plant. The panel regression quantifies the effect.

1. Turns: sales divided by average net plant, for each company-year.
2. Model: gross margin regressed on turns with company and year fixed effects (within-company comparison, net of industry-wide annual swings), standard errors clustered by company.
3. Result: each additional turn is associated with 5.5 percentage points higher gross margin (standard error 0.9 points; 406 company-years; 95% interval 3.7 to 7.3 points).

A new plant starts well below the industry median of about 2 turns, as its assets are on the books before sales arrive. Moving from 1 turn to 2 implies a gross margin gain of about 5 points. Heavy capital spending in one year had no significant effect on the following year's margin (coefficient +0.07 per unit of spending over sales).

Timing of capital spending and growth

Both directions of the relationship are estimated with the same company and year effects.

• Spending as a share of the previous year's sales rises by 0.20 for each unit of growth in the same year (standard error 0.04, 368 company-years). Prior-year growth is not significant (+0.03). A maker growing 20% spends about 4% of prior-year sales more in that year.
• Sales growth rises with same-year spending (0.70 per unit, standard error 0.16); spending lagged one, two or three years is not significant (coefficients -0.22, -0.09 and +0.17).

Capital spending follows orders: an announced plant reflects demand the company already sees. The Thai investors fit this pattern. The 28 makers with a plant in Thailand grew 13.6% in 2024 while the other 10 contracted 3.4%, and in 2025 they spent 16.6% of sales on plant against 9.9% for the others.

Forecasting next-year growth

Each model was trained only on years before the forecast year and scored on that year, for every year from 2019 to 2025: 249 out-of-sample forecasts of a company's following-year growth. Inputs were 13 variables from the accounts, including current and prior-year growth, spending, margins, turns, size and industry growth.

The best model has a mean absolute error of 15.1 points, 12% lower than the no-change baseline, and calls the direction of growth correctly in little more than half of cases. Next-year growth depends on orders outside the accounts, such as a customer's new server programme, in line with research on company growth that finds "scant persistence in growth beyond chance".^[9] A plant's capabilities and existing customer base are better guides for buyers than its recent growth.

Installed capacity

1. Each maker's net plant at end-2025 multiplied by its own median turns over 2016 to 2025, summed: US$64.4 billion of annual sales capacity at historical turnover.
2. Against 2025 sales of US$50.9 billion, installed plant could support 26.5% more.
3. For the 28 makers with a plant in Thailand: US$49.9 billion of capacity against US$40.8 billion of sales, 22.3% more.

Construction in progress was 15.2% of the Chinese makers' net plant at end-2025, against 6.4% in 2016, much of it in new plants abroad, including Thailand.

Thai exports and the industry, 2016 to 2026

The 12 Taiwanese makers with a complete monthly record publish sales about ten days after month-end; Thai customs figures reach UN Comtrade two to three months later.^[2][7] From January 2016 to May 2026 the two growth rates moved together, with a monthly correlation of 0.65.

As a leading indicator for Thai exports, a model fitted only on published months and scored from 2019 had a mean error of 11.7%, against 11.2% for repeating the growth rate of three months earlier. In 2026 the two series diverged. From January to May, Thai exports rose 39.4% while the Taiwanese makers' sales rose 24.7%.^[7][2] In July, Thai printed circuit exports reached US$232 million, up 78.8%, while the makers' sales rose 26.0%.^[8][2] A model fitted on the makers' sales put July at US$168 million, with an 80% range of US$146 to 191 million; the outturn exceeded the top of that range.

The divergence reflects new capacity in Thailand: the makers' sales track the industry, while Thai exports track the industry plus the plants opened there since 2024. The export forecast extends the trade data to 2030.

Outlook

The 28 makers with a plant in Thailand could support about a fifth more sales (22.3%) on installed plant, and margins at new plants rise as orders are added, which favours buyers seeking capacity commitments, samples and qualification support. At the median, company sales are about THB 1.07 billion a year for each THB 1 billion of capital spending over the previous ten years, with margin rising about 5 points per additional turn; utilisation is the main driver of plant economics in these accounts. Because spending follows orders, the size of an announced investment, read against these ratios, indicates the scale of output to come.

Methodology and limitations

Annual figures for 26 Chinese makers come from the Eastmoney datacenter, which carries statements filed with the Shanghai and Shenzhen exchanges;^[1] quarterly and monthly figures for 16 Taiwanese companies from FinMind, which carries filings on the Market Observation Post System;^[2] and KCE Electronics' figures from its One Reports for 2021, 2022 and 2025, filed with Thailand's Securities and Exchange Commission.^[3][4][5] Exchange rates are FRED annual and monthly averages.^[6] Thai exports are monthly returns to UN Comtrade and, for July 2026, the Office of Industrial Economics.^[7][8] Data are as of 26 September 2026; Taiwanese monthly sales are used to July 2026. Regressions use company and year fixed effects with standard errors clustered by company; forecasts are out of sample, trained on earlier years only. Plant counts come from the MPBxChange register of PCB plants in Thailand.

Reproducibility. The 43 companies are, in Shanghai and Shenzhen: Shennan Circuits, Victory Giant Technology, WUS Printed Circuit, Shenzhen Kinwong Electronic, Dongshan Precision, Olympic Circuit Technology, Aoshikang Technology, Shengyi Electronics, Suntak Technology, Sihui Fushi Electronics, Guanghe Technology, Changzhou Aohong Electronic, Jiangsu Allfavor Intelligent Circuit, Founder Technology, Huizhou China Eagle Electronic, Guangdong Ellington Electronics, Jian ManKun Technology, Jiangxi Welgao Electronics, Shenzhen Zhongfu Circuit, Shenzhen Fastprint Circuit Tech, Bomin Electronics, Shenzhen Sunshine Circuit, Guangdong Champion Asia, Guangdong Kexiang Electronic and the laminate maker Shengyi Technology, plus Avary Holding, collected but left out of the analysis because Zhen Ding consolidates it; in Taipei: Compeq Manufacturing, Gold Circuit Electronics, Unimicron Technology, Zhen Ding Technology, Dynamic Holding, Chin-Poon Industrial, Unitech Printed Circuit Board, Apex International, Nan Ya Printed Circuit Board, Kinsus Interconnect, Tripod Technology, HannStar Board and Flexium Interconnect, and the laminate makers Elite Material, ITEQ and Taiwan Union Technology; and KCE Electronics in Bangkok. The 13 forecasting inputs are: this year's sales growth; last year's sales growth; capital spending over sales; last year's capital spending over last year's sales; gross margin; operating margin; sales over average net plant; depreciation over the previous year's net plant; inventory days; the logarithm of sales in US$; the board makers' aggregate sales growth that year; whether the company is listed in Taipei; and whether it has a plant in Thailand. Ridge regression, a random forest and gradient boosting are refitted for each forecast year on earlier years only; a forecast counts as right on direction when it and the outcome fall on the same side of that year's median growth. The 0.65 correlation is the Pearson correlation, with no lag, between the year-on-year growth of Thailand's monthly HS 8534 exports and the year-on-year growth of the summed monthly sales in US$ of the 12 Taiwanese board makers with a near-complete monthly record (all 13 except Dynamic Holding, whose record starts in August 2022), over the 125 months from January 2016 to May 2026 in which both are available.

• Coverage is limited to listed companies with open filings. Japanese and Korean makers, several with Thai plants, are outside the sample, and Thailand is represented by one listed maker.
• Consolidated accounts combine plants in several countries; the ratios describe companies rather than individual plants.
• The ten-year spending measure requires ten years of cash flows, available for 30 of the 38 makers; younger companies appear in the other measures only.
• Growth in US$ includes exchange-rate movements; ratios are unaffected.
• Regression coefficients are within-company associations net of annual industry swings.
• The forecast test comprises 249 predictions over seven years. Inputs such as order books could improve on the accounts.
• Compeq's and HannStar Board's 2020 capital spending is stored with a reversed sign in the source; the amounts, which match their nine-month figures, are treated as outflows.
• July 2026 is from Thailand's Office of Industrial Economics, pending UN Comtrade publication, and is subject to revision.