Shennan Circuit

Where we land

Shennan Circuit makes the circuit boards inside AI servers and network switches, and that work has changed the shape of its profits. Yearly operating profit, what is left after the costs of making the products and running the company, has risen from 2.5bn to 5.7bn, and the analysts who follow the company keep raising their numbers for what comes next. The shares trade at 31 times earnings, below most of the companies it competes with. Our view is that this price still underplays how fast earnings are moving.

We award it silver. Analysts rank it among the very best companies we follow, and on the quality of the business, how profitable and soundly run it is, it stands with the strongest we rate. It misses gold for one reason only: its profits are not backed by cash coming into the business as fully as we require for our highest award.

What the business is

Most of the company is one business. Printed circuit boards, the layered boards that hold a device's chips and carry signals between them, make up 69% of revenue. Assembly, mounting components onto boards for customers, adds 18%. The remaining 13% is semiconductor packaging and testing, including the small, dense substrates that connect a chip to the board beneath it.

The economics of that mix have shifted sharply. Of each sale, 35% is left after the direct cost of making the product, a measure called gross margin, and 18% remains as operating profit. Over five years the gross margin has had a middle value of 24%, typically moving only about 3 points either side. Today's figure is well beyond that range, and much of the argument about this company, for and against, turns on whether the jump is a new level or a peak.

Why now

The engine is demand for complex, high-speed circuit boards used in AI servers, network switches and the optical modules that move data between them. These boards have many more layers and far tighter tolerances than ordinary ones, and few suppliers can meet the specification. When demand runs ahead of the factories able to make them, the companies that can tend to win both volume and price, the pattern visible in Shennan Circuit's margins. It has sat in our Conviction 10 portfolio since October 2025 and still earns its place there.

The case for

The strongest argument is what the analysts expect. Consensus earnings per share, the profit attributable to each share, is 8.81 for this year and 13.98 for next, a rise of about 59%. That is a large step for a company of this size.

The direction of travel matters as much as the level. In the last 45 days there were 16 changes to those estimates, every one an upgrade. When forecasts move only one way for a sustained stretch, it usually means the people closest to the company keep finding orders and pricing better than they had assumed.

Those forecasts rest on growth already delivered. Sales rose 51% over the past year, on top of a three-year compound growth rate of 36%, so this is an acceleration on an already fast pace rather than one good year after a weak one. On sales growth it sits close to the top of the whole market.

Where the growth could come from

The main engine is most likely AI data-centre boards. Our revenue figures do not split data-centre and AI boards from telecom and other boards, but sales up 51% and every recent forecast change pointing upward fit demand for high-layer, high-speed boards better than anything else. These boards carry better prices and margins than standard ones, so as their share rises earnings can grow faster than revenue, exactly what the 59% earnings forecast implies. Confirmation would be the company reporting a rising share of sales from data-centre and server customers alongside a higher circuit-board gross margin.

The second possible engine is packaging substrates, the 13% line. Chinese chipmakers are pushing to buy advanced substrates at home rather than from Japan, Taiwan and Korea, and the company is ramping newer high-end substrate lines. Advanced substrates for processors and AI chips are hard to make and command high prices, but new lines usually weigh on profit while they fill before lifting it. This is early, with orders but little revenue yet. The signs it was working would be the substrate business turning from a drag on profit into a contributor, and named qualification wins at domestic chip designers.

Car electronics are a smaller option. Chinese carmakers are packing more electronics into each vehicle, for driver assistance, radar and electric drive controls, all of which need boards. Automotive is a minority of the circuit-board line and does not drive earnings today, but car programmes bring steadier, longer-running orders that could cushion the company if data-centre spending slowed. It would show as car-related board sales becoming a visible share of the business, with new capacity running full on vehicle programmes.

Running through all of this is copper, the core input in copper foil and copper-clad laminates for both boards and substrates. Copper averaged $14,473 a tonne in September 2026, up 45% over a year though little changed over three months. With AI board demand strong, the company may well pass higher costs on. If it cannot, sharply dearer copper would eat into exactly the margin gains that support the earnings forecast.

The case against

The first concern is that today's profitability looks exceptional rather than normal. Beyond the gross margin's break from its narrow 24% history, return on equity, the profit earned on shareholders' money, is 25%, against a five-year median return on invested capital of 11.6%. Margins that break out of a long, narrow range can fall back into it.

History shows how quickly that can happen. Between September 2022 and September 2023, twelve-month operating profit fell from 1.6bn to 1.1bn, a drop of about a third, before climbing to 5.7bn by September 2026. The recovery is impressive, but this is a business whose earnings can reverse, which matters when the market is pricing in another large step up.

The peer comparison is less flattering than the headline multiple suggests. At 31 times earnings Shennan Circuit is cheaper than Unimicron at 48, Gold Circuit at 36, and Avary and Kinwong at 34; only Victory Giant, at 28, is cheaper. Its gross margin matches Gold Circuit and Victory Giant at about 35-37%, but its 18% operating margin is well short of the 26% those two reach. Less of each sale survives the cost of running the business, so part of the discount may reflect a less efficient operation.

In absolute terms the shares are not cheap at all. The price is 7.7 times book value, the accounting value of assets minus debts, and the whole business, including debt, is valued at 29.5 times operating profit. On those measures it sits near the bottom of the market for value, a valuation that depends on elevated margins and forecast growth both holding.

The share price itself has cooled. The shares are up only 9% over the past year despite earnings more than doubling, and have fallen 25% over six months even as the upgrades kept coming. Their strength relative to the wider market remains high but has faded sharply over both three and twelve months, which looks like a sustained loss of enthusiasm rather than a one-off dip.

What the price assumes

One way to judge a share price is to ask what growth it needs. Suppose the share price stayed flat for five years and the shares ended trading at 18.5 times earnings, the median for every profitable listed company we cover. Earnings would need to compound at about 11% a year, reaching 12.63 a share.

Against that, the bar looks low. Analysts' forecast for next year alone, 13.98, already exceeds what the price needs five years out. Operating profit has grown about 84% a year over the past three years; carried forward, that would put earnings per share at 160.95 in five years, 1275% of the requirement. That is the company's past, not a forecast, and it starts from the 2023 dip, which flatters the rate. Nobody should expect such a pace to persist, but even a small fraction of it would clear the hurdle. The market is treating current profits with suspicion, and that gap is what our verdict rests on.

The other side of the medal

The single risk we would keep in view is that the AI story cannot yet be seen directly in the numbers, while the main raw material is testing the company's pricing power. Because our figures fold data-centre boards together with telecom boards, the shift towards higher-value AI work is inferred from the pace of growth and the jump in margin, not read off a line. Meanwhile dearer copper is asking whether customers will keep accepting price rises.

Those two strands meet in the margin. If AI demand is as strong as the pattern suggests, price rises stick and margins hold. If it is weaker, or some of today's strength is telecom or one-off orders, copper would expose it quickly by squeezing margins back towards their old range. That would bite hard, because on price against assets, sales and earnings the shares are already dearer than most of the companies we rate.

What to watch

The clearest signal is the gross margin set against the copper price. A gross margin holding at about 35% or above while copper stays high would show customers absorbing higher material costs, possible only when demand for these boards is genuinely tight. Copper moves daily and well ahead of the company's results, so a further leg up without a matching margin hold would be the earliest warning.

The second is the direction of analysts' forecasts, which change before reported results do. If analysts start cutting next year's numbers, or the gross margin slides back toward its five-year median of around 24%, the reading of a lasting step-change would be wrong and the valuation would look stretched. Beyond that, any disclosure of a rising share of sales from data-centre and server customers, or of the substrate business beginning to add to profit, would turn our inference into something visible.

The numbers

The same industry, for comparison

Research for information, not personal investment advice or a recommendation to buy, sell or hold any security: see Important information. Data as at 31 October 2026.