AT&S (ATS.VI): The Only European Building Leading-Edge AI Chip Substrates, Up 11x as Customers Fund the Next Three Factories

A year ago, $ATS.VI, AT&S was a roughly $0.8 billion Austrian printed-circuit-board maker that most investors outside the European small-cap circuit could not have placed on a map. Today it is a roughly $8.8 billion AI-infrastructure compounder, up around 11x, and on 13 June 2026 it agreed key terms with AMD and a second tech giant on long-term commitments that finance a $1.7 to $2.3 billion build-out of its Kulim, Malaysia Phase 2 site and a brand-new manufacturing site presumed to be Phase 3.

The factor framework reads it as Buy at StockRank 80 of 100, with Momentum carrying the rank. The Quality and Value scores are still modest because the operating economics are rebuilding from the 2024 trough. The catalysts are visible, the customer commitment is real, and the insider activity called the bottom: two insiders bought near multi-year lows in December 2025 and are now up roughly 6x. This post is an attempt to lay out the moat, the catalyst chain, the honest bear case, and what the company has chosen to do with the cash that other boards would have returned.

What an IC Substrate Actually Is

The first thing to get right on AT&S is what the product is and what it sits underneath, because the category is not the consumer-electronics PCB business most generalists associate with the company name.

An IC substrate is a micron-precise board that sits directly under a leading-edge AI or HPC chip. Its job is to fan thousands of electrical connections out from the dense ball-grid array on the underside of the silicon to the system board around it, without losing signal integrity, without warping under thermal load, and without becoming the latency bottleneck of the accelerator above it. As AI chips have grown denser and hotter, the substrate has gone from a packaging afterthought to the constraining layer of the entire accelerator stack.

A modern AI accelerator costs roughly $30,000 at the wafer level. The substrate underneath it is, in absolute-dollar terms, a small fraction of that bill of materials. But if it fails, the accelerator above it fails. That asymmetry, more than the unit economics of the substrate itself, is the foundation of the moat.

The Moat in Hamilton Helmer's Vocabulary

Two of the seven Powers in Hamilton Helmer's framework apply to AT&S simultaneously, and they reinforce each other:

• Switching Costs. Once AT&S is qualified as the substrate supplier under a given accelerator, switching to a competing substrate means re-running the entire reliability qualification on a part that sits beneath a $30,000 piece of silicon. The customer is not going to do that for a 5 percent unit-cost saving. Price becomes a secondary variable in a procurement conversation that is dominated by the cost and the calendar of re-qualification. This is the highest-quality switching cost in the semiconductor packaging stack.
• Cornered Resource. The core dielectric film that the substrate is built on, Ajinomoto Build-up Film (ABF), comes from a single supplier in the world: Ajinomoto. The number of firms capable of running ABF at the leading edge is five. AT&S is one of them; the others are concentrated in Taiwan (Unimicron, Nan Ya) and Japan (Ibiden, and one further major Asian counterpart). Capital intensity, multi-year qualification cycles, and the ABF sole-source upstream mean the five-player oligopoly does not get a sixth member on any realistic timeline.

We covered the broader category of these stacked moats in our explainer on what an economic moat actually is. The closest structural analogue inside our catalog is Frontken (0128.KL): the same shape of qualified-into-a-customer's-process moat in a different layer of the semiconductor supply chain. The closest functional analogue is the US-listed AI-infrastructure compounder Penguin Solutions (PENG), whose moat sits at a different layer of the same AI hardware stack (the memory bottleneck rather than the substrate bottleneck).

The 13 June 2026 Catalyst: Customer-Funded Capacity

On 13 June 2026, AT&S agreed key terms with AMD and a second tech giant to fit out Kulim, Malaysia Phase 2 and a brand-new manufacturing site, presumed to be Phase 3. The total committed build is between $1.7 and $2.3 billion. The financing structure is the part most worth slowing down on.

Capacity at this scale is not being funded off AT&S's balance sheet on speculation that AI accelerator demand persists. It is being funded by the long-term purchase commitments of the customers themselves. AMD and the unnamed second customer are pre-paying for the option to receive supply at a future date, in a form that makes the factory possible to build today.

Customer-funded capacity is rare in industrial hardware. It surfaces when an upstream supplier is strategically essential to a downstream customer whose own product roadmap depends on having confirmed supply. The customer would rather pay forward, even with the financing cost that implies, than risk being gated by the supplier's prudential capex discipline. The structure is, by its existence, the customer's admission that the substrate is load-bearing in the customer's own product plan.

For AT&S, the read is that the moat-on-paper converted into a moat-in-contracts. The customer has voted with capital that the qualification position is worth paying to preserve. The Phase 3 site, on top of the Phase 2 expansion, doubles the optionality: additional capacity that can serve incremental customers without re-qualifying the anchor accounts.

The Capex Triple and the Guidance Lift

Behind the customer signing, management tripled FY26/27 capex from roughly €0.4 billion to between €1 and €1.2 billion. Revenue growth guidance was lifted from a range of 30 to 35 percent to a range of 45 to 55 percent. The EBITDA margin guidance was set at 32 to 37 percent.

CEO Michael Mertin framed the position publicly as AT&S being “the partner of choice for the next wave in growth.” That is the kind of phrase that lives or dies on the next four quarters of execution. The signed customer commitments and the tripled capex tie the company to delivering on it; the guidance lift is the public commitment to investors that management believes the demand visibility now justifies the spend.

Capital Allocation: Every Euro Into Capacity

AT&S sits at the opposite end of the capital-return spectrum from the mature-cash-return compounders in our catalog. The dividend was cut to zero in 2023 and has stayed there. There are no buybacks. Every euro of operating cash flow is being directed into the next tranche of capacity.

The right way to read that policy is not as an absence of shareholder alignment, but as a deliberate choice between two forms of it. A board that returns cash is signalling that the reinvestment runway is exhausted. A board that retains and reinvests every euro is signalling that the reinvestment runway is the most valuable use of capital available, full stop. Five qualified leading-edge substrate makers in the world, customer-funded Phase 3 capacity, AMD on a long-term commitment, ABF sole-sourced upstream: this is the rare case where the no-dividend stance reads as conviction rather than as cash discipline.

The insider signal sits alongside the capital allocation. Two insiders bought near multi-year lows in December 2025. Those positions are now up roughly 6x. Insider purchases at depressed prices from the individuals closest to the customer pipeline are the highest-quality bottom-calling signal in equity research. They do not prove the thesis is right; they do indicate that the people inside the company believed the qualification position was worth a great deal more than the market was willing to pay in late 2025.

The substrate sits under a $30,000 AI accelerator. Once you are qualified, switching means re-testing reliability on a part beneath very expensive silicon. Nobody does that lightly. Price becomes secondary.

The MoatMap Scorecard: Q44 V41 M79, StockRank 80

Here is the AT&S MoatMap StockRank:

• Quality: 44/100. Modest. The trailing operating metrics are still carrying the 2024-trough mix. The Quality factor reads the historical financials before the customer-funded expansion shows up in margins, which it will not until FY26/27 begins to deliver.
• Value: 41/100. Modest. After the 11x rerating, AT&S no longer screens as cheap on trailing multiples. The Value factor reflects that the market has already begun to price in the AI-infrastructure repositioning; the question is whether the next leg of forward earnings power gets there fast enough.
• Momentum: 79/100. Strong. The 11x rerating is the single largest single-name Momentum signal in the European mid-cap universe right now, and the June 13 AMD signing reset the underlying narrative on top of the rerating. The factor is reading both layers.
• Composite StockRank: 80/100. Buy on the composite. Momentum carries the rank, with Quality and Value still rebuilding from the trough. The opposite shape from the cash-rich derated Greater-China names elsewhere in our catalog.

We covered how to read this Momentum-led / Quality-rebuilding profile in our guide to factor investing. Worth noting separately: AT&S only joined the MoatMap ranked universe this week, as part of our six-market expansion that added Austria, Ireland, Portugal, South Africa, Greece, and New Zealand. ATS.VI is the most-watched single name in the Austrian slice of that expansion.

The Question Worth Sitting With

Every long thesis on AT&S has to answer the same question honestly.

Customer-funded factories look great. They also tie AT&S's fate to a small number of very large customers. The same key-customer dependence triggered a profit warning in 2024 when one anchor customer pulled forward an order shift. The next time a single customer's product roadmap stumbles, AT&S will be exposed to the same dynamic, in larger absolute dollars because the contracted volumes are bigger and the dedicated capacity is purpose-built.

Will AT&S continue to diversify its client base?

The bull read says that the June 13 signing already converts one-customer dependence into two-customer dependence (AMD plus the unnamed second tech giant), that the Phase 3 site is the structural answer (capacity that does not yet have a name on it but will), and that the ABF-bottleneck plus the five-firm oligopoly means every new leading-edge AI customer eventually has to qualify with one of the five. Diversification will arrive mechanically as the industry scales.

The bear read says that customer concentration is not a problem you solve by adding a second customer; it is a structural feature of the leading-edge substrate market, that the qualification cycle that makes the moat strong also makes new customer onboarding slow, and that the next downturn in AI infrastructure spending hits AT&S harder than the diversified peers because the dedicated capacity cannot easily be redirected to lower-end products. The 2024 profit warning is, in the bear read, the template rather than the exception.

Both reads are defensible. The factor framework, with StockRank 80 and Momentum 79, is weighting the bull read heavily. The insider purchases that called the December 2025 bottom are weighting the bull read heavily. The market, by paying 11x in one year, has already paid for a meaningful portion of the bull outcome.

Companion Reading

AT&S sits inside our growing AI-infrastructure cluster with some natural neighbours:

• Penguin Solutions (PENG) for the closest functional analogue: a different layer of the same AI hardware stack, the same qualified-into-a-customer's-process moat shape, the same bottleneck-of-the-roadmap framing.
• Frontken (0128.KL) for the closest structural cousin: an Asian semiconductor supply-chain compounder with the identical Cornered-Resource-plus-Switching-Costs moat shape, qualified into TSMC's chamber cleaning process the way AT&S is qualified into AMD's substrate process.
• Plover Bay (1523.HK) for the small-cap-to-mid-cap rerating arc shape: a hardware-bottleneck moat compounding through a product-cycle inflection, with founder/operator capital allocation choosing reinvestment over distribution.

The Bottom Line

AT&S is the clearest example in our catalog of a substrate-layer AI-infrastructure compounder where the customer is willing to pay forward for the right to supply. The moat is the qualification position underneath very expensive silicon. The catalyst was the June 13 AMD and second-tech-giant signing. The capex triple and guidance lift are the public commitment to executing on the contracted demand. The dividend-to-zero policy is the deliberate choice to compound capacity rather than return cash. The insider buying called the December 2025 bottom.

The honest risk is the same risk that triggered the 2024 profit warning: concentrated customer dependence in a market structurally short of qualified suppliers. The bull case is that the June 13 signing is itself the partial answer, and that the Phase 3 site is the rest of it.

For investors considering AT&S as a single-name position, our guide to reviewing your portfolio for weak spots is the right framework for sizing a Momentum-led AI-infrastructure name where the rerating has already taken some of the upside off the table but the qualification moat is genuinely durable.

For the full breakdown including the AMD contract terms, the Kulim Phase 2 and Phase 3 capex schedule, the ABF supply-chain mechanics, the FY26/27 margin walk, and the insider trade history, the AT&S Deep Dive is the place to go.

Credit where due: thanks to Fenix Vanlangerode for being early on this name, with a substrate thesis published near €18 in mid-2025. Worth reading.

This article is for informational purposes only and is not investment advice. The author may be long names covered on MoatMap.