Business

Know the Business — RS Technologies (3445)

Figures converted from JPY at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.

RS Technologies is a three-businesses-in-one trench coat: a globally dominant reclaim-wafer specialist (37% margin, ~36% of revenue, ~71% of segment operating income), a sub-scale Chinese prime-wafer challenger that is China-only and subsidy-supported, and a bolted-on equipment / camera-module / battery rollup running at 5% margin. The market keeps treating consolidated FY2025 numbers as one cyclical wafer story; the reclaim engine inside is far higher quality than the blended P&L suggests, and the prime-wafer arm is far more China-policy-dependent than the segment label implies.

The right question is not "what's RST worth at a wafer-cycle multiple," but "what would each of these three businesses be worth standing on its own, and where is consolidated reporting hiding value or risk."

1. How This Business Actually Works

Takeaway: Reclaim is a near-defensive chemistry-and-polish business whose feedstock is essentially free (a customer's used wafer). Prime is a capital-heavy Chinese fab supplier whose feedstock is polysilicon and whose margin is tied to Chinese fab utilization. Equipment is a thin-margin rollup. The economic engine is reclaim, not wafers in aggregate.

A foundry like TSMC or a memory IDM like Samsung needs three kinds of wafers: prime wafers that become chips, plus test wafers and monitor wafers (the ~20% of wafer consumption that polices the process). Around 80% of those test wafers are recycled rather than bought new, because a reclaimed wafer sells for one-quarter to one-half of a new test wafer. RST takes the customer's scrap, strips off prior films with proprietary chemistry, re-polishes it to spec, and sells it back. Feedstock is free. The whole margin is in chemistry, polish allowance, and how many times you can recycle a single wafer.

The other two businesses run on different mechanics. Prime wafers (GRITEK in China for 8-inch power-semi wafers, plus a 39%-owned equity-method affiliate SGRS that is still pre-commercial in 12-inch) buy polysilicon, pull crystal ingots, slice and polish virgin substrates, and sell into mostly-Chinese fabs at competitive prices with state support. The Equipment & Materials segment is a roll-up: DG Technologies (silicon consumables for dry-etch tools), Union Electronics (trading/agency), RSPDH (optical pickup and in-car camera modules — acquired 2024, generated a one-off negative goodwill gain), and LE System (vanadium-redox-flow battery electrolyte). FY2025's revenue jump in this segment was almost entirely RSPDH's first full year of consolidation.

Reclaim barely flexes through cycles — 35.1% in the worst year (2020 COVID disruption), 40.6% at the 2022 peak. Prime swings 11.9% → 26.4% over the same window. Equipment trades at single digits.

Where incremental profit comes from. When Sanbongi and Tainan are at full utilization (which they were all four quarters of FY2025, per the Q4 briefing), every additional wafer slot adds revenue at near-segment margin because feedstock is free and the variable cost is mostly chemistry and labor. That is why management is bringing Sanbongi Plant 7 forward a year (now FY2026, was FY2027) and adding 70K wafers/month of new monthly reclaim capacity in 2026 — they have customer LTAs pre-signed and they cannot ship enough today.

Where margin gets pressured. Prime-wafer ASP is now the swing factor. In FY2025 8-inch shipment volume rose 20% YoY but average ASP fell about 10% YoY (general-purpose IGBT and memory wafers in China are the weak slice; power-semi wafers held up). Equipment margin is structurally low because optical-pickup modules and trading-company revenue have nowhere near reclaim economics.

2. The Playing Field

Takeaway: RST has no clean listed peer. The four prime-wafer giants are bigger, more cyclical, and operate the part of RST that is least profitable. Ferrotec is the only listed company whose mix even loosely resembles RST. The peer table shows RST cheap on multiples — but every peer is dominated by prime wafers, while RST's earnings power lives in reclaim. The "cheapness" is partly mismeasurement.

Peer figures sourced from Yahoo Finance quote-summary 2026-05-15 (TTM revenue and EBITDA), reported here in USD via Frankfurter/ECB rates. RST market cap from ¥6,640 close × 26.46M shares = ¥175.7bn ≈ $1,108M; EV is market cap less net cash of ~$485M; EBITDA is operating income $91M + D&A $35M = $126M. Multiples (EV/Sales, EV/EBITDA, EBITDA margin) are unitless and unchanged across currencies.

Three things the peer set reveals:

(1) RST is the cheapest in the group on EV/Sales (1.3x) and EV/EBITDA (5.0x), but cheapness here is not yet a mispricing. Half its revenue is the low-quality Equipment segment, which deserves an EV/Sales closer to 1x, and the Prime segment is at a Chinese-policy multiple. The pure reclaim engine is what could re-rate the stock; it is currently buried inside a blended multiple.

(2) Reclaim has no listed pure-play. Mimasu Semiconductor and the Taiwanese reclaim names are either too small or too diversified for a clean peer. Ferrotec is the only relevant comp — and it trades at 2.1x EV/Sales / 10.9x EV/EBITDA with worse mix (reclaim is only a slice of its business) and 13% segment-level op margins. Carving RST's reclaim out standalone would allow marking it against semi-materials specialists, not against SUMCO and Siltronic.

(3) The prime-wafer peer set shows what good and bad look like. Shin-Etsu and SUMCO together set 300mm wafer pricing globally and earn 9–28% operating margins depending on cycle position. RST's prime business runs 20% — credible inside the Chinese market but with no realistic path to global 300mm scale. GlobalWafers' 28x EV/EBITDA tells you what equity markets will pay for a clean prime-wafer pure-play in an upcycle — and how much that multiple can reset down when shipments slow.

3. Is This Business Cyclical?

Takeaway: Reclaim is the closest thing in semis to a defensive segment — in the 2019 memory glut and 2020 COVID year, reclaim margin actually rose (38–38%). Prime is fully cyclical. The blended P&L looks moderately cyclical; the right way to read RST through a downturn is reclaim flat-to-up, prime down 1,000 bps, equipment unchanged at thin margins.

The mechanism is counterintuitive. In a downturn, fabs cut wafer starts → they cannibalize their internal pool of test wafers more aggressively → they actually increase the share of test wafers that are reclaimed (because reclaim is cheaper than buying new test wafers). RST's reclaim volume holds; pricing is sticky because chemistry is qualified per customer. Meanwhile prime-wafer customers slow purchases, ASP softens 5–15%, and prime-segment op margin compresses by ~1,000 bps.

Two patterns the chart makes visible:

(a) Reclaim margin range FY2017–FY2025: 35.1% → 40.6%, a 5.5-point band. That is a quality semiconductor-materials margin, narrower in range than any IC manufacturer. The high point was the 2022 AI/memory peak; the trough was 2020 COVID. In the 2023 mature-node downturn — when SUMCO and Siltronic saw EBITDA fall double digits — RST's reclaim segment dipped only 100 bps from FY2022.

(b) Prime margin range FY2018–FY2025: 11.9% → 26.4%, a 14.5-point band. That is a fully cyclical wafer business. The 11.9% trough in FY2020 was driven by Chinese 8-inch IGBT ASP weakness and capacity ramp inefficiency; the 26.4% peak in FY2022 was the AI-server pre-build. FY2025's 19.9% with 8-inch ASP down ~10% YoY suggests we are mid-cycle on the down-leg.

The other cycle exposure is working capital and capex. RST's gross cash swung from $515M in FY2022 → $543M in FY2024 → $617M at FY2025 year-end while long-term debt rose to $90M because management is pre-funding the ~$359M capex plan for 2026–2028; net cash (cash less total debt) was ~$485M at FY2025 year-end. The business throws off enough cash ($95M operating CF FY2025, $47M FCF) to self-fund, but the next three years' growth depends on a customer demand forecast that has to materialize — if it doesn't, the new Sanbongi Plant 7 / Tainan Plant 2 / China lines come up at low utilization.

4. The Metrics That Actually Matter

Takeaway: Ignore consolidated revenue growth. The five things that drive value here are reclaim segment margin, reclaim capacity utilization, 8-inch prime ASP, the SGRS 12-inch equity-method loss, and ROIC trajectory. If you only watch one, watch reclaim segment operating margin — that's the one the consolidated number is most likely to obscure.

Relative health scored 0–1 within RST's own range — 1.0 marks the peak reading for that metric over the window.

Notice what the table is not showing: revenue growth and gross margin. Headline revenue grew 30% YoY in FY2025 — almost entirely RSPDH's first full-year consolidation (camera modules), which is the lowest-quality revenue in the mix. Gross margin compressed 200 bps for the same reason. Neither number tells you whether the business is getting better or worse. Reclaim margin and ROIC do.

5. What Is This Business Worth?

Takeaway: Consolidated multiples mismeasure this company. The right lens is sum-of-the-parts because the three segments have wildly different margins, growth profiles, capital intensity, geographic exposure, and acquisition multiples. The reclaim engine deserves a quality-materials multiple. The prime business deserves a China-policy multiple. The equipment business deserves a thin-margin diversified-industrial multiple. There is also an embedded listed stake — GRITEK on the Shanghai STAR Market (688521.SH) — that complicates the picture and should be sized before forming any consolidated valuation view.

The practical implication for an underwriter: when you see a "5x EV/EBITDA" headline for RST, ask immediately whose EBITDA. If you mentally split the FY2025 $91M segment EBIT into $65M (reclaim) + $27M (prime) + $10M (equipment) − $11M (other), and apply roughly 12x / 6x / 5x respectively, the math lands at ~$780M + $160M + $50M − $55M ≈ $925M of operating-business value, plus $485M of net cash, ≈ $1,410M of equity. At ~$42 per share the market cap is ~$1,121M. That suggests a ~25% gap to a clean SOTP — roughly half the reclaim multiple the market is not paying for, and roughly half the net cash that is earmarked but not yet deployed. It is also why this name's biggest swing factor is not next quarter's prime ASP but whether reclaim margins hold above 38% through the FY2026–28 build-out.

6. What I'd Tell a Young Analyst

The trap with RST is that consolidated metrics blend three businesses you would never value the same way. Read every press release with that filter. When net sales jumped 30% in FY2025, almost all of it was the RSPDH camera-module roll-on — the lowest-quality slice of revenue and the part most likely to fade once the negative-goodwill year is behind it. When ROIC fell from 12.7% to 10.8%, that's the capex pre-build phase, not a structural deterioration — wait for FY2028 readings before judging.

Watch four things, in order. (1) Reclaim segment operating margin quarterly — the company discloses it; if it stays above 38% while Sanbongi Plant 7 ramps, the quality-compounder thesis is intact. (2) 8-inch prime ASP commentary — management calls out the gap between power-semi pricing (firm) and IGBT/memory-grade pricing (weak), and this is the leading indicator for prime segment margin. (3) Sanbongi Plant 7 + Tainan Plant 2 capex disbursement — $109M Sanbongi + $156M Tainan committed over 2026–28; slippage or scope creep is the first sign customer demand has softened. (4) GRITEK's standalone Shanghai disclosures (688521.SH) — quarterly results there are the cleanest read on the prime segment without RST's HQ-allocated overhead distorting it.

The thesis breaks if: reclaim margin cracks below 35% structurally (chemistry cost-up, customer reclaim-cycle counts shifting), Sanbongi Plant 7 ramps at low utilization (demand miss), or US-China escalation forces GRITEK to write down its China prime book. The thesis works if: reclaim margin holds at 37–39% through the build-out, FY2027–28 revenue clears ~$660M with mid-teens operating margin, and management starts buying back stock when net cash crosses $500M again. Everything else is noise.

The mid-term plan published in February 2026 — $529M / $662M / $725M revenue and $97M / $110M / $120M op income for 2026 / 2027 / 2028 — implies operating margin grinding down from 18.6% to 16.5% as the rollup grows faster than reclaim. Track whether the mix actually plays out that way. If reclaim outgrows the plan, op margin should beat. If equipment outgrows the plan, op margin meets the plan but the multiple does not change. Same revenue, different P&L quality, totally different equity value.