한미반도체 — Research Report (2026-07-18)

Hanmi Semiconductor · KRX: 042700 · researched via web (no US filing; DART is the primary Korean source)

TL;DR

• What it is: the world's #1 maker of "TC bonders" — the machine that stacks DRAM chips into HBM, the memory that feeds every AI GPU.
• Overall take: a genuine bottleneck monopolist riding the AI wave, but priced for perfection and losing share — a great business at a scary price.
• 2025 revenue ₩5,767억, op. margin 43.6% (record year, best-in-class profitability).
• 71.2% global TC-bonder share (dominant, but eroding — see bear case).
• Net cash ≈ ₩2,762억, debt ratio 17.8% (basically no debt — can't go bankrupt).
• Trailing P/E ~108×, forward ~29–67× (expensive on any measure vs semis ~20–28×).
• Strongest bull: HBM4 is ramping and Hanmi is SK hynix's lead bonder supplier → order surge in H2 2026.
• Strongest bear: SK hynix is dual-sourcing (ASMPT already ~half of HBM4 bonders); Hanmi's share of that account may fall to 20–30%.
• One thing to watch next: the HBM4 volume split between Hanmi and ASMPT at SK hynix — that ratio is the stock.

1. Business overview

So what: Hanmi makes the single most critical assembly machine for HBM — and until 2024 it had a near-monopoly on it.

Hanmi Semiconductor builds back-end semiconductor equipment — the machines used after a chip is fabricated, to cut, stack, bond and package it. Its crown jewel is the TC bonder (Thermal Compression bonder): a machine that presses DRAM chips onto each other under precise heat and pressure to stack them vertically.

That stacking is exactly how HBM (High Bandwidth Memory) is built — the tall "skyscraper" of DRAM chips that sits next to an AI GPU (like Nvidia's) to feed it data fast. No HBM without bonding; no leading-edge AI accelerator without HBM. That is Hanmi's position in the value chain.

Jargon check. Back-end = the packaging/assembly half of chipmaking (vs "front-end" = the fab where transistors are etched). DRAM = the standard memory chip. HBM = many DRAM chips stacked into one tall high-speed module. TC bonder = the machine that does the stacking with heat + pressure.

The revenue model matters: Hanmi sells big machines in batches when a customer expands HBM capacity, then goes quiet between expansion cycles. This makes quarterly results violently lumpy (proven below).

2. Financials — 3-to-5-year picture

So what: 2025 was a record blowout, but Q1 2026 was an "earnings shock" — this is a cyclical, order-driven business, not a smooth grower.

Annual

Multi-year 2021–2024 line items weren't cleanly available on the pages I could fetch (Naver/WISEreport summary pages omit them) — verify on DART's 사업보고서 before acting. The shape is clear from coverage: Hanmi was a modest ₩1,000–2,500억 revenue tool-maker until the 2023–2025 HBM boom roughly doubled-to-tripled its revenue and pushed margins to record highs (43.6% in 2025). (파이낸셜뉴스, 한국일보)

(A) = actual, reported. Consensus = analyst estimate, labeled.

The cyclicality, made visible (quarterly)

This table is the whole risk in one place:

Revenue fell -65.5% YoY in Q1 2026 and op. profit -87.9% YoY, then op. profit rose +51% the very next quarter to an all-time high. (헤럴드경제, 한국경제, 파이낸셜뉴스)

Plain-English lesson: Hanmi's earnings gap up and down with the timing of customer HBM-capacity expansions. A weak quarter is often a transition air-pocket between two chip generations (HBM3E finishing, HBM4 not yet ramping), not a broken business — but you cannot value it on one quarter.

Cash & balance sheet

(judal/valueline via search, economic22)

Operating cash flow / capex / FCF: clean multi-year OCF and capex lines weren't available on the fetched summary pages — pull from DART 현금흐름표 to complete this. Directional read: with 40%+ operating margins, a near-debt-free balance sheet, ₩2,762억 of cash, and the ability to pay a record ₩759억 dividend (below), Hanmi is clearly a strong cash generator — but the exact FCF and its conversion (FCF/EBITDA) are the one gap in this report. (estimated — flag for DART verification)

3. Unit economics — the one number the business runs on

So what: ~₩30억 per TC bonder × how many machines a customer's HBM ramp needs. Count the machines, you've forecast the revenue.

The business runs on machines shipped × price per machine:

(네이트 공시, TrendForce)

Because one contract can equal ~8% of annual sales, order disclosures (단일판매·공급계약) on DART are the highest-signal thing to watch — each one is a chunky, forecastable revenue slug. This is the §0 "backlog vs market cap" lens: watch the order flow, not the quarterly noise.

4. Metrics table — with the "why it matters here"

Jargon: P/E = price ÷ per-share earnings. PBR = price ÷ per-share book (net asset) value. ROE = net income ÷ equity, i.e. return on shareholders' money. EBITDA = operating profit before depreciation/amortization — a rough cash-earnings proxy.

4b. Sector valuation benchmarks — is it cheap or expensive?

So what: on every yardstick, Hanmi trades far above the semiconductor-equipment sector — it is priced as a scarce monopoly, and the whole niche is priced for a multi-year HBM boom.

Which metric: Hanmi is profitable and asset-light with net cash, so P/E is the right lens (not EV/EBITDA — no leverage to neutralize; not EV/Sales — well past pre-profit).

Source for sector range: broad semiconductor / semi-cap-equipment P/E screens (Damodaran sector data, semi-ETF medians). ASML/BESI/AMAT sit ~25–40× forward at cycle mid-points; the S&P semis composite historically ~20–28×.

Comparability caveat: the cleanest peers are BESI (hybrid-bonding leader) and ASMPT (Hanmi's direct TC-bonder rival) — both also carry AI/HBM premiums. So Hanmi isn't uniquely mispriced within its niche, but no clean cheap comp exists — every HBM-bonding name is expensive; the question is who keeps the share.

What a reasonable buyer might pay: a strategic acquirer valuing the monopoly-that-was might justify 30–40× forward earnings on a durable 40%+ margin. But at ~108× trailing / high-20s-to-60s forward, the market has already priced in HBM4 success and assumed Hanmi holds its share. Given the dual-sourcing erosion (below), a disciplined buyer would want the price to reflect a 20–30% SK hynix share, not the old 100%. Verdict: expensive, and the expensiveness assumes a monopoly that is already breaking.

5. Ownership, management & insider signals

So what: the chairman owns a third of the company and keeps buying more with his own money — a strong "skin in the game" signal, though governance isn't spotless.

(헤럴드경제, 서울경제, 중앙이코노미)

Read: the chairman repeatedly buying his own stock with personal money — including after a big run-up — is one of the strongest management-conviction signals available. The NPS "against" votes and the wide pay-cap are worth noting but are common in founder-controlled Korean firms; not a thesis-breaker.

6. Value-chain position — bottleneck or commodity?

So what: Hanmi sells a genuine bottleneck tool — but its customer (SK hynix) holds the pricing power, and is deliberately breaking the monopoly by dual-sourcing.

• Bottleneck? Yes, technically. HBM cannot be built without bonding, and Hanmi's TC bonder was the industry standard from 2016. That's why it earns 40%+ margins.
• But who has pricing power over whom? The customer does. SK hynix is a giant; Hanmi's HBM revenue leaned on essentially one buyer for years. A customer that concentrated can — and did — decide to bring in a second supplier to cut its dependence and its costs.
• The erosion, in numbers: SK hynix now runs ~50 TC bonders for HBM4, ~half from ASMPT (Singapore). Analysts see Hanmi's share of SK hynix's bonder business falling toward 20–30%. For Micron's HBM4, tighter placement-accuracy needs may push it fully to BESI — a lost account. (TrendForce, lumenalpha)
• The next war — hybrid bonding: for future ultra-tall HBM (16-high HBM4E and beyond), the industry may shift from TC bonding to hybrid bonding, where BESI leads and Hanmi is a challenger. Hanmi is investing ₩100억 to ship hybrid-bonding equipment by end-2027 — but here it's the underdog. Timing keeps slipping: 16-high HBM4E now looks like the earliest real adoption point. (TrendForce)

Jargon: Dual-sourcing = a buyer deliberately using ≥2 suppliers so no single one has leverage. Hybrid bonding = a next-gen, copper-to-copper stacking method (no solder bumps) for very tall/dense HBM — a different machine where BESI, not Hanmi, currently leads.

The competitive map:

7. Bull / Base / Bear — attributed to named views

So what: the sell-side is genuinely split — some see ₩325k–420k, JPMorgan says "underweight." That disagreement IS the story.

🐂 Bull — Merrill Lynch (BofA), target ₩420,000; 아시아경제/Click e-Stock cites a ₩325,000 target. HBM4 is ramping, Q2 2026 was a record (51.9% margin), and Hanmi remains SK hynix's lead TC-bonder supplier with a fresh ₩442억 order. If HBM4→HBM4E volumes scale and Hanmi holds a meaningful share, 2026 revenue could jump toward ₩8,000억+ and beyond. (아시아경제)

⚖️ Base — CGSI / CGS International (with a ~₩230k–240k consensus band from 유진투자/리딩투자). 2026 revenue ~₩8,040억, op. profit ~₩4,108억, ~51% margin — a strong recovery year off the Q1 air-pocket, but not the ₩2조 (₩20,000억) revenue management floated. The gap between management's ₩2조 aspiration and the ₩8,040억 consensus is itself a caution flag. (더벨)

🐻 Bear — JPMorgan, "Underweight" (Mar 2026). JPMorgan flagged valuation as stretched. The structural bear case: SK hynix dual-sourcing drops Hanmi's share to 20–30%, Micron migrates to BESI, and the eventual hybrid-bonding transition favors BESI — so Hanmi's 40%+ margins are unsustainable and a ~108× trailing / high-20s+ forward multiple has no support. (economic22)

Catalysts (watch these):

• HBM4 order cadence from SK hynix (each 단일판매 disclosure on DART).
• The Hanmi-vs-ASMPT volume split at SK hynix — the single most important number.
• Samsung HBM4 qualification — a genuine "second big customer" would re-rate the stock.
• Hybrid-bonding equipment launch (targeted end-2027) — does Hanmi actually win here?

What would prove the thesis wrong (either way):

• Bull broken if: Hanmi's SK hynix share confirms below ~30% and Samsung stays in-house and hybrid bonding goes to BESI → the monopoly is gone, margins compress, 100× P/E collapses.
• Bear broken if: Hanmi wins Samsung, holds >50% at SK hynix, and ships competitive hybrid bonders on time → it stays the bottleneck and grows into the multiple.

📚 What this company teaches

1. A monopoly's biggest risk is its own customer, not a competitor. Hanmi didn't lose share to a better machine — SK hynix chose to dual-source to escape dependence. When one buyer is >50% of your revenue, they hold the pricing power no matter how good your product is. This is customer-concentration risk — the flip side of the "multi-customer" checklist item.

2. Lumpy capital-equipment earnings must be valued across a cycle, not on a quarter. Q1 2026 (-88% op. profit) and Q2 2026 (record) are the same business — an air-pocket between two chip generations. Judging it on either single quarter is a trap; for tool-makers the order book, not one income statement, tells the truth.

3. In a value chain, watch where the next bottleneck moves. TC bonding was the choke point for HBM3/4; hybrid bonding may be the choke point for HBM4E+. The incumbent of today's bottleneck (Hanmi) is the underdog of tomorrow's (where BESI leads). Margin follows the bottleneck, and the bottleneck moves with technology.

Self-test question: If SK hynix places its next HBM4 bonder order split 50/50 with ASMPT instead of favoring Hanmi, what happens to (a) Hanmi's revenue growth and (b) the justification for its ~108× P/E — and which of the three lessons above does that illustrate?

📖 Glossary

• Back-end (semiconductor) — the packaging/assembly stage of chipmaking (cut, stack, bond, test), after the "front-end" fab that etches transistors. Hanmi is a back-end equipment maker.
• DRAM — the common type of memory chip; the building block stacked into HBM.
• HBM (High Bandwidth Memory) — many DRAM chips stacked vertically into one tall module, placed beside an AI GPU to feed it data at very high speed.
• HBM3E / HBM4 / HBM4E — successive generations of HBM; each new generation triggers a fresh wave of equipment orders.
• TC bonder (Thermal Compression bonder) — the machine that stacks/bonds DRAM chips using heat and pressure; Hanmi's flagship product.
• Hybrid bonding — a next-generation, solder-free (copper-to-copper) stacking method for very tall/dense HBM; BESI currently leads it, Hanmi is a challenger.
• Dual-sourcing — a buyer deliberately using two or more suppliers so no single supplier holds pricing leverage. What SK hynix is doing to Hanmi.
• P/E (price-to-earnings) — share price ÷ earnings per share; how many years of current profit you pay for the stock. High = growth priced in (or overvaluation).
• PBR (price-to-book) — share price ÷ net asset value per share; >1 means the market values the business above its accounting net worth.
• EV/EBITDA — enterprise value (equity + debt − cash) ÷ EBITDA; a debt-neutral whole-company valuation. Less useful here because Hanmi has net cash.
• EBITDA — operating profit before depreciation and amortization; a rough proxy for operating cash generation.
• FCF (free cash flow) — operating cash flow minus capex; the cash left for dividends, buybacks or debt paydown. (Exact figure to be pulled from DART for this name.)
• OCF (operating cash flow) — cash generated by the core business before investing/financing.
• Net debt — total debt minus cash. Negative for Hanmi = net cash = financially anti-fragile.
• Net cash — the company holds more cash than debt.
• ROE (return on equity) — net income ÷ shareholders' equity; the return on owners' money.
• Operating margin — operating profit ÷ revenue; here it doubles as a pricing-power gauge.
• Order backlog / 단일판매·공급계약 — contracted future revenue; on Korea's DART, large single supply contracts must be disclosed — the highest-signal event for a lumpy equipment maker.
• DART — Korea's electronic corporate-disclosure system (dart.fss.or.kr), the primary source for Korean filings.
• NCF (Non-Conductive Film) — a bonding-material approach Samsung uses in-house; part of why Samsung self-supplies rather than buying Hanmi bonders.
• SEMES — Samsung's in-house equipment subsidiary.
• BESI / ASMPT / Hanwha — rival bonding-equipment makers (Netherlands / Singapore / Korea respectively).

Sources: 파이낸셜뉴스 · 한국일보 · TrendForce (SK hynix–ASMPT split) · TrendForce (hybrid-bonding timeline) · 네이트 공시 (₩442억 order) · 헤럴드경제 (Q1 쇼크) · 헤럴드경제 (insider buying) · 아시아경제 (target price) · economic22 (valuation/JPM) · lumenalpha (supply chain).

⚠ Verify on DART before acting: 2021–2024 annual line items, and OCF/capex/FCF (the one data gap). Numbers labeled (A)=reported, consensus=estimate. Not financial advice — a personal research note.

I tried to save this to investing/042700-2026-07.md (per the workflow's file convention) but the write needs your permission — want me to save it? One honest gap to flag: the multi-year 2021–2024 line items and clean OCF/capex/FCF weren't on the summary pages I could reach, so those are marked for DART verification rather than guessed.