Amkor reported a record quarter, beat earnings by nearly 50%, and lost a quarter of its market value the next day. The market did not reject the quarter. It rejected one sentence about 2027 — and in doing so priced the shares slightly below what the cash flows support.
The author holds no position in AMKR as at the date of publication, and held none while the research was produced. No position will be taken for 72 hours after publication.
The author has received no compensation from Amkor Technology or any affiliate, holds no position exceeding 0.5% of the issuer’s share capital, and has no business relationship with the company. This report was not shown to the issuer before publication.
The selloff was about one sentence. Revenue guidance missed by roughly 4% while EPS guidance beat by 15%. The market punished the revenue line, which tells you investors are underwriting a growth story here, not an earnings story.
The damage is self-inflicted and temporary. Half the communications shortfall is a deliberate relocation of SiP production from Korea to Vietnam. Moving a factory is a cost you choose, on a schedule you control.
Concentration is the real risk, not guidance. Apple was 29.8% of FY2025 sales. A transfer of one customer’s production line moved $3.7bn of market value in a day — which is the concentration risk demonstrating itself, not a guidance problem.
Everything difficult about owning Amkor traces back to a single number. Per the FY2025 Form 10-K, the ten largest customers were 72% of net sales, with Apple alone at 29.8% and Qualcomm at 11.1%. In the second quarter of 2026 the top ten fell to 66%, which is genuine progress driven by the computing ramp — but the structure has not changed.
This matters more than any multiple. When a single customer changes where a product is built, Amkor absorbs the transition cost, the idle capacity and the timing gap. That is precisely what happened this quarter, and it is why a company that beat earnings by 50% lost a quarter of its value the next day.
Amkor generates cash and has done so through multiple cycles, so we value it on unlevered free cash flow rather than on an asset base. Every assumption is exposed below. The single most important one is capital intensity: this is a business that must keep spending to hold its position, and the gap between EBITDA margin and capex is where the equity value actually lives.
Terminal value is 78% of enterprise value. Anyone telling you a DCF on a cyclical is precise is selling something.
Tax held at 15%, reflecting Amkor’s jurisdictional mix. Net cash held at $0.4bn. Both are derived estimates and should be reconciled against the next 10-Q before being relied upon.
Same model, different inputs. Click any card to load it and see which assumption is carrying the weight. The spread here is wide because it is a cyclical with a genuine structural tailwind — narrow ranges on businesses like this are a sign of false confidence, not rigour.
Amkor is an OSAT — outsourced semiconductor assembly and test. Chipmakers design and fabricate silicon; Amkor takes the finished wafer, cuts it, packages it, connects it and tests it. It is the second-largest firm in an industry where the top ten companies generated roughly $41.6bn of revenue in 2024, with ASE Technology at about 45% of that group.
For most of its history this was a commodity business. That changed when transistor scaling slowed and cost per transistor stopped falling reliably. The industry’s answer was to stop building one large monolithic die and instead stitch several smaller chiplets together inside a single package — 2.5D interposers, high-density fan-out, silicon bridges, stacked high-bandwidth memory. Packaging went from the last and cheapest step in the process to a genuine performance bottleneck.
That is the whole investment case. Amkor currently has five HDFO customers with ten active engagements and four products launching in 2026, plus eleven 2.5D customers with four launches. A commodity subcontractor does not have a design-win pipeline. This one now does.
| Metric | Q2 2026 | Q1 2026 | Q2 2025 | Why it matters |
|---|---|---|---|---|
| Net sales | $1,898m | $1,685m | $1,511m | Record. Up 26% year on year and 13% sequentially. |
| EPS | $0.70 | — | — | Against roughly $0.47 consensus — a 49% beat. |
| Gross margin | +250bp | — | — | Sequential expansion, driven by the advanced packaging mix. |
| Q3 revenue guide | $1.95–2.05bn | — | — | Versus roughly $2.09bn expected. This is what was sold. |
| Q3 EPS guide | $0.72–0.82 | — | — | Versus roughly $0.62–0.64 consensus. Comfortably above, and ignored. |
Communications will decline high-single-digits sequentially in Q3, breaking the normal seasonal pattern of a peak iOS build quarter. Management split the cause roughly evenly between memory supply constraints with changed build patterns, and the deliberate relocation of SiP production from Korea to Vietnam. Then the CEO said the drag “is probably going to extend into Q4 and even into the first half of next year.” The stock had already fallen 6.5% during the session before the release and drifted 0.4% lower after hours. It lost 24.8% the following day.
The sequencing is the tell. The market did not reject the quarter — it rejected the shape of 2027. And the asymmetry between the revenue guide and the EPS guide says something specific about the shareholder base: a company beating on earnings and missing on revenue gets sold only when its owners are paying for growth rather than for cash generation.
| End market | % of revenue | Sequential | Detail |
|---|---|---|---|
| Computing | 22% | +20% | AI data centre, HPC, HDFO data-centre CPU ramp. Record quarter. |
| Automotive & industrial | 22% | +17% | ADAS, electrification, infotainment. Record quarter. |
| Consumer | 14% | +15% | IoT, wearables, connected home, AR and gaming. |
| Communications | — | — | iOS ecosystem up double digits; Android down 20% on memory shortages. The segment guided to decline in Q3. |
The mix shift is the story the selloff obscured. Computing and automotive together are 44% of revenue and both set records. Communications — the segment with the concentration problem and the SiP transfer — is the one shrinking. If you believe the transfer is temporary, the business is moving in the direction you would want.
Demand is pulling 2.5D and HDFO capacity faster than it can be built. That is a seller’s market for the handful of firms that can do the work.
US restrictions prevent Chinese OSATs such as JCET from buying the tools needed for the highest-end AI packaging. That structurally reserves the top of the market for ASE, Amkor and TSMC.
The Arizona campus carries $400m of proposed CHIPS Act direct funding, access to $200m of CHIPS loans, and an Advanced Manufacturing Investment Credit whose rate rose to 35% for qualified property placed in service after 2025.
TSMC’s CoWoS and SoIC capture the most attractive AI packaging work inside the foundry. The ten-year agreement makes TSMC a partner and a competitor at once, and that tension is unresolved.
Communications returning to normal seasonality by Q2 2027 as guided; disclosed volume under the TSMC agreement or the NVIDIA partnership; Arizona reaching qualification ahead of schedule; or the top-ten customer share falling below 60% without a revenue decline.
The SiP drag extending beyond the first half of 2027; any indication Apple is moving share to a competing OSAT rather than relocating within Amkor; capex intensity rising above 13% of revenue without a matching margin response; or TSMC taking 2.5D volume back in-house as CoWoS capacity expands.
We rate AMKR fairly valued. The cash flows support roughly $50 against a $45.68 quote — close enough that this is not a mispricing to act on, but it is emphatically not the broken company the tape suggested on 28 July.
Other companies design computer chips and print them onto silicon wafers. Amkor does the step after that: cutting the wafer up, wrapping each chip in a protective package, wiring it so it can talk to the rest of the device, and testing that it works. It is the finishing shop of the chip industry.
For decades chips got faster by making the transistors smaller. That has largely stopped working. So the industry now builds several smaller chips and joins them together inside one package instead. The joining-together part is exactly what Amkor does, which means a boring final step has become one of the harder and more valuable ones.
The results were excellent. But the company told investors that next quarter’s sales would be lower than expected, partly because it is moving one production line from Korea to Vietnam, and that this would drag on into 2027. Investors who had been paying for fast growth decided they were wrong about the speed, and sold.
Nearly a third of Amkor’s sales come from a single customer. When that customer changes how or where its products are made, Amkor feels it immediately and has little say in the matter. That is the risk you are taking. The offsetting opportunity is that AI chips need exactly the advanced packaging Amkor has spent years building.
None of this tells you whether to buy it. Our arithmetic says the shares are worth roughly what they cost today — so the question is not whether it is cheap, but whether you believe the factory move is temporary.