Earnings Call Highlights: Amkor Technology (AMKR) | 2026 Q2

2 days ago

7 min read

1. This is an "AI advanced packaging" earnings report, not a phone packaging house's report. Amkor set a quarterly revenue record and gross margin jumped, with the real engine being record highs in computing and automotive/industrial, not its traditionally largest segment, communications. Management even let Q3 communications decline "counter-seasonally" to shift capacity toward AI.

2. Two long-term agreements tie Amkor to the core of the AI supply chain: a 10-year advanced packaging agreement with TSMC and a multi-year strategic collaboration with NVIDIA. With CPO (Co-Packaged Optics) formally added to its technology platform for the first time, Amkor is moving from "behind-the-scenes assembler" onto the critical path of AI system performance.

3. Full profit leverage: EPS more than doubled YoY. First-half operating income doubled and EPS tripled, showing how much financial flexibility this business has once utilization rises and the mix shifts toward advanced packaging. But depreciation from the new Arizona fab in 2027/2028 is the next headwind, as management itself flagged.

This article is for technology and industry trend analysis only and does not constitute investment advice.

1. Three signals to understand first

First, the source of growth has changed. Amkor's hallmark has been communications (SiP packaging, mainly for iPhone), but this quarter's records came from computing and automotive/industrial. Management even proactively guided Q3 communications revenue down; this is not collapsing demand but a "choice" to free up Korean fab capacity for fast-growing computing programs.

Second, Amkor is being locked into the AI supply chain through long-term agreements. The 10-year agreement with TSMC and the multi-year collaboration with NVIDIA show that advanced packaging is now on the critical path of system performance; customers want early co-development plus long-term capacity commitments, not quarter-by-quarter orders. That kind of visibility did not exist in past packaging industry cycles.

Third, CPO has entered Amkor's official technology vocabulary for the first time. This is the most important point for STT readers, covered in detail below.

2. Revenue and financials

Core quarterly P&L:

• Revenue of US$1,898M (about $1.9B), YoY +26%, QoQ +13%, a quarterly record and above the top of guidance.

• Gross margin of 16.8%, up more than 250 bps QoQ and about 4.8 percentage points YoY (12.0% a year earlier); gross profit of $319M, up 33% QoQ.

• Operating income of $200M, operating margin of 10.5% (including a $21M one-time gain on a real estate sale).

• Net income of $174M (just $54M a year earlier).

• Diluted EPS (GAAP) of $0.70 ($0.22 a year earlier, more than doubling YoY).

• EBITDA of $400M, EBITDA margin of 21%.

What the numbers mean matters more than the numbers themselves: first-half revenue up 26% YoY, gross margin up 360 bps, operating income doubled, EPS tripled. This is the financial leverage an asset-heavy packaging house should show when utilization rises from the "50s%" to the "70s%." Management said several technology platforms are fully loaded, meaning this wave is driven by mix upgrade, not price cuts for volume.

The balance sheet strengthened too: cash and short-term investments of $2.5B at quarter end, total liquidity of $3.6B, total debt of $2.5B, and debt/EBITDA of just 1.8x; in May it issued $1.15B of 0% convertible notes, lining up funding for further expansion (especially Arizona).

Quarterly revenue by end market:

• Communications 42% (about $797M): QoQ +6%, with double-digit growth in the iOS ecosystem but Android down 20% QoQ (pressured by both memory supply and demand).

• Computing 22% (about $418M): QoQ +20%, a quarterly record, from a broad data center customer base.

• Automotive & Industrial 22% (about $418M): QoQ +17%, also a record, with ADAS as the main driver.

• Consumer 14% (about $265M): QoQ +15%, driven by IoT.

Amkor's shift from back-end assembler up to the core of value is the most important structural change in advanced packaging over the past year; we laid out the full context in Optical Packaging in Transition (1): Why This $14B Market Suddenly Got Hot.

3. Technology and business highlights: CPO, advanced packaging, and two long-term agreements

For STT readers, this section is the main event.

CPO entered Amkor's official technology platform list for the first time. When discussing "technology leadership," CEO Kevin Engel named three categories of advanced packaging and test technologies the company is advancing: 2.5D, High-Density Fan-Out (HDFO), and "emerging technologies such as co-packaged optics", and said these engagements span AI infrastructure, high-performance computing (HPC), and networking applications.

When the world's largest OSAT formally writes CPO into its technology platform instead of a watch list, it means the CPO battlefield has extended from "optics" to "packaging," and packaging is the real threshold for volume production.

This is not a position we made up but an industry consensus: much of CPO's success or failure lies not in the optical components themselves but in heterogeneous integration and packaging yield. We have made this point repeatedly in the CPO Is Won in Packaging, Not Optics series and in The First Year of CPO Commercialization Begins: TSMC COUPE and the 200G EML Bottleneck; Amkor's statement adds endorsement from another heavyweight player.

Advanced packaging progress: 2.5D has 11 customers and multiple programs; HDFO has 5 customers and 10 active projects; each has 4 product launches this year. The most closely watched data center CPU program (on HDFO) began ramping in Q2 and will keep ramping in the second half; it was the main driver of this quarter's computing record and the basis for Q3 computing guidance of nearly 30% sequential growth. For the more advanced bridge technology, management put the timeline at 2028.

Two long-term agreements tying Amkor to the AI core:

• 10-year advanced packaging agreement with TSMC: building an integrated solution in Arizona "from advanced silicon manufacturing to advanced packaging and test," strengthening the US semiconductor supply chain.

• Multi-year strategic collaboration with NVIDIA: focused on advanced packaging and test supporting next-generation AI infrastructure, aligned with long-term technology roadmaps. Financially, it is a prepayment received around 2027 and returned quarterly as US services are delivered, with a contract term of 5 to 10 years.

NVIDIA personally tying packaging and test capacity into its AI roadmap follows the same playbook as its steady moves to consolidate optics and interconnect upstream; we covered this in NVIDIA Buys Into Optics: When Optical Communications Goes from Growth Stock to Index Heavyweight.

Capacity footprint: Arizona Phase I is "fully committed"; the new Songdo fab in Korea completes at year end, and the Gwangju expansion targets data center and advanced packaging after 2028; Vietnam is expanding SiP and NAND memory capacity, with SiP moving from Korea to Vietnam to free Korean space for rapidly ramping computing programs.

4. Management outlook: Q3 guidance and the profit path

Official guidance for next quarter (Q3 FY2026) (consistent with the earnings card):

• Revenue of $1.95B–$2.05B (midpoint about $2.0B, implying about +5% QoQ).

• Gross margin of 18.5%–19.5% (midpoint about 220 bps above Q2, mainly from a mix more weighted to advanced packaging).

• Operating expenses of about $140M.

• Net income of $180M–$205M, diluted EPS of $0.72–$0.82.

• Full-year 2026 CapEx maintained at $2.5B–$3.0B: 65%–70% for facility expansion (including Arizona Phase I) and 30%–35% for HDFO, test, and other advanced packaging capacity. Full-year effective tax rate about 20%.

Q3 end-market cadence: computing up nearly 30% QoQ (AI data center plus HDFO CPU ramp), auto/industrial up mid-single digits QoQ, consumer up mid-double digits QoQ, while communications down high-single digits QoQ, counter-seasonally.

Management's language should be read at two levels of certainty. Gross margin expansion was attributed to "product mix" (compute up, comms down), a mix shift that is already happening and highly credible. But when asked whether Q4 would step up again seasonally, the CEO said it "depends on mix and utilization" and "if sustained, we don't expect major changes"; that is conditional, not a commitment. More importantly, management proactively flagged that depreciation from US manufacturing in 2027/2028 will be a headwind to gross and operating margins, effectively pre-warning the market: the long-term target looks like "2028 only catches up to 2026" because Arizona capacity must absorb depreciation first, then climb in utilization.

5. Supply chain and customer clues

The call contained several signals worth noting for STT readers:

• Strong iOS, weak Android: Android was already weak in Q2, mainly due to memory supply and overall demand; iOS has seasonal pull-ins, but management cited GfK/Gartner expectations of declining smartphone shipments this year and admitted it is "feeling it."

• Moving SiP to Vietnam is "a one-time relocation plus a structural headache": the move leaves some capacity "idle" during crating, shipping, and requalification, pressuring utilization in the short term; and a headwind in one application area will persist into Q4 and even the first half of 2027, not ending in a single quarter.

• Tight memory supply and rising memory prices could in turn affect phone unit volumes, an external variable on the communications side.

• The data center CPU program is the largest single ramp driver, with customers at different stages (volume production, ramping, qualification), meaning visibility on this line is durable.

In other words, Amkor is reshuffling capacity by giving up communications in the short term to capture computing in the long term. This is an active mix upgrade, not a passive demand decline.

Summary

Amkor made one thing very clear this quarter: advanced packaging is now on the critical path of AI system performance, and Amkor wants to be the core supplier on that path, locked in by long-term agreements. CPO joining the platform for the first time, the TSMC and NVIDIA long-term agreements, and record computing revenue all point the same way.

Over the next three quarters, STT will watch these three indicators:

1. The ramp slope of HDFO/data center CPU: the key to whether computing can sustain nearly 30% sequential growth, and a thermometer for the advanced packaging mix upgrade.

2. When CPO-related engagements move from "emerging technology" to "revenue-contributing platform": still at the early co-development stage; any concrete customer or volume production timeline would be a re-rating point.

3. The race between Arizona depreciation and utilization: whether 2027/2028 gross margin is dragged down by depreciation at the new US fab depends on how fast utilization climbs; this is the biggest uncertainty in the long-term financial model.

The CPO ramp timeline has recently been nudged toward 2028 by the market, but that is not necessarily bad for a "pick-and-shovel" packaging house like Amkor: what gets extended is the shelf life of existing technology. We make this case more fully in CPO Ramp Pushed to 2028, but It's Not Bad News.