Single-day -24.74%, which sentence caused AMKR to crash?

NEXT PICK · Market Insights

Single-day -24.74%, which sentence caused AMKR to crash?

Tuesday, July 28, 2026

A company that had just delivered quarterly revenue growth of 26% year-on-year and earnings per share of $0.70—nearly half of the market expectation of $0.47—was hit to $45.06 at the second day's opening and closed at $45.69, a single-day drop of -24.74%, with trading volume expanding to 3.6 times the three-month daily average.

The previous day's closing price was $60.71, and five days ago, it had surged 17% in a single day after securing a $1.5 billion multi-year package order from NVIDIA.

If even "exceeding expectations" can't bring a price increase, then why is the market pricing in? The answer is hidden in the earnings call that most headlines ignore—not revenue, not gross margin, but the communications business and a capital expenditure plan of $2.5 to $3 billion.

Latest price $45.69 ▼ -24.74%

Q2 Revenue 1.9 billion USD ▲ +26% year-on-year

2026 Capital expenditure $2.5–3 billion ▲ Approximately 2.5 times operating cash flow

After chip manufacturing, there is another process that no one can avoid

Founded in 1968 and headquartered in Tempe, Arizona, Amkor is the world's second largest independent outsourced semiconductor packaging and test manufacturer, commonly known in the industry as OSAT. ASE Technology holds about 28-30% of the share, ranking first, followed closely by Amkor with about 15-18%.

What it does is the entire process from wafer to finished product that can be installed in servers and smartphones: wafer bumps, wafer probe testing, wafer thinning, package design, flip-chip assembly, fan-out packaging, system-in-package packaging, all the way to final testing and programming. Customers do not need to coordinate processes among multiple suppliers.

To put it more bluntly

TSMC is responsible for weaving the fabric, Amkor handles cutting, sewing, labeling, quality inspection, and then delivering the finished product to the brand. For decades, this process was treated as low-margin odd jobs.

But AI turned it into a bottleneck. 2.5D/3D heterogeneous integration stacks multiple chips and HBM onto the same substrate, upgrading advanced packaging from "back-end miscellaneous" to physical constraints for computing power expansion.

Amkor's client list includes NVIDIA, Qualcomm, and MediaTek. Full-year revenue for 2025 is $6.71 billion, up 6.2% from $6.32 billion in 2024; After entering 2026, the pace accelerated sharply, with Q1 1.68 billion up 27.3% year-on-year, and Q2 1.9 billion up 26% year-on-year.

What truly changed its identity were the last two contracts

A ten-year advanced packaging agreement with TSMC, along with NVIDIA's $1.5 billion multi-year upfront payment—to expand its $2 billion packaging and testing plant in Peoria, Arizona, which will begin construction in the second half of 2025.

So here's the question

Why is a company that has been doing "odd jobs" for 58 years suddenly being bet on by two top decision-makers in the AI industry chain at the same time?

A cheap 20x P/E ratio versus expensive $3 billion in expenses

Let's first look at static numbers. A stock price of $45.69, corresponding to a market cap of $11.33 billion, rolling earnings per share of $2.23, and a P/E ratio of about 20.5 times; Its market capitalization is only 1.7 times the $6.71 billion revenue for the full year of 2025.

The median price-to-sales ratio of the semiconductor packaging sector is usually in the 3-5 times range, and Amkor is significantly discounted by relative metrics. The stock price is about 14.5% below the 200-day moving average of $53.41, and is 37% below the 50-day moving average of $72.91. The RSI (14) is only 27.5, indicating that technicals are already deeply oversold.

From this perspective, the negative side seems to have been fully priced in by a 20% drop in one day. But the cash flow section makes people pause.

**Here, a widely circulated claim must be corrected

Amkor's operating cash flow in 2025 is $1.1 billion, but free cash flow is only $308 million. ** The 800 million yuan between the two was entirely consumed by capital expenditure.

Management's full-year 2026 capital expenditure guidance is $2.5 to $3 billion, with 65-70% allocated to plant expansion, mainly Arizona Phase I, and the remaining 30-35% allocated to high-density fan-out, testing, and advanced packaging equipment.

A company with about $1.1 billion in annual cash flow plans to spend $2.5 to $3 billion—the so-called 'self-sustaining, no external financing' judgment is untenable in 2026. With 2.5 billion in cash and short-term investments on hand, total liquidity of 3.6 billion, debt also at 2.5 billion, debt-to-EBITDA ratio of 1.8 times, this balance sheet can hold up, but it's tight.

What about NVIDIA's $1.5 billion prepayment? Management made it clear during the call: the money is expected to arrive in 2027 and will be returned to customers as the service is gradually delivered. It is a cash flow time smoother, not a free profit.

So, is a 20.5x P/E ratio cheap or a trap? In a window where profits are just taking off but cash will be swallowed by capacity for two or three years, static valuation itself should not be the anchor for decision-making.

The narrative has been confirmed, but production capacity has yet to be realized

Amkor is currently on the 'eve of a breakout' rather than a 'breakout.' The validation of the business model has been completed: NVIDIA's upfront payments, TSMC's ten-year agreement, and the rigid demand for advanced packaging in AI data centers—these are no longer stories.

But production capacity is still a construction site. The Arizona plant has not yet started production, and Q3 revenue guidance is $1.95 billion to $2.05 billion, below the consensus forecast of $2.12 billion, indicating that the pace of capacity release is not keeping pace with the expansion of demand narratives.

The market itself is voting with price

The stock price has fallen below the 200-day moving average and is still 37% away from the 50-day moving average, indicating the moving average system has shifted from bullish alignment to bearish pressure. The +116.4% return over the past twelve months was erased by the -46.5% plunge over the past thirty days.

What's even more challenging is management's own warning—during Arizona's ramp-up period, from 2027 to 2028, it will face headwinds in gross margin due to underutilization. This means there is still a road between "production" and "making money," which looks unsightly on paper.

TSMC's friends, or TSMC's next lunch?

Amkor faces four layers of opponents. At the top is TSMC, whose CoWoS and InFO advanced packaging have become the preferred solutions for NVIDIA and AMD, enjoying a valuation premium of 8-10 times, while Amkor only offers 1.7 times.

Right now, the two seem more like a coordinated force than a fierce battle

TSMC's Arizona production line produces front-end wafers, while Amkor's Arizona factory takes over back-end packaging and testing. The two sides even signed a ten-year agreement. But as long as TSMC continues to expand its packaging capacity, there is a possibility that in the medium to long term, it could backfire upstream at Amkor's high-end market share.

The second layer is ASE Technology Holdings, the world's largest OSAT, which faces Amkor head-to-head in flip-chip integration, system-in-package, and fan-out packaging, with scale advantages providing stronger cost dilation capabilities; However, its North American production capacity layout is weaker than Amkor's, and it has suffered geographical disadvantages from the "Made in America" policy dividend.

The third layer is Intel foundry. Foveros' 3D packaging and EMIB technology routes do pose threats, but its own ongoing losses, yield, and customer trust remain to be verified. In the short term, insufficient capacity may cause orders to spill over to Amkor.

The fourth layer consists of Chinese mainland market leaders like JCET and Tongfu Microelectronics, which have price advantages in traditional QFN and BGA cost-sensitive packaging, but lack sufficient technological accumulation in high-end segments like 2.5D/3D and HBM stacking, and export controls restrict their access to advanced equipment.

Where exactly is Amkor's moat built? It's not a patent wall, it's switching costs plus geographic positioning. Once a chip is introduced into an OSAT production line, packaging design documents, testing procedures, yield tuning, and automotive-grade reliability certification are deeply coupled, requiring a 6 to 12 month validation cycle to change suppliers.

NVIDIA's prepaid system has made this binding systematic—pay first, then get the goods, which is the toughest form of lock-up. Combined with the CHIPS Act and the policy barrier of "Made in America," there is no second OSAT in the U.S. with equivalent advanced packaging capacity in the short term.

This moat isn't narrow, but it's not deep either. It can block new entrants, but it cannot stop TSMC from moving down the value chain.

The ceiling is high, but the market share ceiling is even lower

The global semiconductor packaging and testing market is expected to reach approximately $40 billion to $45 billion by 2025, with the portion outsourced to OSAT at approximately $35 to $38 billion. The real growth is in advanced packaging—2.5D/3D, fan-out, system-in-package, and HBM stacking—which is expected to reach about $12 to $15 billion by 2025, with industry estimates reaching $25 to $30 billion by 2028, with a compound annual growth rate of around 20%.

Amkor's focus on high-end packaging and system-level testing corresponds to a serviceable market of about $18 to $22 billion, with advanced packaging related to AI and high-performance computing accounting for about $8 to $10 billion, making it the fastest-growing subset.

Its $6.71 billion revenue in 2025 corresponds to about 17-19% of the global OSAT market, but in the most lucrative segment of advanced packaging, excluding TSMC's own production capacity and competition from ASE, it can actually secure a market share of about $3 to $4 billion.

How will you proceed in the next three years? In a pessimistic scenario, AI packaging demand growth is expected to drop to 15%, and the company's revenue is expected to grow at a compound annual rate of 8-10%, reaching about $8.5 billion by 2028; Under the baseline scenario, demand remains above 20%, Arizona is expected to start production in 2027, with revenue projected to compound 12-15% annually, and to reach $9.5 to $10 billion by 2028; Under optimistic conditions, the benefits of localized policies are fully unleashed, and by 2028, the market is expected to surpass 11 billion yuan.

The only turning point for the three-tier path is when that factory will reach full production.

Reevaluate packaging from OEM to strategic assets

The logic of the bulls is not based on "how much was earned this quarter," but on the fact that the nature of the business is changing:

▲ Bull Case

Cash needs to be eaten up by factories for two years first

The bears' rebuttals are also backed by hard data, and most come from the company's own conference calls:

▼ Bear Case

Three time windows determine whether they are turning points or traps

Over the next six to eighteen months, the pace of stock prices will be almost entirely dominated by the chain of evidence of capacity realization:

•	Q3 2026 Financial Report: Validates whether the 18.5%-19.5% gross margin guidance can be implemented, and whether the decline in the telecommunications business is seasonal or structural
•	Arizona Peoria plant construction progress and production timeline announced: timeline uncertain, but the direction is positive
•	The capacity corresponding to NVIDIA's prepayments is substantially released and converted into quarterly revenue increments: the prepayments themselves are expected to arrive in 2027
•	NVIDIA GB300 fully localized and CoWoS packaging implementation: validating Arizona production line yield, affecting neutral to positive outcomes
•	Gross margin performance during the 2027 ramp-up period: Management has already warned of headwinds, which is the most easily misjudged by the market

It took only one month to go from tripling to halving

Looking at the long timeline, Amkor's return over the past twelve months is +116.4%, with the 52-week range rising from $20.87 all the way to $96.68—the market reassessed the proposition that "advanced packaging is the AI bottleneck" in just one year.

The driving force behind the rise is quarterly expansion of profit margins

Operating profit margin will rise from 2.39% in Q1 2025 to 7.98% in Q3, and reach 10.5% in Q2 2026, with net profit up 218% year-on-year. On July 23, NVIDIA announced its prepayment status, causing its stock price to surge about 17% in a single day, with the narrative shifting from "cyclical recovery" to "AI strategic assets."

Then came the -46.5% drop for the month backward. The post-market earnings report released on July 27 showed revenue, gross margin, and earnings per share all exceeded expectations, except for the Q3 revenue guidance that missed by about $70 million; The next day, it gapped up to $51.80 at the open, plunged to a low of $45.06, and closed down 24.74%. Some brokerages adjusted their target prices on the same day and maintained a neutral rating.

Looking back over the year, the right and wrong are actually clear: the bulls' judgment that "packaging is the bottleneck, not a commodity" was fully confirmed by NVIDIA's actions; The bears' concerns about "low absolute profit margins and a capital expenditure black hole" are also fully valid. Neither side is wrong; the fault lies with those who put two things into the same quarter and trade them.

⚠️ Risk Notice • Capital expenditures of $2.5-3 billion in 2026 far exceed operating cash flow, turning free cash flow significantly negative • Management has warned that insufficient capacity utilization in Arizona for 2027-2028 will weigh on gross margins • The telecommunications business declined quarter-on-quarter due to production lines moving to Vietnam and storage shortages, which may not be a single-quarter phenomenon • With highly concentrated customers, any adjustment in outsourcing by either NVIDIA or TSMC would have a shock • AI supply chain costs have risen across the board, with upstream price hikes and downstream price pressures squeezing profit margins in both directions • The stock price fell 46.5% in a month, with extremely high volatility, making position management more difficult than fundamental judgment

🟡 Neutral The profit inflection point is real, but the cash needs to be eaten up by the factory for two years.

💬 Discussion It's not a bottom-fishing price, it's a window to watch. Its real-time buy and sell positions can be viewed via AMKR on the official account.

Source

NextPick real-time market snapshots, Amkor official earnings and investor relations announcements, SEC 8-K filings, Q2 2026 earnings call minutes, and public coverage from mainstream financial media.