All comparisons

Compare the evidence

Micron vs TSMC vs Amkor: HBM and packaging capacity

Compare HBM memory, foundry integration and outsourced packaging. Announced investment, qualification, high-volume production and accepted customer deliveries remain separate evidence stages.

Read horizontally within each row. Product ratings, planned capacity, completed deliveries and current financials describe different stages.

On narrow screens, scroll the comparison horizontally. Each figure keeps its period, accounting definition and source. GAAP cash flow, adjusted margins, segment gross profit and EBITDA are not interchangeable.

Micron vs TSMC vs Amkor: HBM and packaging capacity: capacity, delivery, economics and evidence boundaries
Research dimensionMicron TechnologyMU · Checked 2026-10-28TSMCTSM · Checked 2026-10-28Amkor TechnologyAMKR · Checked 2026-11-02
Industrial role

Memory including high-bandwidth memory

Semiconductor foundry and process technology

Semiconductor packaging technology

Constraint to test

Qualified HBM output matched to accelerator requirements

Qualified fabrication matched to process-node demand

Can packaging technology convert into qualified production output?

Reported evidence and sources

Micron announced high-volume production of HBM4 designed for NVIDIA Vera Rubin.

TSMC documents dedicated foundry technologies; AMD announced Venice production on TSMC 2nm.

Amkor describes semiconductor packaging technologies including wafer-level packaging.

Capacity and delivery

Quantitative accepted output and qualification status are not established in this profile.

Quantitative accepted output and qualification status are not established in this profile.

  • NVIDIA and Amkor target more than 300% growth in South Korean interposer-wafer capacity by end-2028. The release does not report current qualified output, yield or accepted shipments. Source ↗
  • Q3 Computing revenue reached a record and full-year capex guidance was about $2.5bn. These are financial and investment signals; the release does not quantify qualified interposer output, package yield, customer allocation or commissioned AI capacity. Source ↗
Reported economics

Not quantified in this profile.

Revenue
NT$1,494.24bn / US$46.72bnQ3 2026 · TIFRS consolidated; issuer USD translation · Source ↗
Gross margin
66.4%Q3 2026 · TIFRS · Source ↗
Net sales
$1.991bnQ3 2026 · GAAP consolidated revenue · Source ↗
Gross margin / operating margin
18.3% / 11.3%Q3 2026 · GAAP consolidated measures · Source ↗
FY2026 capital expenditures
About $2.5bnFY2026 outlook · Management outlook; not commissioned output · Source ↗
Evidence coverage

Public product or operations profile. Quantitative economic exposure has not been reviewed here.

Revenue attributable to this specific constraint, pricing power and current valuation have not been quantified in this profile. A separate financial review is needed.

No quantified share of revenue attributable to the specific constraint. No evidence-strength score is assigned.

Primary-source operating or segment results; capacity evidence retains its stated stage.

Consolidated results provide financial context. Revenue attributable to this specific physical constraint is not separately established.

No quantified share of revenue attributable to the specific constraint. No evidence-strength score is assigned.

Primary-source operating or segment results; capacity evidence retains its stated stage.

Q3 consolidated revenue and margin include multiple end markets and package types. Record Computing revenue does not disclose AI-specific advanced-packaging revenue, yield or accepted unit output.

No quantified share of revenue attributable to the specific constraint. No evidence-strength score is assigned.
Counter-evidence (dated; earlier periods are historical)

No separately sourced counter-evidence review is included in this baseline. The conditions below are research tests.

No separately sourced counter-evidence review is included in this baseline. The conditions below are research tests.

  • Consolidated net sales were roughly flat year over year and Communications was weak. Capex supports multiple end markets, so neither spending nor Computing revenue alone proves AI-package scarcity or accepted output. Source ↗
Next to monitor
  • HBM generation-specific shipments and yields
  • Customer qualification and allocation
  • Capital spending versus cash generation
  • Node-specific capacity and yields
  • Advanced packaging availability
  • Customer qualification and geographic ramp-up
  • Package-specific qualification, yield and accepted shipments
  • Computing revenue conversion versus AI-specific advanced-packaging output
  • Capex deployment, gross margin and cash as new capacity ramps
What would weaken the case?
  • Qualified supply catches up with demand
  • Pricing or inventory indicates easing constraints
  • Additional qualified output eases allocation
  • Customer demand shifts to available alternatives
  • Qualified competing packaging capacity expands
  • Device demand or upstream die supply limits utilization
  • Capital intensity rises without package-specific output or margin conversion

Open in Excel or another spreadsheet. Each entry preserves its source, date and accounting definition; research interpretations are labeled.

How to use this comparison

Start with the constraint, inspect the dated sources, then follow the company to monitor new evidence. A documented industrial role does not establish a shortage or an attractive share price.

This comparison uses open company research checked on the dates shown. Valuation and expected returns require a separate review.