Topicsadvanced-packaging

How AI Compute Gets Built: Advanced Packaging

A sourced reference collection on advanced semiconductor packaging — CoWoS, chiplets, and the UCIe interconnect standard — the capacity constraint that limited AI chip production more than logic fabrication did.

Assertions
14
Sources consulted
15
Read in full
8/15
Cited as evidence
8

15 sources sit behind this page — including any that arrive with a concept this page shares with another collection. 8 were retrieved and read in full, and only those can back an assertion. 3 could not be retrieved, and 4 were surfaced and deliberately set aside. Every one of them is named in the register below, with the reason in view. How we source this.

Background

Our own synthesis, written to orient you — not evidence. Every factual statement here is asserted and sourced further down this page.

For most of the current AI build-out, the step that limited how many accelerators could be built was not lithography. It was packaging — the assembly of logic dies and memory stacks onto a shared substrate. TrendForce reported in June 2026 that the CoWoS supply-demand gap stood at around 20%, narrowing to about 10% by year end, with TSMC running 120,000 to 140,000 wafers a month and outsourced partners adding 50,000 to 60,000 more. Those figures come with a caution the report states itself: they are attributed to unnamed institutional investors via a newspaper, not to TSMC.

That a packaging number is the headline at all is the change worth understanding. TSMC puts it plainly on its own site: packaging technologies “were once considered simple backend processes, almost an inconvenience”, and are now “critical to a product’s performance, function, and cost”. Its 3DFabric family — SoIC, CoWoS and InFO — deliberately spans both wafer-level frontend and backend work, because the boundary between making a chip and packaging one has stopped being a boundary. The cheapest-subcontractor step became the most advanced foundry’s competitive advantage.

The next structural change is the shape of what the work is done on. A round wafer wastes area at its edges when the products cut from it are large rectangles, and accelerator packages are now large rectangles. Chip-on-Panel-on-Substrate moves onto a square panel: TSMC has standardised on 310 by 310 millimetres, with pilot production targeted for 2027, mass production in the second half of 2028, and glass core substrates after 2030. Taiwan panel makers already run fan-out panel-level packaging up to 620 by 750 millimetres for mature products, so the base exists; what is new is bringing leading-edge accelerator work onto it. Every one of those dates is a forecast.

Underneath the capacity story sits the design idea that created the demand. A chiplet is a die built to be combined with others rather than to work alone, and UCIe is the industry standard for connecting them across vendors — released in 2022 and revised through 2025. The reason this matters more each year is covered on the GPU page: once a design reaches the reticle limit, continuing means building from several dies, and the quality of the connection between them becomes as important as the arithmetic inside them.

What is missing is named rather than implied. TSMC’s SoIC page returned an error, so this collection cannot state bond pitches or interconnect densities — the numbers that would show concretely how much finer direct bonding is than microbumps, and the most useful thing a later pass could add. Intel’s advanced packaging page and a Semiconductor Engineering article both refused the fetcher, so nothing is asserted here about EMIB, Foveros or hybrid bonding. And several widely-circulated articles credit the panel-level move with lifting area utilisation past 90% and cutting cost 20 to 30%; those figures do not appear in the research this collection read, and are not repeated.

Figures

Every number below is asserted and sourced elsewhere on this page.

Packaging capacity, and the gap it is closing

Reported monthly CoWoS wafer capacity in 2026. The industry total is the sum of the foundry and its outsourced partners.

CoWoS wafers per month, thousands

~130k

TSMC (120-140k)

~55k

OSAT partners (50-60k)

~185k

Industry total

TrendForce, 15 June 2026. The report attributes these figures to unnamed institutional investors via Economic Daily News rather than to TSMC. Ranges are shown at their midpoints; all are estimates for a year in progress, not measured output.

Concepts

The vocabulary this subject is built from, and what we can show about each.

Advanced Packaging

packaging technology

TSMC has publicly stated a CoWoS capacity CAGR above 80% for 2022-2027; the widely-cited 120,000-140,000 wafers-per-month figure for 2026 is an analyst estimate attributed to unnamed institutional investors, not a TSMC disclosure.

ReportedSupported by the sources below, not yet editor-reviewed.
1 source1 retrieved & read

Advanced packaging and HBM — not logic dies — were the binding constraint on AI chip production in 2025, with the four largest AI chip designers consuming ~90% of global CoWoS and HBM supply but only ~12% of advanced logic die production.

ReportedSupported by the sources below, not yet editor-reviewed.
1 source1 retrieved & read

Advanced packaging integrates multiple separately-manufactured dies into one package with dense short interconnects; TSMC's CoWoS platform, which places an SoC and multiple HBM stacks on a shared interposer at over 2x reticle size, is the dominant implementation for AI accelerators.

ReportedSupported by the sources below, not yet editor-reviewed.
1 source1 retrieved & read

Chiplet

component

A chiplet is a small IC containing a defined subset of system functionality, combined with others in one package instead of a monolithic die — enabling IP reuse, per-function process-node choice, and known-good-die testing before assembly.

ReportedSupported by the sources below, not yet editor-reviewed.
1 source1 retrieved & read
  • SupportsQualifiesRetrieved & read
    Chiplet

    WikipediaOther

CoWoS (Chip-on-Wafer-on-Substrate)

packaging technologyshared from another collection — see its own page for what it asserts

Panel-Level Packaging

packaging technology

TrendForce reported in June 2026 that TSMC has standardised CoPoS on a 310 x 310 mm panel, with pilot production targeted for 2027, mass production in the second half of 2028 and glass core substrates after 2030 — moving leading-edge packaging off the round wafer onto a square panel.

ReportedSupported by the sources below, not yet editor-reviewed.
1 source1 retrieved & read

Silicon Interposer

packaging technologyshared from another collection — see its own page for what it asserts

The Packaging Capacity Constraint

process

TrendForce reported in June 2026 a CoWoS supply-demand gap of around 20% narrowing to about 10% by year end, with TSMC at 120,000-140,000 wafers a month and OSAT partners adding 50,000-60,000 — figures the report attributes to institutional investors via Economic Daily News, not to TSMC.

ReportedSupported by the sources below, not yet editor-reviewed.
1 source1 retrieved & read

TSMC describes packaging as having been 'once considered simple backend processes, almost an inconvenience' and now 'critical to a product's performance, function, and cost', with its 3DFabric family spanning both wafer-level frontend and backend technologies.

ReportedSupported by the sources below, not yet editor-reviewed.
1 source1 retrieved & read
  • SupportsPrimary evidenceRetrieved & read
    TSMC 3DFabric

    TSMCCompany

Through-Silicon Via (TSV)

interconnectshared from another collection — see its own page for what it asserts

UCIe (Universal Chiplet Interconnect Express)

interconnect

UCIe is an open die-to-die chiplet interconnect specification co-developed by AMD, Arm, ASE, Google Cloud, Intel, Meta, Microsoft, Qualcomm, Samsung and TSMC, letting packages exceed single-die reticle limits and mix chiplets from different vendors; version 1.0 states up to 1.35 TB/s per mm² at 45µm bump pitch and about 0.5 pJ per bit.

ReportedSupported by the sources below, not yet editor-reviewed.
2 sources2 retrieved & read

Timeline

What actually happened, in order, with sources.

Coverage
  • 2 Taiwan
  • 4 Unattributed

Where this topic’s events took place, as far as our sources establish it. Events with no single location — a standards publication, say — and events we have not yet attributed are both counted as unattributed rather than omitted.

  1. Jun 17, 2026

    Packaging Moves Off the Round Wafer

    manufacturing developmentTaiwan

    TrendForce reported on 17 June 2026 that TSMC has standardised CoPoS on a 310 x 310 mm panel format, targeting pilot production in 2027, mass production in the second half of 2028 and glass core substrates after 2030 — all forecasts.

    ReportedSupported by the sources below, not yet editor-reviewed.
    1 source1 retrieved & read
  2. Jun 15, 2026

    The Packaging Bottleneck Starts to Ease

    capacity announcementTaiwan

    TrendForce reported on 15 June 2026 that the CoWoS supply-demand gap was around 20%, narrowing to about 10% by year end, with TSMC at 120,000-140,000 wafers a month plus 50,000-60,000 from OSAT partners — attributed to institutional investors via Economic Daily News rather than to TSMC.

    ReportedSupported by the sources below, not yet editor-reviewed.
    1 source1 retrieved & read
  3. Aug 5, 2025

    UCIe 3.0 Specification Released

    standards milestone

    The UCIe 3.0 specification was released on August 5, 2025 and is the current version.

    ReportedSupported by the sources below, not yet editor-reviewed.
    2 sources2 retrieved & read
  4. Aug 6, 2024

    UCIe 2.0 Adds 3D Packaging Support

    standards milestone

    The UCIe 2.0 specification, released August 6, 2024, added 3D packaging support to increase bandwidth density and power efficiency beyond 2.5D side-by-side integration.

    ReportedSupported by the sources below, not yet editor-reviewed.
    1 source1 retrieved & read
  5. Aug 8, 2023

    UCIe 1.1 Specification Released

    standards milestone

    The UCIe 1.1 specification was released on August 8, 2023.

    ReportedSupported by the sources below, not yet editor-reviewed.
    1 source1 retrieved & read
  6. Mar 2, 2022

    UCIe 1.0 Specification Released

    standards milestone

    The UCIe 1.0 specification was released on March 2, 2022, establishing an open die-to-die chiplet interconnect standard rated at up to 1.35 TB/s per mm² at 45µm bump pitch.

    ReportedSupported by the sources below, not yet editor-reviewed.
    1 source1 retrieved & read

Source register

All 15 sources behind this page — what we read, what we tried to read and could not, and what we looked at and set aside, with the reason in view for each. A concept shared with another collection brings its own references with it, so some entries here were surfaced for a neighbouring topic rather than this one.

Cited as evidence
8
Tried, could not read
3
Surfaced, set aside
4
Cited sources 8 distinct links

Original publisher links. Files open on the publisher’s site; we do not host copies. A linked document is not an additional source or an independent verification.

Tried, could not read3

We attempted these and were refused or served nothing. Nothing on this page rests on them; they are published so the gaps are checkable rather than invisible.

Surfaced, set aside4

These came up while researching and were deliberately not used. We do not claim to have read them — each is listed with why it was passed over, so the shape of the survey is visible and not just its conclusions.

Coverage & limits

What this page does and does not claim.

Second packet through the pipeline, rebuilt in September 2026 to the reference-collection template and reweighted toward current developments: the capacity constraint and the move to panel-level packaging now lead, with the UCIe standards history retained as background rather than as the spine. Fourteen sources were consulted: nine were retrieved and read, three were attempted and could not be read, and four were surfaced and set aside with a stated reason. Two of the reads are new and recent — TSMC's own 3DFabric page, obtained in a browser after an HTTP 403, and TrendForce's June 2026 report on CoPoS. One set-aside entry is unusual and deliberate: a widely-circulated article attributes to that TrendForce research a 30% cost reduction and 90%-plus utilisation figure which does not appear in it, and it is listed so a reader meeting those numbers elsewhere can see that this collection looked and could not substantiate them. Capacity figures throughout carry the attribution chain their source states — institutional investors via Economic Daily News, not TSMC. Named gaps: TSMC's SoIC page returns HTTP 404, so no bond pitch or interconnect density is asserted anywhere here; Intel's advanced-packaging page and a Semiconductor Engineering article both returned HTTP 403, so this collection still asserts nothing about EMIB, Foveros or hybrid bonding. The CoWoS, interposer and through-silicon-via concepts are shared with the HBM collection rather than duplicated. Not yet editor-reviewed; every assertion reads as reported.

Source check, 2026-09-17. Numeric-presence checks passed for 14 assertions using available source text, which may be cached. This is not verification of their meaning. What this check does and does not prove →

  • Not editor-reviewed unless labelled. Assertions marked Reported are assembled from the sources shown and have not yet been checked by an editor. Only Primary source and Corroborated mean a human verified them.
  • Disagreements are preserved, not resolved. Where sources conflict, both accounts appear and the assertion is marked Disputed.
  • Retrieval status is disclosed per source. A source we could not open is never counted as evidence for an assertion.

This page is also available as structured data: /api/v1/topics/advanced-packaging