Global AI Value Chain — Equity Deep Research

Global AI Value Chain — Listed-Company Deep Research (2026-08)

A full teardown of the global AI value chain, from upstream equipment to downstream applications. Every figure is anchored to the 13 August 2026 close and a single FX set, so comparisons work across markets.

Coverage

Dimension	Count	Detail
Markets	9	US, mainland China, Hong Kong, Taiwan, Japan, Korea, Netherlands, Germany, France
Listed companies	105	combined market cap ≈ $44.2 trillion
Key private companies	30	including OpenAI, Anthropic and other nodes in the deal network
Core pass-throughs	10	fixed five-section template each, plus 12 ownership and supply-chain diagrams
Length	155 pages	13 chapters; 80 tables, 29 charts
Reference date	2026-08-13	closing prices and FX anchored to the same moment

The longest single-point dependency chain

The report's "three-layer, six-segment hourglass" framework identifies three choke points: TSMC CoWoS packaging, the HBM trio, and ASML EUV lithography. The longest single-point dependency chain runs four layers deep, with no second supplier at any layer:

Layer	Company	Function	Second supplier
1	TRUMPF / Zeiss	EUV light source and optics	none
2	ASML	EUV lithography systems	none
3	TSMC	leading-edge nodes and CoWoS packaging	none
4	NVIDIA	AI accelerators	none

Source: chapter 4, chain map and value distribution. A capacity or policy change at any layer propagates to everything downstream.

Hard numbers inside (a sample)

Metric	Value	Basis and meaning
2026 capex guidance, five US hyperscalers	≈ $802.5bn	sum of guidance midpoints; +86% versus 2025 actual, the source variable for the whole chain
NVIDIA FY2027 Q1 revenue	$81.6bn (+85%)	China share down from 21.9% to 5.6%
Micron FQ3 FY2026 revenue	+346% YoY	GAAP gross margin 84.6%, at a forward P/E of about 6.6x
NVIDIA share of TSMC CoWoS capacity	≈ 60%	set to exceed 20% of TSMC revenue in 2026, passing Apple for the first time
Gross margin at the two ends of one chain	NVIDIA ≈ 75% vs FII 7.15%	the spread in value capture along a single chain
Oracle remaining performance obligations	$638bn (+363%)	about $300bn from OpenAI alone; full-year free cash flow −$23.7bn

Source: company SEC filings and results releases. Every figure carries an evidence grade and a source inside the report.

Ten core-company pass-throughs

Segment	Companies
Compute silicon and accelerators	NVIDIA, Broadcom, Cambricon
Wafer fabrication and advanced packaging	TSMC, SMIC
Semiconductor equipment	ASML
Memory and HBM	SK hynix
Cloud platforms and foundation models	Microsoft
Optical modules and server assembly	Innolight, Foxconn Industrial Internet

Each follows the same five-section template: operating data → top ten holders → board and management → two-way supply-chain pass-through → valuation and key variables.

A dedicated chapter on circular deals

One chapter is set aside for the hardest part of the AI chain to characterise: OpenAI's four-way counterparty relationships, Anthropic being invested in and supplied by four giants at once, and NVIDIA's outbound investment network split transaction by transaction into closed and announced. Bull and bear arguments are quantified side by side, without a position.

Cycle ranking and scenarios

Eleven segments are ranked by cycle position, from memory and HBM (1st) to AI software applications (11th), each with three scenarios (50% / 20% / 30%) and observable trigger signals.

What makes this one different

Practice	What it means
Four-level evidence grading	every key figure carries [A]/[B]/[C]/[D]: A = primary filings, B = major media, C = analyst estimates, D = unverified. Conflicting figures are shown side by side, never silently resolved
Anti-misreading manual	18 widely circulated factual errors corrected; 23 open questions listed with both readings
Strict basis management	fiscal-year mismatches flagged; three capex definitions separated; A/H market caps marked non-additive; P/Es distorted by one-off gains labelled as distorted

Contents (13 chapters)

Chapters	Subject
1–2	Summary and core conclusions · Methodology and data verification
3–4	Industry overview · Chain map and value distribution
5–7	Upstream: compute hardware · Midstream: cloud and foundation models · Downstream: applications and software
8–9	Core-company deep dives · Cross-holdings and investment network
10–12	Financial and valuation comparison · Risk factors · Investment views
13	Appendices: verification worksheets and source list

How it is produced & disclaimer

Compiled independently by the TopX research team from public information, then reviewed and revised before release by professional investment managers and fund managers with asset-management experience. Reviewers take part in a personal capacity, do not represent their respective employers, and provide no investment advisory services through TopXEA; TopXEA holds no investment-adviser licence in any jurisdiction. Primary filings (SEC, exchanges, company disclosures) are the first-priority source, key figures require corroboration from at least two independent sources, claims carry [A]/[B]/[C]/[D] evidence grades, and known limitations are disclosed item by item inside the report. This is research content — not investment advice — and is not tailored to any investor; historical data and forward projections do not guarantee future results. Digital goods are non-refundable once delivered.