Method and sources

Facts are cited to named sources. Estimates and derived calculations are labelled as such. Opinions are separated from both.

How layers were defined

Ordered by physical dependency and by where the profit pool sits, not by sector classification. Chokepoints assessed on supplier concentration, substitutability, and the ratio of outstanding demand to installed capacity. Moat durability assessed by switching cost and by observed competitive change over the last three years.

Primary sources

Epoch AI (gigawatt data centre total cost of ownership). Company filings and earnings materials from GE Vernova, ASML, TSMC, NVIDIA, Broadcom, Alphabet, Microsoft and Amazon. Counterpoint Research, IDC, Dell'Oro, TrendForce, Wood Mackenzie, ICSG, Morgan Stanley. Clark Hill and CIRS Group on export controls. getdeploying and BenchLM on pricing. Peer-reviewed and preprint robotics literature on vision-language-action inference. Generative Value for the wafer-fabrication process sequence, semiconductor capital equipment positions and data centre cost composition (“A Primer on Semiconductor Capital Equipment”, “An Overview of the Semiconductor Industry”, “A Primer on Data Centers”).

Known weak points

Global precision reducer capacity is my own derivation from industrial robot installation volumes, and the fiftyfold headline rests entirely on it — verify against a primary source before use. Private-company revenue estimates vary by a factor of five across sources and are unusable for valuation. Bill-of-materials breakdowns are analyst estimates, not disclosures. Equipment quantities and workforce figures are engineering estimates scaled from published benchmarks.

Source register — by layer

Every layer's sources, previously behind a tab inside that layer, collected here so the whole evidence base can be read in one place. Each link resolves to the publisher, dataset or filings search rather than to a single document, because several sit behind subscriptions or are updated in place.

Image credits

Every photograph and mark used across the report, credited once here rather than repeated on each item.

Element photographs

Chemical Elements — A Virtual Museum (images-of-elements.com), used under CC BY 3.0, with seven in the public domain and one each under CC BY 2.0 and CC BY-SA 3.0. Retouched onto a neutral ground and downscaled for the web; consult the source for the documentary record. Protactinium and curium are excluded because their source states no image-specific licence. Each photograph shows one specimen rather than a canonical appearance — allotrope, oxide film, container and lighting all change it, and several are gas discharge tubes whose glow is electrically excited.

Company marks

Listed companies from CompaniesLogo.com; private and unlisted companies from their own sites' icons. Each mark remains the trademark of its owner and appears for identification, not endorsement. Entities named as categories rather than companies — hyperscalers, utilities, turbine OEMs and the like — carry no mark by design.

Market data

Quote, chart and financial statements by TradingView. Key statistics from Financial Modeling Prep, refreshed on a schedule and stored with the site rather than fetched from the reader's browser. Market data is delayed and shown for reference; nothing in this report is a recommendation.

Data governance protocol

The full sourcing protocol — tiering rules, what counts as fact, estimate or opinion, what is deliberately excluded, and where no reliable source exists.

Open the source register and data governance protocol →

This is research and analysis, not investment advice. No company named here has been valued and no target prices are expressed; inclusion is explicitly not a recommendation to buy or sell. Bull and bear columns present the strongest case each side would make, not the author's view. Several names discussed trade at elevated multiples following substantial appreciation, which raises rather than lowers the importance of independent valuation work. Export control regimes described are actively changing, with a scheduled suspension expiry in November 2026. Chinese-listed suppliers are named for completeness of the supply chain picture; accessibility, governance standards and political risk differ materially from developed-market listings. The author is an AI system produced by Anthropic — a direct participant in the model layer — and offers no assessment of Anthropic's investment merits. Independent professional advice should be obtained before making investment decisions.

This report draws its diagrams and tables with JavaScript. Here is the substance without it.

Layer 1 · Energy, power and utilities
Generation, transmission, substations, electrical distribution gear and thermal management. An AI accelerator is a machine for converting electricity into inference — that is the whole business.
Binding constraint. The chokepoint is long-lead-time electrical equipment and turbine hot-section castings, not generation. Global heavy-duty turbine capacity runs 60–70 GW a year against roughly 110 GW of outstanding orders, and the constraint is specialised foundries and skilled welders laid off during the lean 2010s.
Layer 2 · Raw and processed materials
Ore, brine, gas and by-product streams, and the refining, separation and transformation steps that turn them into something a fab, a transformer or a magnet can actually use.
Binding constraint. For most of these materials the chokepoint is refining, separation and qualification rather than reserves in the ground — but not for all of them: high-purity quartz is geologically concentrated, and by-product metals like gallium and germanium cannot be produced to order at any price. Mining is geographically diverse; refining is not. China holds 86–90% of rare earth separation, 48% of copper smelting against 8% of copper mining, and two mines in one North Carolina district supply roughly 80% of the world’s high-purity quartz.
Layer 3 · Semiconductor production ecosystem
Design software and IP, wafer fabrication equipment, materials and chemicals, the foundry itself, and advanced packaging and test. Everything required to turn sand into a working advanced logic or memory die.
Binding constraint. The bottleneck moved. For years it was lithography; it is now advanced packaging — bonding a logic die to twelve stacks of HBM on a silicon interposer. TSMC controls over 90% of global CoWoS output, and CoWoS allocation is effectively the revenue ceiling for every accelerator vendor.
Layer 4 · Compute, memory and networking
The accelerators, the memory that feeds them, and the networking that binds them into a single training fabric. Highest revenue growth in the infrastructure stack, and the shortest moat half-life.
Binding constraint. Three separate chokepoints, and memory is currently the tightest. The most profitable chip company in history is guiding margins down because its memory suppliers raised prices — value transferring up the supply chain against the buyer’s wishes is what a genuine bottleneck looks like.
Layer 5 · Data centres, cloud and edge
The physical buildings and the operating businesses that sell capacity: hyperscale cloud, colocation and GPU-rental neoclouds. Four genuinely different business models that are routinely conflated.
Binding constraint. This is no longer a technology sector. It is a leveraged infrastructure sector wearing technology multiples. Once the build-out is financed with debt, SPVs and depreciating collateral rather than operating cash flow, the right analytical toolkit is project finance, not software.
Layer 6 · Data and knowledge infrastructure
The data assets and the systems that govern them: ingestion, storage, cataloguing, cleaning, annotation, provenance and curation. It supports evaluation, retrieval and monitoring as much as training — and it was previously compressed into a single card inside the model layer.
Binding constraint. The binding constraint here is legal rather than physical. Clear, transferable rights to train on third-party content are contested across several jurisdictions simultaneously, and lineage sufficient to evidence those rights cannot be retrofitted once data is already mixed into a corpus. That is a chokepoint with no lead time to shorten.
Layer 7 · AI models and inference
Frontier training, post-training and alignment, inference serving, and the data pipelines that feed all of it. Where capital expenditure is converted into capability.
Binding constraint. There may not be one. Model weights are a depreciating asset; a fab is not. The durable assets at this layer are proprietary data and distribution, not the weights themselves — which is why the vertically integrated players look structurally stronger than the pure-play labs.
Layer 8 · Agentic software and applications
Orchestration frameworks, agent runtimes, vector and operational databases, observability, security, and the application software customers actually pay for.
Binding constraint. Data gravity and systems of record. The scarce input is context, not intelligence — model capability is available to everyone at collapsing prices, but a company’s customer records, permission structures and process knowledge are not. Whoever owns the context owns the deployment.
Layer 9 · Connectivity and communications
The communications layer that links racks, buildings, continents and machines. It was missing from the earlier map: optics sat inside compute silicon, carrier networks appeared nowhere, and the path between the data centre and the edge was drawn as an arrow.
Binding constraint. The chokepoints are geographic rather than manufactured. Metro conduit sits on rights of way that would not be granted again, and three firms outside China can lay a transoceanic system. Neither constraint is relieved by capital expenditure, which is precisely what makes them durable.
Layer 10 · Embodied AI and autonomous systems
Robots, vehicles, drones and devices, plus the actuators, reducers, sensors and batteries beneath them. Three industries at three completely different stages of maturity, routinely conflated.
Binding constraint. The geography inverts here. Layers 3 and 4 concentrate in the Netherlands, Taiwan, Japan and Korea. Layer 10 concentrates in China, which holds 63% of humanoid component supply and over 70% of rare earth magnet production. Adding this layer to a semiconductor-heavy portfolio adds a second, oppositely-directed geopolitical exposure — it does not diversify the first.