The full AI supply chain mapped to public tickers — ~150 companies across 17 layers, from $ASML lithography and $MU/SK Hynix HBM through $COHR photonics and $GEV/$OKLO power to $CRWV neoclouds — with the bottleneck in each layer and a live basket to track it.
More Macroplane supply-chain guides · Baskets · Investment theses · Macro trends
The AI supply chain is the end-to-end network of public companies that design, manufacture, power, cool, and operate the hardware behind artificial intelligence — from EUV lithography and HBM memory through optical interconnect, power and cooling, all the way to the neocloud operators renting out GPUs. This page maps that chain layer by layer, names the publicly-listed leader in each, and links every layer to a live basket on Macroplane so you can drop in a ticker and trace who depends on whom.
Most "AI stock" lists stop at $NVDA and a handful of hyperscalers. The actual buildout runs through roughly 150 public companies across 17 supply-chain layers, and the most asymmetric exposure is usually three or four links upstream of the names everyone already owns. This is the map.
👉 Trace any link live: every layer below is a basket on Macroplane with continuously-updated performance, and every company has a supplier/customer graph you can open from its ticker page.
Each layer is an equal-weight index, so the spread between the leader and the laggard is doing the talking. In Custom Silicon, $MRVL ran +56% while Alchip ($3661.TW) fell 15% over the same window — the basket view makes that dispersion obvious before you open a single name.
The AI supply chain isn't a list — it's a directed graph. GPUs need HBM; HBM needs advanced packaging; packaging needs metrology tools; the whole fab needs EUV lithography; the finished accelerators need optical interconnect to talk to each other, power to run, and cooling to not melt — then a neocloud rents the cluster out by the hour. Money and risk concentrate at the bottlenecks: the single-supplier choke points where one company gates an entire layer (think $ASML in EUV, $TSM in leading-edge foundry, SK Hynix/$MU in HBM).
Open any ticker and you get exactly this: suppliers feeding in from the left, the company in the centre, customers pulling from the right, with revenue-exposure percentages on the edges that matter. The example above is $AAOI in the Photonics / CPO layer — 68 relationships traced from the filings.
We group the 17 layers into five tiers, top of the stack to bottom.
The chips themselves and the memory that feeds them.
You can't have chips without the machines that print them.
Once you have thousands of accelerators, the bottleneck becomes moving data between them.
The companies view behind each basket carries the per-name detail — market cap, P/E, 52-week range, and a short-term sparkline — so a layer reads at a glance: $COHR and $LITE anchor the large-cap end, while the sub-$3B names ($POET, Sivers at $SIVE.ST, $LWLG) are where the optical-interconnect optionality sits.
The buildout is increasingly gated not by chips but by megawatts and heat.
The demand side and the materials underneath everything.
If you only remember four choke points, remember these:
The investing edge is that these bottlenecks sit upstream of the obvious names. When you open a company on Macroplane, the supplier/customer graph makes the dependency explicit — you can see, for example, which neocloud's economics ride on which memory supplier, or which photonics name is single-threaded through one laser vendor.
The broad AI/robotics ETFs (BOTZ, AIQ, ROBO, IRBO) give you a cap-weighted slug of mostly mega-cap compute and a long tail of loosely-related software. They systematically underweight the upstream bottleneck layers — packaging, photonics, power — where the asymmetry lives, and they can't show you why a name matters (its position in the chain).
| Broad AI ETF | The Macroplane map | |
|---|---|---|
| Coverage | Mega-cap compute + software tail | All 17 supply-chain layers, ~150 names |
| Upstream bottlenecks | Underweighted | Each is its own basket |
| Shows why a name matters | No | Supplier/customer graph per ticker |
| Customize / weight | No | Clone any basket, set weights |
| Cost | 0.5–0.75%/yr | Free to track |
This map is a hub. For a deeper read on any branch, the layer-specific guides go further:
The AI supply chain is the network of companies that produce the physical infrastructure for artificial intelligence: chip designers, memory makers, the equipment that fabricates and packages chips, the optical and electrical interconnect that links them, the power and cooling that runs the data center, the contractors that build it, and the cloud operators that rent it out. It's a dependency graph, not a single industry — each layer is a customer of the one upstream of it.
By bottleneck control: $ASML (EUV lithography), $TSM (leading-edge foundry), and SK Hynix ($000660.KS), $MU, and Samsung ($005930.KS) in HBM memory. By compute: $NVDA, $AMD, $AVGO, and $MRVL. By the increasingly-binding power constraint: $GEV, $CEG, $VST, $OKLO, and $SMR. The full set spans ~150 names — the baskets on Macroplane organize them by layer.
The tightest choke points are EUV lithography (single supplier, $ASML), leading-edge foundry capacity ($TSM and its CoWoS advanced-packaging lines), HBM memory (three suppliers, sold out in advance), and — increasingly — electrical power and grid interconnection, which is why nuclear and on-site generation names re-rated as AI trades.
Broad AI ETFs are cap-weighted toward mega-cap compute and dilute into loosely-related software, underweighting the upstream layers (packaging, photonics, power) where the asymmetric exposure sits. They also can't show you a company's position in the chain. Mapping the supply chain lets you target a specific layer and see each name's supplier/customer dependencies directly.
Each layer above is a live basket on Macroplane with continuously-updated performance, and every company page shows its supplier/customer graph, deal flow, and financials. Start from any ticker and traverse upstream or downstream to see who depends on whom.
The AI supply chain is the network of companies that produce the physical infrastructure for artificial intelligence: chip designers, memory makers, the equipment that fabricates and packages chips, the optical and electrical interconnect that links them, the power and cooling that runs the data center, the contractors that build it, and the cloud operators that rent it out. It's a dependency graph, not a single industry — each layer is a customer of the one upstream of it.
By bottleneck control: $ASML (EUV lithography), $TSM (leading-edge foundry), and SK Hynix ($000660.KS), $MU, and Samsung ($005930.KS) in HBM memory. By compute: $NVDA, $AMD, $AVGO, and $MRVL. By the increasingly-binding power constraint: $GEV, $CEG, $VST, $OKLO, and $SMR. The full set spans ~150 names — the baskets on Macroplane organize them by layer.
The tightest choke points are EUV lithography (single supplier, $ASML), leading-edge foundry capacity ($TSM and its CoWoS advanced-packaging lines), HBM memory (three suppliers, sold out in advance), and — increasingly — electrical power and grid interconnection, which is why nuclear and on-site generation names re-rated as AI trades.
Broad AI ETFs are cap-weighted toward mega-cap compute and dilute into loosely-related software, underweighting the upstream layers (packaging, photonics, power) where the asymmetric exposure sits. They also can't show you a company's position in the chain. Mapping the supply chain lets you target a specific layer and see each name's supplier/customer dependencies directly.
Each layer above is a live basket on Macroplane with continuously-updated performance, and every company page shows its supplier/customer graph, deal flow, and financials. Start from any ticker and traverse upstream or downstream to see who depends on whom.