The AI stack, 19 layers deep

The AI market is not one trade — it is a stack of trades with different margins, capex profiles, and risks. Click any segment for the deep-dive: companies, market size, bottleneck status, upstream/downstream dependencies, and the strongest names by composite rank.

Total tracked TAM: $2.60T·Severe-bottleneck segments: 6·Aggregate forecast CAGR (weighted): 17.2%

Demand

Consumer AI

AI-native chat, search, creation, and assistant products targeting individuals.

Workflow incumbents and AI-native challengers embedding AI into sales, service, engineering, finance, and operations.

Apps & security

Model security, data protection, identity, runtime defense, posture management, and AI governance tooling.

AI software

Data platforms, vector stores, observability, eval, orchestration, and the plumbing that turns models into reliable production systems.

Frontier labs and challengers building LLMs, multimodal, reasoning, and agentic systems.

Distribution

Hyperscale clouds, sovereign clouds, and neoclouds reselling GPU capacity, storage, networking, and managed AI services.

Compute systems

AI servers, racks, ODMs, liquid-cooling integration, and the supply chain that assembles raw chips into deployable compute.

Colocation REITs, hyperscale-grade campuses, and GPU-as-a-service operators leasing compute to AI labs and enterprises.

Turbines, transformers, switchgear, behind-the-meter generation, liquid cooling, and the utilities serving new data center load.

Silicon & memory

GPUs, TPUs, custom hyperscaler ASICs, and merchant inference silicon designed for training and serving large models.

HBM, DDR, LPDDR, NAND, and enterprise storage stacks. HBM in particular is the gating factor for accelerator bandwidth.

Switches, optical interconnects, NICs, DPUs, and the optics modules connecting GPUs, racks, and data centers.

Manufacturing

Contract manufacturers that turn chip designs into silicon, at progressively smaller process nodes.

CoWoS, HBM stacking, 2.5D/3D interposers, substrates, and the assembly that ties chips, memory, and IO together.

Lithography, deposition, etch, ion implantation, inspection, metrology, and test equipment used inside fabs.

Licensable CPU, GPU, interconnect, and memory-controller architectures used as building blocks for custom silicon.

Software toolchains used to design, verify, and tape out modern chips, plus reusable IP blocks shipped with the tools.

Inputs

Silicon wafers, specialty gases, photoresists, advanced substrates, rare earths, and high-purity chemicals that feed semiconductor manufacturing.

Disrupted

Companies whose business models face revenue or margin pressure from AI substitution: outsourced IT services, BPO, low-end content, and seat-priced software.

Country dominance by layer

Where each layer of the AI supply chain physically lives. Click any cell to drill into companies in that segment × country intersection.

Country dominance heatmap

Approximate global share by segment (revenue / installed capacity). Click any cell to see companies in that segment from that country.

Low

High

Segment	🇺🇸USA$33.80T	🇹🇼Taiwan$2.40T	🇰🇷S. Korea$2.41T	🇯🇵Japan$387.16B	🇳🇱Netherlands$646.15B	🇩🇪Germany$584.69B	🇫🇷France$84.77B	🇬🇧UK$507.40B	🇨🇳China$343.49B	🇮🇳India$173.38B
Raw inputs bottleneck 3/5	18%	5%	5%	55%	—	5%	—	—	10%	—
EDA bottleneck 4/5	70%	—	—	—	—	18%	—	5%	5%	—
Semi IP bottleneck 3/5	50%	—	—	—	—	—	—	40%	5%	3%
Equipment bottleneck 5/5	38%	—	—	28%	28%	3%	—	—	3%	—
Foundries bottleneck 5/5	12%	65%	12%	2%	—	—	—	—	9%	—
Packaging bottleneck 5/5	12%	50%	12%	8%	—	—	—	—	17%	—
Accelerators bottleneck 4/5	85%	—	—	—	—	—	—	—	8%	—
Memory bottleneck 5/5	22%	—	50%	12%	—	—	—	—	12%	—
Networking bottleneck 4/5	65%	—	—	3%	2%	2%	—	2%	18%	—
Servers bottleneck 2/5	22%	60%	—	—	—	—	—	—	12%	—
Data centers bottleneck 5/5	55%	—	—	2%	2%	3%	—	4%	18%	4%
Power & grid bottleneck 5/5	35%	—	—	4%	—	15%	12%	2%	18%	5%
Cloud bottleneck 3/5	65%	—	—	—	—	—	—	—	22%	—
Models bottleneck 3/5	70%	—	—	—	—	—	4%	—	22%	—
Security bottleneck 2/5	75%	—	—	—	—	—	—	—	8%	4%

Color intensity scales with share. Cells below 2% rendered as —.·Aggregate mcap shown under each country is sourced from real Yahoo data for tracked tickers headquartered there.