What evidence supports the company roles?

Company roles link to public documents. Source labels on this page include: NVIDIA data center; AMD Instinct GPUs; Broadcom custom silicon; Cerebras CS-4; Intel Xeon processors; Arm for cloud AI.

L4Upstream

Compute

Q: Who sits in this layer?

Listed alphabetically, not ranked: NVDA: Supplies data-center GPUs and rack-scale systems for AI training and inference. AMD: Supplies Instinct data-center GPUs used for AI training and inference. AVGO: Designs customized system-on-chip ASICs for customers' own systems, including AI accelerators. CBRS: Builds wafer-scale AI systems, currently marketed for fast inference. INTC: Supplies Xeon data-center CPUs used for AI, HPC and general-purpose workloads, and Gaudi AI accelerators. ARM: Licenses CPU architecture and Neoverse designs that partners build into data-center and cloud AI chips.

GPUs, AI accelerators, and general-purpose compute silicon for training and inference.

Stack position

Where Compute sits in the AI stack

L3b · Foundry
Contract wafer fabrication and advanced packaging that turn accelerator designs into silicon.
L4 · Compute
GPUs, AI accelerators, and general-purpose compute silicon for training and inference.
L4 · HBM / Memory
HBM/DRAM bandwidth beside accelerators, plus enterprise NAND/HDD capacity. US-listed tickers or ADRs only.
L4 · Custom Silicon
ASICs and custom accelerator designs (in-house or merchant silicon) for specific AI workloads.

Icons mark layer roles — not company logos. Position only; not a price chart.

Why this layer matters

The visible centre of the chain, and the layer most people stop at. It is also the layer that cannot run alone: an accelerator needs memory bandwidth, packaging, interconnect, optics, storage, power and cooling around it. Any one of those lagging and the silicon idles.

What is the bottleneck?

Design leadership concentrates here, but the near-term ceiling is usually set elsewhere — packaging, HBM supply, or power — rather than by accelerator design.

A structural observation about supplier concentration and lead times — not a view on any security.

Who sits in this layer?

Company
NVDAData-center GPUsHGXNVL72
NVIDIA
Role
Supplies data-center GPUs and rack-scale systems for AI training and inference.
Evidence
Source: NVIDIA data center
GPU/accelerator platform for AI training and inference.
Stock research
Company
AMDInstinct GPUs
Advanced Micro Devices
Role
Supplies Instinct data-center GPUs used for AI training and inference.
Evidence
Source: AMD Instinct GPUs
Data-center GPUs and CPUs for AI training and inference.
Stock research
Company
AVGOCustom ASICs
Broadcom
Role
Designs customized system-on-chip ASICs for customers' own systems, including AI accelerators.
Evidence
Source: Broadcom custom silicon
Custom AI accelerator (XPU) and networking silicon designer.
Stock research
Company
CBRSCS-4WSE-3
Cerebras Systems
Role
Builds wafer-scale AI systems, currently marketed for fast inference.
Evidence
Source: Cerebras CS-4
Wafer-scale AI accelerator systems for training and inference.
Stock research
Company
INTCXeon CPUsGaudi accelerators
Intel
Role
Supplies Xeon data-center CPUs used for AI, HPC and general-purpose workloads, and Gaudi AI accelerators.
Evidence
Source: Intel Xeon processors
Data-center CPUs and accelerators; also operates contract fabrication.
Stock research
Company
ARMNeoverseCPU architecture licensing
Arm Holdings
Role
Licenses CPU architecture and Neoverse designs that partners build into data-center and cloud AI chips.
Evidence
Source: Arm for cloud AI
CPU architecture and IP licensed into datacenter and edge silicon.
Stock research

The other side

Concentration in accelerator design may not last. Large cloud operators design their own chips, inference can run on cheaper or more specialised parts, and software that makes models portable across vendors lowers the cost of switching. Each of these spreads compute spending across more designs instead of concentrating it.

What could invalidate this picture?

If packaging, HBM and power stop being the limit, accelerator design itself becomes the ceiling and this layer matters more. If custom chips take a growing share of new deployments, or inference moves to lower-cost hardware, design leadership here matters less.

Public 13F filings citing these securities

Research context, not a recommendation. These are facts on file with the SEC; the classification that puts a security in this layer is ours, and the filer made no such claim.

ARK Investment Managementreported a position inAMD
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Appaloosa Managementreported a position inAMD
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Bridgewater Associatesreported a position inAMD
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Coatue Managementreported a position inAMD
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Duquesne Family Officereported a position inAMD
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Fundstrat Capitalreported a position inAMD
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-13 · SEC filing
Tiger Global Managementreported a position inAMD
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
ARK Investment Managementreported a position inAVGO
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Appaloosa Managementreported a position inAVGO
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Bridgewater Associatesreported a position inAVGO
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Coatue Managementreported a position inAVGO
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Fundstrat Capitalreported a position inAVGO
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-13 · SEC filing
Tiger Global Managementreported a position inAVGO
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
ARK Investment Managementreported a position inCBRS
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Coatue Managementreported a position inCBRS
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Tiger Global Managementreported a position inCBRS
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
ARK Investment Managementreported a position inNVDA
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Appaloosa Managementreported a position inNVDA
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Bridgewater Associatesreported a position inNVDA
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Coatue Managementreported a position inNVDA
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing
Fundstrat Capitalreported a position inNVDA
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-13 · SEC filing
Tiger Global Managementreported a position inNVDA
Form 13F-HR for the period ending 2026-06-30, filed 2026-08-14 · SEC filing

Rows below are public SEC Form 13F-HR filings, reported by the filing entity for the period shown and filed later — they are never current holdings. Form 13F covers long US equity positions only: it excludes short positions, cash, non-US holdings and most derivatives, and a reported position may have been sold since. Assigning a ticker to a layer is XingAI's classification, not the filer's. Nothing here is a recommendation, and an institution reporting a position is not a reason to take one.

Coverage boundary

Coverage boundary: US-listed tickers or ADRs only (same rule as TSM / ASML). Important suppliers without a US listing are omitted even when they matter to the stack. That is a coverage rule, not a quality ranking.

No hand-listed notable omissions for this layer yet. The US-list / ADR rule still applies.

Checked 2026-10-09. Not exhaustive.

Same track

L4 Custom Silicon L4 HBM / Memory

Descriptive positioning only. The positioning label says whether a company's business is exposed to a layer, never whether to buy it. Layers with no exposure are not recommendations to add any.