Independent research starter · reviewed 4 August 2026

Map the AI market before you form a view.

Technology value, adopter ROI, supplier economics and security returns are different questions. Use this neutral map to collect filings and operating evidence, write a thesis or counter-thesis, decide what would falsify it, and reach your own conclusion.

Follow the eight gatesBrowse public companies
01 · Start here

Keep four questions on separate sheets.

A useful system can help a worker, a customer can see a return, and a supplier can report revenue. None of those observations alone settles what an owner of a security receives.

01

Usefulness

Does a bounded workflow become faster, safer, more accurate or more accessible after review and operating cost?

Collect: accepted-work evidence.
02

Customer economics

Does the adopter retain value after licences, integration, data, human review, governance and change management?

Collect: repeatable payback evidence.
03

Supplier economics

Which layer receives revenue, and what remains after power, depreciation, model renewal, sales expense and competition?

Collect: cash-conversion evidence.
04

Security economics

What growth, margins, reinvestment and durability appear embedded in a market price, and how does a less-perfect outcome change the case?

Collect: downside sensitivities.

The site’s claim-boundary guide explains why task results do not establish sector-wide outcomes.

02 · Follow the money

Follow the money through eight gates.

This is an illustrative path, not a chain of ownership or a claim that one layer captures more value. Demand can enter at several gates, and roles can overlap.

  1. 01

    Power, land & datacentres

    Grid access, cooling, land, construction, capacity and financing.

    Ask: what is actually deliverable?
  2. 02

    Chip equipment & foundries

    Tools, fabs, packaging and manufacturing capacity.

    Ask: where is the constraint?
  3. 03

    Compute, memory & networking

    Accelerators, memory, interconnects and systems.

    Ask: what can substitute?
  4. 04

    Cloud & distribution

    Managed capacity, platforms, developer reach and enterprise channels.

    Ask: does usage become cash?
  5. 05

    Models

    Training, post-training, inference and model operations.

    Ask: can capability sustain a premium?
  6. 06

    Data & context

    Permissions, records, evaluation, governance and integration.

    Ask: what is hard to recreate?
  7. 07

    Applications & workflows

    Repeatable jobs, adoption, outcomes, renewal and support.

    Ask: who owns the job-to-be-done?
  8. 08

    Edge & devices

    Local runtimes, embedded systems, privacy, latency and resilience.

    Ask: where should work run?
03 · Evidence before narrative

Build a file that can disagree with you.

Start with primary documents. Separate reported measures from adjusted measures, management explanations, independent research and your own scenarios.

Useful evidence labels

Official filing
Financial statements, notes and regulatory filings; check definitions and accounting policy.
Official IR
Company presentations, earnings materials and investor-relations pages; useful context, not independent proof.
Research view
An external method or interpretation; retain its date, scope and assumptions.
Your scenario
A stated condition to test, not a forecast or a conclusion.

Research sequence

  1. Map the customer workflow and buyer urgency.
  2. Read revenue quality and customer concentration.
  3. Bridge adjusted measures to cash and per-share outcomes.
  4. List substitutes, switching costs and supply constraints.
  5. Trace capex, depreciation, financing and replacement cycles.
  6. Write a counter-thesis before collecting confirming evidence.
  7. Set a dated decision rule and falsifier.
  8. Record what changed at the next filing.

Use the full cash bridge. Revenue and a headline adjusted metric are not cash conversion. Read the reconciliation, capital expenditure, depreciation, stock compensation or dilution, debt and replacement cycle together.

Keep a countercase alive. A thesis is a provisional explanation. It becomes more useful when the researcher can say which evidence would make them stop relying on it.

04 · Public-company directory

Research public companies by region and market layer.

A diversified, non-exhaustive directory of public suppliers, platforms and adopters referenced across this site. Region means headquarters or primary corporate base, not an exchange listing or broker availability. Each card shows one primary research area only; participation across areas can overlap.

Geographic area

Showing 124 companies.

Advantest

Test and measurement systems for semiconductor production.

Region
Asia-Pacific
Primary area
Compute, memory & networking
Official investor relations

AGCO

Precision-agriculture and autonomous farm-equipment systems provider.

Region
North America
Primary area
Robotics & automation
Official investor relations

Arista Networks

Data-centre networking equipment and software provider.

Region
North America
Primary area
Compute, memory & networking
Official investor relations

BE Semiconductor Industries (Besi)

Semiconductor assembly and packaging-equipment supplier.

Region
Europe
Primary area
Equipment & foundries
Official investor relations

Deere & Company

Agriculture and construction equipment with autonomous and precision systems.

Region
North America
Primary area
Robotics & automation
Official investor relations

GXO Logistics

Contract-logistics operator and named live humanoid-robotics deployment partner.

Region
North America
Primary area
Robotics & automation
Official investor relations

Hewlett Packard Enterprise

Enterprise compute, storage and networking provider.

Region
North America
Primary area
Compute, memory & networking
Official investor relations

Kubota

Agricultural equipment with autonomous and precision-farming systems.

Region
Asia-Pacific
Primary area
Robotics & automation
Official investor relations

Mercedes-Benz Group

Vehicle manufacturer and named humanoid-robotics deployment partner.

Region
Europe
Primary area
Robotics & automation
Official investor relations

Midea Group

Industrial and home-technology group with robotics and automation operations, including KUKA.

Region
Asia-Pacific
Primary area
Robotics & automation
Official investor relations

Renault Group

Vehicle manufacturer and named humanoid-robotics deployment partner.

Region
Europe
Primary area
Robotics & automation
Official investor relations

Rockwell Automation

Industrial-automation hardware, control and software provider.

Region
North America
Primary area
Robotics & automation
Official investor relations

Toyota Motor

Vehicle manufacturer and named commercial-robotics deployment partner.

Region
Asia-Pacific
Primary area
Robotics & automation
Official investor relations

Twilio

Communications software and customer-engagement platform operator.

Region
North America
Primary area
Applications & workflows
Official investor relations

Yaskawa Electric

Industrial-robotics, motion-control and automation provider.

Region
Asia-Pacific
Primary area
Robotics & automation
Official investor relations
Directory boundary.Inclusion, alphabetical order, card prominence and the availability of editorial material do not indicate endorsement. This is not a recommendation or a verdict on any issuer. Open the official route, then read the underlying filings and form your own view.
05 · Private/public-route watch

Separate listed shares from private-company status.

“Official public route filed/approved — not yet public shares” identifies only a filed or approved primary-source route, not a listing. “Private operating company — no public route shown here” identifies current operating context only. Neither status means shares are available, a transfer is approved or an IPO is planned; these entries remain outside the public-company directory.

1X

Private operating company — no public route shown here.

Region
Europe
Research note
Home humanoid and supervised-robotics operating context covered in the robotics guide.
Dated status
Official operating source checked 4 August 2026.
Official operating source

Agility Robotics

Official public route filed/approved — not yet public shares.

Region
North America
Research note
Digit is covered as a logistics-humanoid operating route.
Dated status
SEC Form S-4 source checked 4 August 2026; a filing is not a public listing.
Official SEC filing

Anthropic

Official public route filed/approved — not yet public shares.

Region
North America
Research note
Private AI-model developer included for research context.
Dated status
Confidential draft S-1 submitted 1 June 2026; any offering remains subject to SEC review, market conditions and other factors.
Official status source

Apptronik

Private operating company — no public route shown here.

Region
North America
Research note
Apollo humanoid and operating-data context covered in the robotics guide.
Dated status
Official Series A disclosure checked 4 August 2026.
Official funding source

Figure AI

Private operating company — no public route shown here.

Region
North America
Research note
Humanoid-robotics operating company included for research context.
Dated status
Official warning checked 4 August 2026: unauthorized stock transfers may be void or not recognized.
Official transfer warning

Mistral AI

Private operating company — no public route shown here.

Region
Europe
Research note
Model developer whose Robostral navigation research appears in the robotics guide.
Dated status
Official funding disclosure checked 4 August 2026.
Official funding source

NEURA Robotics

Private operating company — no public route shown here.

Region
Europe
Research note
Physical-AI, training-infrastructure and humanoid context covered in the robotics guide.
Dated status
Official Series C disclosure checked 4 August 2026.
Official funding source

OpenAI

Official public route filed/approved — not yet public shares.

Region
North America
Research note
Private AI-model developer included for research context.
Dated status
Confidential draft S-1 submitted 8 June 2026; timing remains undecided.
Official status source

Unitree Robotics

Official public route filed/approved — not yet public shares.

Region
Asia-Pacific
Research note
Humanoid and quadruped operating context covered in the robotics guide.
Dated status
CSRC approved the official route on 1 July 2026; that approval is not public shares.
Official CSRC approval

Wandercraft

Private operating company — no public route shown here.

Region
Europe
Research note
Medical-exoskeleton and Calvin humanoid operating context covered in the robotics guide.
Dated status
Official Series D disclosure checked 4 August 2026.
Official funding source
06 · Thesis lab

Turn market stories into falsifiable tests.

These are balanced research prompts, not conclusions. Write a claim, a countercase, evidence to collect and a decision rule before treating a narrative as established.

Inference economics ≠ company returns

Claim
Lower cost per useful task could expand usage enough to improve supplier economics.
Countercase
Lower cost may be passed through, while utilisation, sales expense or capital intensity prevents company-level returns.
Evidence to collect
Usage, realised pricing, gross-margin bridge, utilisation, capex and customer renewal.
Falsifier / decision rule
Revisit the claim if falling unit costs do not translate into durable cash conversion after the full infrastructure bill.

Adjusted profit ≠ cash conversion

Claim
A reported adjusted measure may help describe operating progress.
Countercase
Capex, depreciation, stock compensation, dilution or financing needs may change the economic picture for an owner.
Evidence to collect
Reconciliations, cash-flow statement, capex, useful lives, share count and debt notes.
Falsifier / decision rule
Do not rely on the adjusted measure if it cannot be reconciled to a repeatable per-share cash outcome.

Demand can be circular or concentrated

Claim
Committed infrastructure demand could support an expanding ecosystem.
Countercase
Financing, supplier commitments or a small set of customers may make reported demand less independent or less durable than it appears.
Evidence to collect
Customer concentration, related commitments, contract terms, funding sources and recognised versus committed revenue.
Falsifier / decision rule
Rework the thesis if independent end demand and diversified payment capacity are not visible in primary disclosures.

Infrastructure can overbuild

Claim
More capacity may create option value if demand, power and utilisation arrive together.
Countercase
Fungible capacity, faster hardware cycles or weak contracts can turn scarcity into low-return overbuild.
Evidence to collect
Utilisation, contracted capacity, contract duration, power delivery, hardware residual assumptions and refinancing needs.
Falsifier / decision rule
Challenge the claim if capacity grows faster than contracted, cash-generating demand or if residual-value assumptions weaken.

Cheaper models can expand or compress

Claim
Cheaper or open models may unlock demand that was previously uneconomic.
Countercase
Capability parity can pressure prices and margins faster than new demand grows.
Evidence to collect
Price changes, task substitution, usage elasticity, switching costs, customer outcomes and margin response.
Falsifier / decision rule
Reconsider the claim if falling prices increase activity but fail to improve durable economics for the relevant layer.

Adoption can bottleneck in the workflow

Claim
Better models may create useful new production workflows.
Countercase
Data readiness, integration, trust, governance, review burden and change management may limit paid deployment.
Evidence to collect
Time from pilot to production, renewal, user acceptance, error burden, implementation cost and measurable workflow outcomes.
Falsifier / decision rule
Pause the capability-led narrative if adoption evidence remains pilot-heavy or outcomes do not survive operating costs.

For the adjusted-measure test, start with the SEC’s official non-GAAP interpretations. For a dated external research view on efficiency, compute demand and inference pricing, compare Epoch AI’s compute-demand research view and its inference-price data insight with issuer disclosures; they are not settled conclusions.

07 · Several futures

Use scenarios to find the evidence that matters.

A scenario is a condition to monitor, not a probability or a forecast. It is useful only when it changes the documents, metrics or operating evidence a researcher will collect.

Capacity tight

Demand races ahead of supply.

Collect delivery evidence, customer concentration and the cost of meeting demand.

Capacity ample

Infrastructure arrives early.

Collect utilisation, contract quality, replacement economics and financing evidence.

Capability cheaper

Models become more substitutable.

Collect pricing, workflow outcomes, switching evidence and margin response.

Adoption slower

Workflow change takes time.

Collect production conversion, renewal, implementation cost and acceptance evidence.

Power constrained

Delivery depends on energy and grids.

Collect interconnection, power contracts, cooling, site and timing evidence.

Demand broadens

More sectors reach production.

Collect independent customer evidence rather than assume broad demand from one route.

08 · Research record

Keep the thesis date-stamped and reversible.

A research record can make uncertainty visible without pretending to resolve it. It should preserve what was known, what was assumed and what would change the view.

Before a conclusion

  • Record the date, primary documents and definitions used.
  • Separate facts, management statements, external research and your own assumptions.
  • Write the countercase and the evidence that would strengthen it.
  • Note material conflicts if they ever exist.

At the next update

  • Compare reported outcomes with the prior claim and countercase.
  • Read the cash bridge, customer mix and capital commitments again.
  • State which evidence changed and which uncertainty remains.
  • Archive corrections rather than silently rewriting the record.
09 · Falsification

Name what would break the explanation.

A falsifier is not a prediction service. It is a pre-committed test that helps prevent a researcher from moving the goalposts when new evidence arrives.

Supplier route

Could customer internalisation, substitution, inventory or export exposure change the economics?

Cloud route

Could capex, depreciation, utilisation or customer concentration outweigh recognised demand?

Workflow route

Could adoption, implementation burden, error cost or bundling prevent durable paid use?

Infrastructure route

Could power delivery, financing, contract quality or hardware residual value change the return profile?

The durable rule

Follow evidence, operating constraints and cash bridges—then keep the countercase visible.

This page maps useful public information. It does not identify winners or tell anyone what to do with a security.

Check the AI evidenceStudy adoption frictionMap infrastructure