--- id: ECON-IND-001 title: "The AI Industry Structure" alternative_names: - "AI Value Chain" - "AI Market Structure" category: "Economy & Policy → Industry Structure" volume: IX chapter: 50 entry_type: "Standard Article" level: 3 status: "Draft" last_reviewed: 2026-10-15 version: "0.1.0" prerequisites: - "ECON-COMP-001" related_entries: - "TECH-HW-001" - "TECH-LLM-001" - "ECON-INFRA-001" - "ECON-MEAS-001" - "GOV-EXPORT-001" - "ORG-LABS-001" --- # The AI Industry Structure ## 1. Concise Definition The AI industry structure is the organisation of firms and relationships that supply the inputs to, build, distribute and apply AI systems: semiconductor and equipment suppliers, cloud and data-centre operators, developers of foundation models, and application and integration companies. It is usually described as a layered value chain with concentration at several layers and with financial and contractual ties between them. ## 2. Overview & Significance Training a frontier model requires large quantities of specialised compute ([[ECON-COMP-001]](../vol09_economy/ECON-COMP-001_compute_economics.md)), so the capital intensity of the upstream layers shapes who can compete downstream. Competition authorities have examined whether partnerships between cloud providers and model developers affect access to inputs. This article describes structure and the evidence behind claims of concentration; market sizes and valuations change quickly and are deliberately not tabulated (see `FACT-001` and `MODEL-001`). ## 3. Historical Development Before about 2012 AI research and products were largely developed in academia and in corporate research groups with modest compute. After the deep-learning breakthroughs ([[TECH-DL-001]](../vol03_technical/TECH-DL-001_deep_learning.md)) and the scaling of language models ([[TECH-LLM-001]](../vol03_technical/TECH-LLM-001_large_language_models.md)), development moved toward organisations able to fund large clusters. Bommasani et al. (2021) introduced the term "foundation model" for large models trained on broad data and adapted to many tasks, and observed that a shared base propagates its flaws to every downstream system; that report is the standard reference for the layered view of the field. Between 2019 and 2024 several multi-billion-dollar investments linked cloud providers and model developers. The US Federal Trade Commission's January 2024 orders under Section 6(b) of the FTC Act, sent to Alphabet, Amazon, Microsoft, Anthropic and OpenAI regarding three investment pairs (Microsoft and OpenAI; Amazon and Anthropic; Alphabet and Anthropic), led to a staff report on 17 January 2025 (Federal Trade Commission, 2025). The UK Competition and Markets Authority published an update paper on AI foundation models on 11 April 2024 and invited comment on further partnerships, stating it had reached no conclusion on them at that time (CMA, 2024). Corporate structures of model developers have also changed. This article does not describe individual developers' restructurings because the primary announcements could not be retrieved in this session [VERIFICATION NEEDED for any such claim]; see [[ORG-LABS-001]](../vol10_people_organizations/ORG-LABS-001_frontier_ai_laboratories.md). ## 4. Technical / Conceptual Explanation **A layered model.** 1. *Equipment and fabrication* (lithography tools, foundries, memory), described in [[ECON-INFRA-001]](../vol09_economy/ECON-INFRA-001_energy_data_centers_and_semiconductor_supply_chains.md). 2. *Accelerator and systems design* (GPUs and custom accelerators; see [[TECH-HW-001]](../vol03_technical/TECH-HW-001_ai_hardware.md)). 3. *Cloud and data centres* providing compute capacity, often with committed-spend contracts. 4. *Foundation-model development* (pretraining, post-training). 5. *Model access and distribution*: APIs, consumer assistants, open-weight releases. 6. *Applications and integration*: software vendors, enterprises, public bodies. **Why concentration arises.** Fixed costs of training runs, learning-by-doing, scarce engineering talent, and access to compute are all candidate barriers. The FTC staff report identifies, as competition concerns, that partnerships may affect access to inputs such as computing resources and engineering talent, may raise switching costs for AI developers, and may give cloud providers access to sensitive technical and business information (Federal Trade Commission, 2025). It also describes arrangements in which developers commit a large portion of a cloud partner's investment to that partner's cloud services and receive discounted compute. These are staff findings and concerns rather than adjudicated violations. **Open-weight models** (released parameters) lower entry barriers at layers 5-6 but not at layer 4 for frontier-scale training. ## 5. Types, Components & Classifications | Actor type | Role | Evidence cited | | :--- | :--- | :--- | | Cloud service providers | Capacity and distribution; investors in developers | FTC (2025) | | Independent frontier developers | Foundation models | CMA (2024) | | Accelerator vendors | Supply of GPUs/systems | NVIDIA 10-K (2026) | | Open-weight developers | Reduce downstream entry cost | Bommasani et al. (2021) as framework | | Application firms | Integration and vertical products | not quantified here | ## 6. Applications & Concrete Examples One dated concentration indicator: NVIDIA's annual report for fiscal 2026 (year ended 25 January 2026) states revenue of USD 215.9 billion (up 65% on the prior year), that its revenue is "concentrated among a limited number of direct and indirect customers", and that one direct customer accounted for 22% of revenue and another for 14%, both mainly in its Compute & Networking segment (NVIDIA, 2026). This describes one supplier's customer concentration, not the market share of the AI hardware sector, and does not identify the customers. Counts of foundation models by developer type are tracked by Epoch AI, whose notable-models dataset defines "notable" by criteria including state-of-the-art benchmark improvement, more than 1,000 citations, historical relevance or significant use, and was last updated 2026-10-15 (Epoch AI, 2026). The share of notable models from industry is not quoted here because the figure was not retrieved from the dataset itself. ## 7. Advantages, Limitations & Unresolved Questions - *Cited concerns*: access to inputs, switching costs, information advantages (FTC, 2025); concentration among a small number of firms and the effect of partnerships on existing market power (CMA, 2024). - *Counter-considerations* (argued by commentators; not tested here): partnerships supply capital that new entrants otherwise lack; open-weight releases and falling inference cost broaden access. - *Unresolved*: whether current structure is persistent; how to measure concentration when models, clouds and applications are interlinked; the profitability of the model layer, for which this article uses no figures. ## 8. Current State of Knowledge Evidence is strongest for upstream concentration in specialised hardware and for the existence of cloud-developer financial ties (Well Supported by regulatory reports and company filings). Evidence on downstream competition, durable margins and eventual market shares is Preliminary or Contested. Enforcement outcomes after the January 2025 FTC report were not retrieved. ## 9. Future Trajectory & Strategic Outlook Observed trends include continued capital expenditure on data centres (see [[ECON-INFRA-001]](../vol09_economy/ECON-INFRA-001_energy_data_centers_and_semiconductor_supply_chains.md)) and regulatory attention to partnerships. No quantitative forecast is registered here; readers should consult `FCST-REG-001` and `FACT-001` for dated figures. ## 10. Related Entries & Knowledge Graph - **Foundational Concepts**: - [[ECON-COMP-001] The Economics of Compute and Scaling](../vol09_economy/ECON-COMP-001_compute_economics.md) - **Related Methods & Technologies**: - [[TECH-HW-001] AI Hardware](../vol03_technical/TECH-HW-001_ai_hardware.md) - [[TECH-LLM-001] Large Language Models](../vol03_technical/TECH-LLM-001_large_language_models.md) - **Related economy and governance**: - [[ECON-INFRA-001] Energy, Data Centers and Semiconductor Supply Chains](../vol09_economy/ECON-INFRA-001_energy_data_centers_and_semiconductor_supply_chains.md) - [[ECON-MEAS-001] Measuring the AI Economy](../vol09_economy/ECON-MEAS-001_measuring_the_ai_economy.md) - [[GOV-EXPORT-001] Export Controls and AI Geopolitics](../vol08_safety_governance/GOV-EXPORT-001_export_controls_and_ai_geopolitics.md) - [[ORG-LABS-001] Frontier AI Laboratories](../vol10_people_organizations/ORG-LABS-001_frontier_ai_laboratories.md) ## 11. References & Foundational Literature 1. Bommasani, R., et al. (2021). *On the Opportunities and Risks of Foundation Models* (114 authors; first author Rishi Bommasani). arXiv:2108.07258. https://arxiv.org/abs/2108.07258. [Defines foundation models and the shared-base propagation of flaws; preprint, Stanford CRFM]. 2. Federal Trade Commission. (2025, January 17). *FTC Issues Staff Report on AI Partnerships & Investments Study*. https://www.ftc.gov/news-events/news/press-releases/2025/01/ftc-issues-staff-report-ai-partnerships-investments-study. [Firms studied, partnership terms and competition concerns; information as of September 2024]. 3. Competition and Markets Authority. (2024, April 11). *AI Foundation Models: Update Paper*. https://www.gov.uk/government/publications/ai-foundation-models-update-paper. [Concern about small number of firms and partnerships; publication date confirmed via secondary summary; primary page not opened: VERIFICATION NEEDED for exact URL]. 4. NVIDIA Corporation. (2026). *Form 10-K for the fiscal year ended January 25, 2026*. U.S. Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm. [Revenue USD 215,938 million; customer concentration of 22% and 14%]. 5. Epoch AI. (2026). *AI Models* (notable AI models dataset). https://epoch.ai/data/ai-models. [Definition of "notable" models; page dated 2026-10-15; no share figure used]. ## 12. Evidential Classification Matrix | Section / Claim | Evidence Category | Supporting Source | Confidence / Status | | :--- | :--- | :--- | :--- | | Foundation-model concept | Well Supported | Ref 1 | High | | Cloud-developer partnership terms and concerns | Well Supported (regulator staff report) | Ref 2 | Moderate-High | | Regulatory concern about concentration (UK) | Well Supported | Ref 3 | Moderate | | NVIDIA customer concentration | Established (audited filing) | Ref 4 | High | | Persistence of concentration downstream | Contested | none | Unresolved | ## 13. Revision History - **v0.1.0** (2026-10-15): Initial draft. Citations checked against official pages where marked; not independently fact-checked.