The AI stack
Coverage is partial
Chygraph maps sourced relationships across nine domains of the AI stack, from capital and physical infrastructure to models, organizational use, and governance. Shared objects retain one identity across the map.
This dated collection covers selected arrangements and commitments. Its sources establish particular claims, with their limits retained alongside them. Coverage is partial.
- Governance
- Integration
- System
- Model
- Experience
- Platform
- Hardware
- Substrate
- Capital
The structure could not be loaded
Sources and relationships
These 17 relationships draw on 11 public sources. The collection reflects evidence assembled as of September 24, 2026. Announcements and commitments retain their stated status. A connection does not establish causation, operational dependency, or completed compliance.
Signed commitments
The European Commission lists OpenAI as a signatory to the General-Purpose AI Code of Practice, comprising Transparency, Copyright, and Safety and Security
Listed signatory
The public list establishes OpenAI’s Code signatory status but does not publish its exact signature date or signed form. Full-Code coverage follows the Commission’s distinction between the general signatory list and the separately named chapter-only signatory. Signature is not proof of adherence or compliance.
- The General-Purpose AI Code of Practice (2026-07-31)The 3 chapters of the code; Signatories of the code of practice — OpenAI listed; xAI separately identified as signing only Safety and Security
Share documentation
For covered models, OpenAI commits to provide downstream providers the current model information intended for them, subject to confidentiality safeguards
Documented commitment
A selected commitment drawn from the final Code, attributed to OpenAI as a listed signatory. Fulfilment is not established by these sources.
- Final Code — Transparency chapter (2025-07-10)Commitment 1; Measure 1.2, pages 5–6
Disclose on request
OpenAI commits to provide the AI Office requested model information necessary for its statutory tasks, within the period specified in the request
Documented commitment
A selected commitment drawn from the final Code, attributed to OpenAI as a listed signatory. Fulfilment is not established by these sources.
- Final Code — Transparency chapter (2025-07-10)Measure 1.2, page 5
Respect reservations
When crawling for text and data mining and model training, OpenAI commits to use crawlers that read and follow robots.txt instructions
Documented commitment
A selected commitment drawn from the final Code, attributed to OpenAI as a listed signatory. Fulfilment is not established by these sources. This selects one protocol commitment; it does not exhaust the chapter’s treatment of rights reservations.
- Final Code — Copyright chapter (2025-07-10)Measure 1.3(1)(a), page 5
Handle complaints
OpenAI commits to provide an electronic mechanism for substantiated rightsholder complaints about the chapter’s commitments and to handle them diligently
Documented commitment
A selected commitment drawn from the final Code, attributed to OpenAI as a listed signatory. Fulfilment is not established by these sources. The commitment includes exceptions for manifestly unfounded or already answered identical complaints.
- Final Code — Copyright chapter (2025-07-10)Measure 1.5(1)–(2), page 6
Notify framework
For models within this chapter’s scope, OpenAI commits to give the AI Office unredacted access to its safety framework and confirmed updates within five business days
Documented commitment
A selected commitment drawn from the final Code, attributed to OpenAI as a listed signatory. Fulfilment is not established by these sources.
- Final Code — Safety and Security chapter (2025-07-10)Commitment 1, page 6; Measure 1.4, page 10
Submit model report
For models within this chapter’s scope, OpenAI commits to give the AI Office a safety and security model report by market placement, subject to the Code’s stated exceptions
Documented commitment
A selected commitment drawn from the final Code, attributed to OpenAI as a listed signatory. Fulfilment is not established by these sources. National-security redactions and a conditional delay of up to 15 business days are addressed in Measure 7.7.
- Final Code — Safety and Security chapter (2025-07-10)Commitment 7, pages 18–19; Measure 7.7, pages 22–23
Finance
Crusoe, Blue Owl-managed funds and Primary Digital announced a $15 billion joint venture funding the original Abilene campus expansion
Announced financing
Announced project financing is not a statement of cash fully disbursed or campus completion
- Crusoe, Blue Owl Capital, and Primary Digital Infrastructure enter second phase of $15 billion joint venture (2025-05-21)Opening paragraphs and joint-venture terms
Cool
Crusoe specified direct-to-chip, closed-loop non-evaporative cooling for the eight-building campus
Specified design
This is a published design statement; no measured water or energy consumption is inferred
- Crusoe, Blue Owl Capital, and Primary Digital Infrastructure enter second phase of $15 billion joint venture (2025-05-21)Eight-building design and cooling paragraph
Deliver
Oracle began delivering NVIDIA GB200 racks to the original Abilene site in June 2025
Reported delivery
The number of installed racks and current total operating capacity are not established here
- Stargate advances with 4.5 GW partnership with Oracle (2025-07-22)Paragraph beginning Meanwhile, construction of Stargate I
Run workloads
OpenAI reported early training and inference workloads at Abilene, operating on Oracle Cloud Infrastructure
Reported operation
No named model, user request or downstream service is assigned to this campus
- OpenAI, Oracle, and SoftBank expand Stargate with five new AI data center sites (2025-09-23)Paragraph beginning The new Stargate sites being developed by Oracle
Train
OpenAI describes Astra alignment work spanning pre-training-data composition and grading during reinforcement learning
Reported development
Training-data contents, grader identities and detailed recipes are not disclosed by this summary
- Safety overview: GPT-6 Astra (2026-09-03)Numbered point 3
Evaluate
OpenAI reports a suite of alignment evaluations for Astra, including authorized-scope and boundary behavior
Reported evaluation
The edge records reported evaluation activity, not independent confirmation of performance or universal safety
- Safety overview: GPT-6 Astra (2026-09-03)Numbered point 3 and linked alignment evaluations
Make available
OpenAI documents Astra access in ChatGPT Work and Codex, subject to plan and workspace permissions
Documented availability
Model choice, plan and permissions vary; this does not say every task uses Astra
- ChatGPT Work and CodexOpening paragraphs; Models in Work and Codex; Astra usage
Provide access
Airbnb announced expanded access to OpenAI frontier models including Astra through OpenAI APIs and Amazon Bedrock
Announced access agreement
No provider allocation, contract value or per-request routing is disclosed
- Expanding access to OpenAI frontier models, including GPT-6 Astra (2026-09-23)Opening agreement and access paragraphs
Embed
Airbnb reports Codex use and early Astra work within its engineering tooling, including debugging and system design
Reported organizational use
This does not establish a particular agent configuration or that Astra runs in ChatGPT Work at Airbnb
- Expanding access to OpenAI frontier models, including GPT-6 Astra (2026-09-23)Engineering tooling and early Astra paragraphs
chygraph (or c-hy-graph; i.e., a "complex hypergraph")
Each compound has its own relationships and can participate in others