---
title: The Map of AI
date: August 21, 2026
slug: map-of-ai
status: published
canonical: https://moiz.ai/writing/map-of-ai
html: https://moiz.ai/writing/map-of-ai
markdown: https://moiz.ai/writing/map-of-ai.md
---

# The Map of AI

*Eight layers sit behind the answer on your screen, from the grid to the product in your hand.*

<!-- block:map-ai-block-tldr -->
**TL;DR.** One generated token depends on energy, chips, datacenters, models, inference, routing, agent control, and the surface a person uses. Each handoff controls a different part of the result.

## Agent collaboration brief

This is the complete Markdown twin of the HTML article. If it is pasted into a coding agent, treat the user's current instruction as the task and this document as the working context. Distinguish sourced facts, company-reported figures, editorial judgments, and predictions. Preserve qualifiers unless new evidence justifies a change.

### Current product contract

- HTML article: https://moiz.ai/writing/map-of-ai
- Markdown twin: https://moiz.ai/writing/map-of-ai.md
- Status: published. The HTML article is indexable and listed in the sitemap. The Markdown twin stays noindex and points to the HTML canonical. The article remains absent from the homepage and writing index by editorial choice.
- The living plate is the primary map. Its paper is rendered with `@paper-design/shaders-react`; keep the parchment, engraved type, restrained sepia palette, and Tufte reading frame.
- ChatGPT sits first and Claude immediately after it in the surface row as an editorial ordering choice. Audience reach remains outside that ordering.
- Keep labs and products distinct: SpaceXAI and Mistral AI belong in L4; Grok and Vibe belong in L8. The same rule applies to Moonshot AI/Kimi and Meta Superintelligence Labs/Meta AI. Thinking Machines Lab and Inkling belong in L4; Tinker is a training and customization platform. Its mass-consumer reach remains unverified.
- Every actor name on the living plate links to the official destination stored with that actor in the map registry.
- The token voyage remains a separate secondary component below the living plate. Keep its route off the primary plate.
- Band height and type size on the living plate express editorial emphasis. Market share, reach, and bargaining power remain outside those visual encodings. The plate renders one editorial survey; the alternate-scenario controls were removed.
- Company-reported weekly users, monthly users, downloads, contracted capacity, installed capacity, API traffic, and revenue are different metrics. Never merge them into one unlabeled ranking.

### Repository source of truth

- `src/app/writing/map-of-ai/content.tsx`: article prose, sidenotes, supplements, and research links.
- `src/app/writing/map-of-ai/map-data.ts`: stable factual placements, destinations, one-line claims, evidence state, and review history.
- `src/app/writing/map-of-ai/map-presentation.ts`: editorial ordering, qualitative salience, curated views, and render-only layout configuration.
- `src/app/writing/map-of-ai/evidence.ts`: public receipt metadata and supporting locators only.
- `src/app/writing/map-of-ai/markdown.ts`: this generated handoff and Markdown serialization.
- `src/components/article/living-plate.tsx`: primary interactive plate and shader treatment.
- `src/components/article/token-voyage.tsx`: separate token route.
- `src/components/article/territory-map.tsx`: engraved atlas details used inside the article sections.
- `src/app/writing/article.css`: article-only visual rules. Keep these styles scoped away from the homepage.

Edit the shared source modules so HTML and Markdown stay aligned; the rendered `.md` response is generated from them. For prose work, keep all section titles, research-link labels, predictions, sidenotes, and matrix content represented in this twin.

### Verification before shipping

```sh
bunx tsc --noEmit
bun run lint
bun test
bun run check:article
bun run check:freerange
bun run build
```

Then verify `/writing/map-of-ai` and `/writing/map-of-ai.md` return 200, the HTML canonical is indexable and present in the sitemap, the Markdown twin stays noindex with the HTML canonical, all official links are reachable, and the visual page has no horizontal overflow or console errors.

## The map

The living plate reads from atoms to bits. The diagram gives coding agents the stack topology; the registry below carries every actor, official destination, editorial weight, claim state, and one-line reason for inclusion.

<!-- block:map-ai-block-presentation-legend -->
I sized the bands by editorial emphasis; market share is outside the scope of this map.

<!-- block:map-ai-block-work-harness-rail -->
Inside Surfaces sits a class worth naming separately: work harnesses: choose, route, or govern the model beneath the work. Factory uses automatic model selection per Droid session, inside its own product; Perplexity Computer uses multi-model orchestration, selecting specialist models by subtask; Cursor uses per-request routing across a managed pool, with policy controls; OpenCode uses user-selected providers with configured fallback ordering. Each product uses a different mechanism.

<!-- block:map-ai-block-evaluation-rail -->
Evaluation and observability form a return rail across all eight layers: production traces to scoring and human review to regression suites to agent, tool, and route changes to re-evaluation. The stack carries work forward; the rail carries measurement back.

<!-- block:map-ai-block-hidden-substrate-rail -->
Five systems cross the numbered request path without becoming layers of their own: chip enablement, network fabric, data and context, trust and identity, distribution.

```mermaid
flowchart BT
  GOV["Governance & the commons · rules can reprice every layer"]
  subgraph STACK[Follow one token, bottom-up]
    direction BT
    ENERGY["L1 Energy & grid · PJM, GE Vernova, Constellation, Talen · decides whether the power exists to be bought"]
    CHIPS["L2 Chips & fabs · TSMC, ASML, Nvidia · where compute begins"]
    DC["L3 Datacenters · hyperscalers, neoclouds · buildings move at civil-engineering speed"]
    MODELS["L4 Models · OpenAI, DeepMind, Anthropic, SpaceXAI, Meta, Qwen, DeepSeek, Kimi, Z.ai, Mistral, Cohere, Thinking Machines · fed by Mercor, Surge, Common Crawl, Innodata"]
    INFER["L5 Inference & serving · clouds and specialists · continuous per-token work"]
    ROUTE["L6 Routing & gateways · the youngest layer · brokers model demand"]
    CONTROL(["L7 Agent control · MCP, frameworks, evals · turns tokens into work"])
    SURFACES["L8 Surfaces · ChatGPT, Claude, Gemini, Grok, Kimi, Qwen, DeepSeek, Meta AI, Vibe + app specialists"]
    ENERGY --> CHIPS --> DC --> MODELS --> INFER --> ROUTE --> CONTROL --> SURFACES
  end
  GOV -. rules and procurement can reprice any layer .-> STACK
```

### L1 · Energy & grid

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-energy-pjm | [PJM Interconnection](https://www.pjm.com/) | anchor | 2026-07-26 | 2026-07-26 | verified | pjm-bra-2028-2029 | is the grid operator whose 2028/2029 capacity auction cleared $16.4 billion with every price at the $325 cap, and which still procured 6,831.3 MW less unforced capacity than its own reliability requirement |
| map-ai-placement-energy-ge-vernova | [GE Vernova](https://www.gevernova.com/) | anchor | 2026-07-26 | 2026-07-26 | qualified | ge-vernova-q2-2026, ge-vernova-bofa-2026 | builds the gas turbines that are the fastest route to firm new capacity; its company-reported backlog and slot reservations grew from 100 to 116 GW in a quarter, and it is taking reservations into 2031 |
| map-ai-placement-energy-constellation | [Constellation Energy](https://www.constellationenergy.com/) | anchor | 2026-07-26 | 2026-07-26 | verified | constellation-crane-2024 | is restarting Three Mile Island Unit 1 as the Crane Clean Energy Center under a 20-year power purchase agreement with Microsoft, roughly 835 MW brought back because this demand arrived |
| map-ai-placement-energy-talen | [Talen Energy](https://www.talenenergy.com/) | supporting | 2026-07-26 | 2026-07-26 | verified | talen-aws-8k-2025 | sells an existing nuclear plant's output straight to datacenter load: 1,920 MW to AWS through 2042, which its own 8-K puts at about $18 billion in total revenues at full quantity |
| map-ai-placement-energy-bloom | [Bloom Energy](https://www.bloomenergy.com/) | supporting | 2026-07-26 | 2026-07-26 | qualified | bloom-oracle-2025, bloom-oracle-2026 | sites fuel cells at the datacenter to bypass grid waits: its company-reported Oracle partnership promises onsite power within 90 days, and was expanded in April 2026 to up to 2.8 GW |
| map-ai-placement-energy-siemens-energy | [Siemens Energy](https://www.siemens-energy.com/) | supporting | 2026-07-26 | 2026-07-26 | qualified | siemens-energy-q4-fy2025 | supplies the grid hardware between a generator and a rack; its company-reported order backlog reached a record EUR 138 billion at the close of fiscal 2025, with data centers named as a driver |
| map-ai-placement-energy-hitachi-energy | [Hitachi Energy](https://www.hitachienergy.com/) | supporting | 2026-07-26 | 2026-07-26 | qualified | hitachi-transformer-investment-2025 | makes the transformers and switchgear that gate energization even where generation exists; more than $1 billion of US plant expansion was announced in 2025, including $457 million for a Virginia transformer works |
| map-ai-placement-energy-nextera | [NextEra Energy](https://www.nexteraenergy.com/) | supporting | 2026-07-26 | 2026-07-26 | qualified | nextera-q1-2026 | was selected to build 9.5 GW of new gas-fired generation for large load in Texas and Pennsylvania, and its company-reported data center hub pipeline has passed 30 hubs |

### L2 · Chips & fabs

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-chips-tsmc | [TSMC](https://www.tsmc.com/english) | anchor | 2026-07-20 | 2026-08-12 | qualified | stanford-ai-index-2026, nist-tsmc-2026, tsmc-cowos | fabricates a large share of leading AI accelerators and custom chips and supplies CoWoS packaging that integrates logic with high-bandwidth memory; the evidence establishes concentration while leaving the rest of accelerator fabrication outside the claim |
| map-ai-placement-chips-asml | [ASML](https://www.asml.com/) | anchor | 2026-07-20 | 2026-07-22 | verified | asml-euv | is the sole commercial supplier of EUV lithography systems used for the most advanced semiconductor manufacturing |
| map-ai-placement-chips-nvidia | [Nvidia](https://www.nvidia.com/) | anchor | 2026-07-20 | 2026-07-22 | qualified | stanford-ai-index-2026, nvidia-cuda | designs the dominant merchant accelerators and CUDA; an estimated majority of global AI compute uses Nvidia systems, with material uncertainty |
| map-ai-placement-chips-amazon-trainium | [Amazon (Trainium)](https://aws.amazon.com/ai/machine-learning/trainium/) | supporting | 2026-07-22 | 2026-07-22 | qualified | anthropic-aws | designs Trainium for AWS; Anthropic's company-reported figure said more than one million Trainium2 chips were training and serving Claude |
| map-ai-placement-chips-google-tpu | [Google (TPU)](https://cloud.google.com/tpu) | supporting | 2026-07-20 | 2026-07-22 | verified | google-tpu | designs TPU systems whose capacity reaches customers through Google Cloud's managed channel |
| map-ai-placement-chips-broadcom | [Broadcom](https://www.broadcom.com/) | supporting | 2026-07-20 | 2026-07-22 | qualified | openai-broadcom | co-develops custom accelerators and networking; its OpenAI program carries an announced deployment target, while installed capacity remains undisclosed |
| map-ai-placement-chips-amd | [AMD](https://www.amd.com/en.html) | supporting | 2026-07-20 | 2026-07-22 | verified | amd-rocm | sells Instinct accelerators and the ROCm software stack as a merchant alternative to Nvidia |

### L3 · Datacenters

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-datacenters-microsoft | [Microsoft](https://azure.microsoft.com/) | anchor | 2026-07-20 | 2026-07-22 | verified | microsoft-openai-2026 | operates Azure and remained OpenAI's primary cloud partner at the July 2026 cutoff, while the partnership allowed additional serving providers |
| map-ai-placement-datacenters-amazon-aws | [Amazon (AWS)](https://aws.amazon.com/) | anchor | 2026-07-20 | 2026-07-22 | qualified | synergy-cloud-q1-2026 | operates the analyst-estimated largest cloud-infrastructure platform by first-quarter 2026 revenue share; AI-capacity share remains a separate, unmeasured figure |
| map-ai-placement-datacenters-google | [Google](https://cloud.google.com/) | anchor | 2026-07-20 | 2026-07-22 | verified | google-tpu | operates its own datacenters, network, and TPU fleet and sells TPU capacity through Google Cloud |
| map-ai-placement-datacenters-meta | [Meta](https://about.meta.com/) | supporting | 2026-07-20 | 2026-07-22 | verified | meta-infrastructure | operates large first-party datacenter and training fleets for its products and models |
| map-ai-placement-datacenters-oracle | [Oracle](https://www.oracle.com/cloud/) | supporting | 2026-07-22 | 2026-07-22 | qualified | oracle-stargate | is an OpenAI Stargate infrastructure partner with announced campuses under construction; energized capacity remains undisclosed |
| map-ai-placement-datacenters-coreweave | [CoreWeave](https://www.coreweave.com/) | context | 2026-07-20 | 2026-08-12 | qualified | coreweave-10k, coreweave-q2-2026 | runs a specialist accelerated cloud whose company-reported Q2 2026 disclosure described roughly $104 billion of backlog and 3.7 GW of contracted power alongside a net loss |
| map-ai-placement-datacenters-spacexai | [SpaceXAI](https://x.ai/) | context | 2026-07-22 | 2026-07-22 | qualified | xai-series-e | reported a company-defined H100-equivalent fleet figure without an independent audit in the cited disclosure |

### L4 · Models

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-models-openai | [OpenAI](https://openai.com/) | anchor | 2026-07-20 | 2026-07-22 | verified | openai-gpt54 | develops and distributes frontier models through its API and first-party products |
| map-ai-placement-models-google-deepmind | [Google DeepMind](https://deepmind.google/) | anchor | 2026-07-20 | 2026-07-22 | verified | google-io-2026 | develops Gemini inside a parent that also operates TPUs, cloud infrastructure, and first-party surfaces |
| map-ai-placement-models-anthropic | [Anthropic](https://www.anthropic.com/claude) | anchor | 2026-07-20 | 2026-07-22 | verified | anthropic-models | develops Claude models and distributes them through its own API and products as well as cloud partners |
| map-ai-placement-models-spacexai | [SpaceXAI](https://x.ai/) | supporting | 2026-07-22 | 2026-07-23 | verified | xai-models, spacexai-joins, spacexai-grok | is the current company identity for the lab that ships Grok models and APIs; frontier leadership remains unmeasured by product availability alone |
| map-ai-placement-models-meta-llama | [Meta Superintelligence Labs (Muse · Llama)](https://ai.meta.com/) | supporting | 2026-07-20 | 2026-07-23 | qualified | meta-llama, meta-muse | develops Muse inside Meta Superintelligence Labs and retains the Llama release lineage; Llama download totals are company-reported, while active deployments remain unmeasured |
| map-ai-placement-models-alibaba-qwen | [Alibaba (Qwen)](https://qwen.ai/) | supporting | 2026-07-20 | 2026-07-23 | qualified | alibaba-qwen, huggingface-open-ecosystem-2026 | distributes Qwen weights, APIs, and an app; its download and user totals are company-reported and measure different things |
| map-ai-placement-models-deepseek | [DeepSeek](https://www.deepseek.com/en/) | supporting | 2026-07-20 | 2026-07-23 | verified | deepseek-api, huggingface-open-ecosystem-2026 | publishes model weights and an API; product availability alone leaves frontier leadership unmeasured |
| map-ai-placement-models-moonshot-kimi | [Moonshot AI (Kimi)](https://www.moonshot.ai/) | supporting | 2026-07-23 | 2026-07-23 | qualified | moonshot-kimi, artificial-analysis-models | develops Kimi models and first-party products; a July 2026 independent benchmark snapshot provides a dated capability signal and leaves durable lab rank open |
| map-ai-placement-models-zai-glm | [Z.ai (GLM)](https://z.ai/company) | supporting | 2026-07-23 | 2026-07-23 | qualified | zai-glm, artificial-analysis-models | develops GLM models and first-party agent products; a July 2026 independent benchmark snapshot measures capability, while adoption remains unmeasured |
| map-ai-placement-models-mistral | [Mistral AI](https://mistral.ai/) | supporting | 2026-07-22 | 2026-07-23 | verified | mistral-models, mistral-vibe | ships API-accessible and release-specific open-weight model families and now names its first-party surface Vibe |
| map-ai-placement-models-cohere | [Cohere](https://cohere.com/command) | context | 2026-07-23 | 2026-07-23 | qualified | cohere-command | develops Command for enterprise and private deployment; its specialist inclusion leaves top-five general capability unclaimed |
| map-ai-placement-models-thinking-machines | [Thinking Machines Lab (Inkling)](https://thinkingmachines.ai/) | context | 2026-07-23 | 2026-07-23 | qualified | thinking-machines-inkling | trains the open-weights Inkling model family; inclusion recognizes a current model lab, with top-five capability and mass-consumer reach left unclaimed |
| map-ai-placement-models-amazon-nova | [Amazon Nova](https://aws.amazon.com/nova/) | context | 2026-07-22 | 2026-07-22 | verified | aws-nova | is Amazon's first-party model family alongside third-party models in Bedrock |
| map-ai-placement-models-microsoft-ai | [Microsoft AI](https://microsoft.ai/) | context | 2026-07-22 | 2026-07-22 | verified | microsoft-mai | develops first-party MAI models in addition to distributing partner models |
| map-ai-placement-models-nvidia-nemotron | [Nvidia Nemotron](https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/) | context | 2026-07-22 | 2026-07-22 | verified | nvidia-nemotron | publishes Nemotron model families as part of a broader hardware, software, and serving stack |
| map-ai-placement-models-hugging-face | [Hugging Face](https://huggingface.co/) | context | 2026-07-25 | 2026-07-22 | verified | hf-state-of-os-2026 | hosts the public hub through which most open-weight models and datasets are distributed; the claim covers distribution reach alone |
| map-ai-placement-models-mercor | [Mercor](https://mercor.com/) | supporting | 2026-07-26 | 2026-07-26 | qualified | mercor-techcrunch-2026-07-08, mercor-information-h1-2026 | brokers paid domain experts to the labs that need them; its CEO put the company past $2 billion in annualized gross revenue in July 2026, on $614 million of gross revenue in the first half |
| map-ai-placement-models-surge | [Surge AI](https://www.surgehq.ai/) | supporting | 2026-07-26 | 2026-07-26 | qualified | surge-inc-2025-06-20 | supplies human preference and RLHF data, and is reportedly larger by revenue than the incumbent it is usually compared to, while staying bootstrapped and small enough that no filing exists to check it against |
| map-ai-placement-models-common-crawl | [Common Crawl](https://commoncrawl.org/) | supporting | 2026-07-26 | 2026-07-26 | verified | common-crawl-corpus, common-crawl-skrenta-2025 | is the 501(c)(3) whose free corpus of over 300 billion pages, growing by 3 to 5 billion a month, sits under a large share of open pretraining; one family foundation has supplied most of its funding for fifteen years |
| map-ai-placement-models-innodata | [Innodata](https://innodata.com/) | supporting | 2026-07-26 | 2026-07-26 | verified | innodata-10k-fy2025, innodata-10q-q1-2026 | is the one publicly traded data-engineering supplier here, which means its concentration is checkable: a single customer was approximately 58 percent of its FY2025 revenues, up from 48 percent the year before |

### L5 · Inference & serving

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-inference-microsoft-azure | [Microsoft Azure](https://azure.microsoft.com/en-us/) | anchor | 2026-07-20 | 2026-07-22 | verified | azure-openai | distributes OpenAI and other models with enterprise identity, billing, and compliance features |
| map-ai-placement-inference-aws-bedrock | [AWS Bedrock](https://aws.amazon.com/bedrock/) | anchor | 2026-07-20 | 2026-07-22 | verified | aws-bedrock | provides access to Amazon and third-party models through AWS services and billing |
| map-ai-placement-inference-google-vertex | [Google Vertex AI](https://cloud.google.com/vertex-ai) | supporting | 2026-07-22 | 2026-07-22 | verified | google-model-garden | distributes Google and third-party models through Vertex AI and Model Garden |
| map-ai-placement-inference-together-ai | [Together AI](https://www.together.ai/) | supporting | 2026-07-20 | 2026-07-22 | verified | together-docs | provides inference and fine-tuning services for open and custom models |
| map-ai-placement-inference-fireworks-ai | [Fireworks AI](https://fireworks.ai/) | supporting | 2026-07-20 | 2026-07-22 | verified | fireworks-docs | provides hosted inference and model-adaptation services without implying one ownership model for every catalog entry |
| map-ai-placement-inference-groq | [Groq](https://groq.com/) | context | 2026-07-20 | 2026-07-22 | verified | groq-docs | markets an inference-focused processor and cloud service |
| map-ai-placement-inference-nvidia-nim | [Nvidia NIM](https://docs.nvidia.com/nim/) | supporting | 2026-07-22 | 2026-07-22 | verified | nvidia-nim | supplies packaged model-serving components and inference software as well as chips |

### L6 · Routing & gateways

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-routing-openrouter | [OpenRouter](https://openrouter.ai/) | anchor | 2026-07-20 | 2026-07-22 | qualified | openrouter-docs, openrouter-rankings | exposes many models through one API and publishes rankings drawn from its own traffic, a subset of the wider market |
| map-ai-placement-routing-cloudflare-ai-gateway | [Cloudflare (AI Gateway)](https://developers.cloudflare.com/ai-gateway/) | supporting | 2026-07-20 | 2026-07-22 | verified | cloudflare-gateway | adds observability, caching, rate controls, and provider routing around model calls |
| map-ai-placement-routing-vercel-ai-gateway | [Vercel (AI Gateway)](https://vercel.com/ai-gateway) | supporting | 2026-07-20 | 2026-07-22 | verified | vercel-gateway | integrates routing and failover with Vercel's deployment and AI SDK ecosystem |
| map-ai-placement-routing-litellm | [LiteLLM](https://www.litellm.ai/) | supporting | 2026-07-20 | 2026-07-22 | verified | litellm-docs | normalizes calls across providers through an open-source proxy and SDK; public docs establish breadth while leaving market-wide adoption unmeasured |
| map-ai-placement-routing-microsoft-foundry | [Microsoft Foundry](https://ai.azure.com/) | context | 2026-07-22 | 2026-07-22 | verified | microsoft-router | offers model-routing modes inside Microsoft's broader enterprise AI platform |
| map-ai-placement-routing-aws-bedrock | [AWS Bedrock routing](https://aws.amazon.com/bedrock/) | context | 2026-07-22 | 2026-07-22 | verified | aws-router | offers intelligent prompt routing inside Bedrock |
| map-ai-placement-routing-google-vertex | [Google Vertex AI](https://cloud.google.com/vertex-ai) | context | 2026-07-22 | 2026-07-22 | verified | google-model-garden | bundles model choice and routing into a larger cloud control plane |

### L7 · Agent control

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-harness-mcp | [Model Context Protocol](https://modelcontextprotocol.io/) | anchor | 2026-07-22 | 2026-07-22 | verified | lf-aaif | was introduced by Anthropic and contributed to the Linux Foundation's Agentic AI Foundation, now its neutral governance home |
| map-ai-placement-harness-openai-agents-sdk | [OpenAI Agents SDK](https://openai.github.io/openai-agents-python/) | anchor | 2026-07-20 | 2026-07-22 | verified | openai-agents | provides first-party agent primitives for tools, handoffs, guardrails, sessions, and tracing |
| map-ai-placement-harness-microsoft-agent-framework | [Microsoft Agent Framework](https://learn.microsoft.com/en-us/agent-framework/) | supporting | 2026-07-22 | 2026-07-22 | verified | microsoft-agent-framework | reached version 1.0 in April 2026 |
| map-ai-placement-harness-aws-agentcore | [AWS AgentCore](https://aws.amazon.com/bedrock/agentcore/) | supporting | 2026-07-22 | 2026-07-22 | verified | aws-agentcore | covers runtime, identity, gateways, memory, observability, and evaluation within AWS |
| map-ai-placement-harness-google-adk-a2a | [Google ADK + A2A](https://google.github.io/adk-docs/) | supporting | 2026-07-22 | 2026-07-22 | verified | google-adk, lf-a2a | combines a development kit with support for A2A under Linux Foundation governance |
| map-ai-placement-harness-nvidia-agent-toolkit | [Nvidia Agent Toolkit](https://docs.nvidia.com/nemo/agent-toolkit/latest/) | supporting | 2026-07-22 | 2026-07-22 | verified | nvidia-agent | ships agent tooling tied to Nvidia's broader inference and enterprise stack |
| map-ai-placement-harness-langgraph | [LangGraph](https://www.langchain.com/langgraph) | supporting | 2026-07-20 | 2026-07-22 | verified | langgraph | provides stateful graph-based orchestration without this map assigning it market leadership |
| map-ai-placement-harness-braintrust | [Braintrust](https://www.braintrust.dev/) | context | 2026-07-20 | 2026-07-22 | verified | braintrust | sells evaluation, tracing, and experimentation for AI applications |
| map-ai-placement-harness-mozilla-ai | [Mozilla.ai](https://www.mozilla.ai/) | context | 2026-07-20 | 2026-07-22 | qualified | mozilla-control | contributed the control-layer framing; the citation supports the name alone and carries no market-position evidence |

### L8 · Surfaces

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-surfaces-openai-chatgpt | [OpenAI (ChatGPT)](https://chatgpt.com/) | anchor | 2026-07-20 | 2026-08-12 | qualified | openai-chatgpt-users-2026-08 | reported one billion weekly active users in August 2026; the company-reported weekly figure uses a different period from monthly metrics |
| map-ai-placement-surfaces-anthropic-claude | [Claude (Anthropic)](https://claude.ai/) | anchor | 2026-07-22 | 2026-07-22 | qualified | anthropic-small-business | ships first-party chat, coding, and work surfaces, with no comparable total audience figure located by the cutoff |
| map-ai-placement-surfaces-google-gemini | [Google (Gemini)](https://gemini.google.com/) | anchor | 2026-07-20 | 2026-08-12 | qualified | alphabet-q2-2026 | has a company-reported monthly Gemini app user figure of 950 million for Q2 2026; weekly use follows a different measurement period |
| map-ai-placement-surfaces-spacexai-grok | [Grok (SpaceXAI)](https://grok.com/) | anchor | 2026-07-23 | 2026-07-23 | qualified | spacexai-grok | is SpaceXAI's first-party assistant across web and other products; no comparable public audience total was located |
| map-ai-placement-surfaces-moonshot-kimi | [Kimi (Moonshot AI)](https://www.kimi.com/) | supporting | 2026-07-23 | 2026-07-23 | qualified | kimi-surface, moonshot-kimi | is Moonshot AI's first-party assistant and work and coding surface; comparable reach remains unmeasured |
| map-ai-placement-surfaces-alibaba-qwen | [Qwen](https://qwen.ai/) | supporting | 2026-07-22 | 2026-07-22 | qualified | alibaba-qwen | has a company-reported monthly app user figure above 300 million across platforms |
| map-ai-placement-surfaces-deepseek | [DeepSeek](https://chat.deepseek.com/) | supporting | 2026-07-23 | 2026-07-23 | qualified | deepseek-surface, deepseek-api | offers a first-party chat surface in addition to APIs and model weights; user count remains unmeasured by product availability |
| map-ai-placement-surfaces-meta-ai | [Meta AI](https://www.meta.ai/) | supporting | 2026-07-22 | 2026-07-22 | qualified | meta-ai-users | has company-reported monthly reach above one billion across its products; embedded distribution differs from a standalone assistant |
| map-ai-placement-surfaces-mistral-vibe | [Vibe (Mistral AI)](https://chat.mistral.ai/) | context | 2026-07-23 | 2026-07-23 | qualified | mistral-vibe | is Mistral's public Work, Code, and Chat surface, renamed from Le Chat in June 2026; no comparable reach figure was located |
| map-ai-placement-surfaces-microsoft-copilot | [Microsoft (Copilot)](https://copilot.microsoft.com/) | context | 2026-07-20 | 2026-07-22 | verified | microsoft-copilot | commercializes Microsoft 365 Copilot as a per-user enterprise subscription |
| map-ai-placement-surfaces-cursor | [Cursor](https://www.cursor.com/) | context | 2026-07-20 | 2026-07-22 | verified | cursor-vscode | uses the VS Code codebase and reorganizes the editor around AI-assisted and agentic development |
| map-ai-placement-surfaces-perplexity | [Perplexity](https://www.perplexity.ai/) | context | 2026-07-20 | 2026-07-22 | verified | perplexity-products | builds answer-oriented search and research products without relying on the draft's unsupported attributed quotation |

### The sea · Governance & the commons

| Placement ID | Actor and official destination | Editorial salience | Added | Reviewed | Evidence state | Receipt refs | Why it matters here |
|---|---|---|---|---|---|---|---|
| map-ai-placement-governance-us-government | [US government](https://www.bis.gov/) | anchor | 2026-07-20 | 2026-07-22 | verified | bis-2026, eo-14409 | uses semiconductor export controls while EO 14409 invites frontier developers into an opt-in early-access review |
| map-ai-placement-governance-eu | [European Union](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai) | anchor | 2026-07-20 | 2026-08-12 | verified | eu-gpai, eu-ai-act, eu-ai-act-enforcement-2026 | applied general-purpose AI obligations from August 2025 and began enforcing the Act with new transparency obligations from August 2, 2026, subject to exceptions and transitions |
| map-ai-placement-governance-aaif | [Agentic AI Foundation](https://aaif.io/) | supporting | 2026-07-22 | 2026-07-22 | verified | lf-aaif | provides neutral governance for MCP and other donated agentic projects |
| map-ai-placement-governance-open-source-commons | [The open-source commons](https://opensource.org/) | supporting | 2026-07-20 | 2026-08-12 | verified | osi-definition, osi-open-source-ai, osi-open-weights | uses defined licensing criteria; open weights may retain use or distribution restrictions that fall outside the open-source definition |
| map-ai-placement-governance-chinese-labs | [Chinese labs and firms](https://qwen.ai/) | supporting | 2026-07-22 | 2026-07-22 | qualified | alibaba-qwen, deepseek-api | are material model producers; direct evidence of a single coordinated program remains absent |

## L1 · Electricity

<!-- block:map-ai-block-section-energy-paragraph-01 -->
Source: [U.S. Department of Energy, “Clean Energy Resources to Meet Data Center Electricity Demand”](https://www.energy.gov/oe/clean-energy-resources-meet-data-center-electricity-demand).

<!-- block:map-ai-block-section-energy-paragraph-02 -->
Now, in response to transformations in technologies like artificial intelligence (AI), data center expansion, new domestic manufacturing, and electrification in different sectors, the United States is returning to a period of rising electricity demand, with total energy demand potentially growing ~15-20% in the next decade.

<!-- block:map-ai-block-section-energy-paragraph-03 -->
Data center deployment, partly driven by the need to power new AI applications, is a significant factor of near-term electricity demand growth. The Electric Power Research Institute (EPRI) estimates that data centers could grow to consume up to 9% of U.S. electricity generation annually by 2030, up from 4% of total load in 2023. At a national level, data centers are critical to supporting America’s economic growth by powering businesses and enabling continued leadership in innovation, including for AI applications.

<!-- block:map-ai-block-section-energy-paragraph-04 -->
Data center electricity demand has specific characteristics. It is growing rapidly and varies regionally. Data centers can impact regional grids given the steep increases in load size, may be geographically constrained due to latency requirements, and often require firm power sources to operate continuously. Projections of data center electricity demand growth continue to evolve due to developing use cases and demand for AI and the speed of improvements in energy efficiency.

<!-- block:map-ai-block-section-energy-paragraph-05 -->
A broad suite of tools can be used to meet and manage rising electricity demand and lower overall peak demand. Approaches that span the whole power system include deploying clean generation and storage technologies; leveraging existing nuclear and hydropower infrastructure; redeveloping retired coal power plant sites; enhancing and expanding grid infrastructure; and maximizing energy efficiency and demand resources. Complementary non-technology-based solutions for managing demand growth include proactive planning, innovative tariff structures, optimizing grid performance, adopting alternative financing structures to fund new energy projects, and supply chain and workforce development. Pursuing key enablers like interconnection and regulatory reforms can unlock the barriers to adopting these energy solutions.

<!-- block:map-ai-block-section-energy-paragraph-06 -->
Energy efficiency is a key tool in reducing energy consumption from data center facilities. DOE has long been a leader in developing improved cooling technologies, including for data centers. DOE national labs have built exascale computing facilities with a Power Usage Efficiency (PUE) of 1.03, demonstrating state of the art techniques for data center efficiency. DOE is also leading the Energy Efficiency Scaling for 2 Decades initiative, with a goal to increase the energy efficiency of the microelectronics that are needed for computation at data centers by a factor of 1000 over 2 decades.

## L2 · Chips and fabs

<!-- block:map-ai-block-section-chips-paragraph-01 -->
Source: [National Institute of Standards and Technology, “Why NIST Is Putting Its CHIPS Into U.S. Manufacturing”](https://www.nist.gov/blogs/taking-measure/why-nist-putting-its-chips-us-manufacturing).

<!-- block:map-ai-block-section-chips-paragraph-02 -->
Silicon is the most frequently used raw material for chips, and one of the most abundant atomic elements on Earth. To give you a sense of its abundance, silicon and oxygen are the main ingredients of most beach sand, and a major component of glass, rocks and soil.

<!-- block:map-ai-block-section-chips-paragraph-03 -->
Silicon is a type of material known as a semiconductor. Electricity flows through semiconductors better than it does through insulators, but not quite as well as it does through conductors.

<!-- block:map-ai-block-section-chips-paragraph-04 -->
But that’s a good thing. In semiconductors, you can control electric current precisely, and without any moving parts. By applying a small voltage to them, you can either cause current to flow or to stop, making the semiconductor, or a small region within it, act like a conductor or insulator depending on what you want to do.

<!-- block:map-ai-block-section-chips-paragraph-05 -->
The first step for making a chip is to start with a thin slice of a semiconductor material, known as a “wafer,” often round in shape. On top of the wafer, manufacturers then create complex miniature electric circuits, commonly called “integrated circuits” because they are embedded as one piece on the wafer. A typical integrated circuit today contains billions of tiny on-off switches known as transistors that enable a chip to perform a wide range of complex tasks from sending signals to processing information. Increasingly, these circuits also have “photonic” components in which light travels alongside electricity.

<!-- block:map-ai-block-section-chips-paragraph-06 -->
Manufacturers typically mass-produce dozens of integrated circuits on a single semiconductor wafer and then dice the wafer to separate the individual pieces. When each of them is packaged as a self-contained device, you have a “chip,” which can then be placed in smartphones, computers and so many other products.

<!-- block:map-ai-block-section-chips-paragraph-07 -->
When we talk about chip packaging, we’re referring to everything that goes around a chip to protect it from damage and connect it to the rest of the device. Advanced packaging takes things to the next level: It uses ingenious techniques during the chipmaking process to connect multiple chips to each other and the rest of the device in as tiny a space as possible.

<!-- block:map-ai-block-section-chips-paragraph-08 -->
Advanced packaging enables our devices to be faster and more energy-efficient because information can be exchanged between chips over shorter distances and this in turn reduces energy consumption.

<!-- block:map-ai-block-section-chips-paragraph-09 -->
Measurement science plays a key role in up to 50% of semiconductor manufacturing steps, according to a NIST report. Good measurements enable manufacturers to mass-produce high-quality, high-performance chips.

## L3 · Datacenters

<!-- block:map-ai-block-section-datacenters-paragraph-01 -->
Source: [U.S. Department of Energy, “DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers”](https://www.energy.gov/articles/doe-releases-new-report-evaluating-increase-electricity-demand-data-centers).

<!-- block:map-ai-block-section-datacenters-paragraph-02 -->
The U.S. Department of Energy announced the publication of the 2024 Report on U.S. Data Center Energy Use produced by Lawrence Berkeley National Laboratory, which outlines the energy use of data centers from 2014 to 2028. The report estimates that data center load growth has tripled over the past decade and is projected to double or triple by 2028. U.S. electricity demand is projected to account for data center expansion and the rise of artificial intelligence applications, domestic manufacturing growth, and electrification of different industries. The Department continues to develop advanced technologies and use its resources to meet rising electricity demand in the United States while maintaining a reliable, affordable, and secure national energy system.

<!-- block:map-ai-block-section-datacenters-paragraph-03 -->
The report finds that data centers consumed about 4.4% of total U.S. electricity in 2023 and are expected to consume approximately 6.7 to 12% of total U.S. electricity by 2028. The report indicates that total data center electricity usage climbed from 58 TWh in 2014 to 176 TWh in 2023 and estimates an increase between 325 to 580 TWh by 2028.

<!-- block:map-ai-block-section-datacenters-paragraph-04 -->
DOE resources span the entire power system, from new generation and storage technologies to enhancing and expanding the transmission system to maximizing efficiency and flexibility of demand resources. DOE’s key strategies for meeting data center energy demand include enabling data center flexibility through onsite power generation and storage solutions, using energy community opportunities to re-use infrastructure at retired coal facilities for data centers and associated power infrastructure, engaging with stakeholders on innovative rate structures, and commercializing next-generation geothermal, advanced nuclear, long-duration storage, and efficient semiconductor technologies.

<!-- block:map-ai-block-section-datacenters-paragraph-05 -->
Data center electricity demand is growing rapidly and varies regionally. Data centers can impact regional grids given the steep increases in load size, may be geographically constrained due to latency requirements, and often require firm power sources to operate continuously. Projections of data center electricity demand growth continue to evolve due to developing use cases and demand for AI and the speed of improvements in energy efficiency.

<!-- block:map-ai-block-section-datacenters-paragraph-06 -->
Energy efficiency is a key tool in reducing energy consumption from data center facilities. DOE national labs have built exascale computing facilities with a Power Usage Efficiency of 1.03, demonstrating state of the art techniques for data center efficiency.

## L4 · Models

<!-- block:map-ai-block-section-models-paragraph-01 -->
Source: [National Institute of Standards and Technology, *Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile*](https://doi.org/10.6028/NIST.AI.600-1).

<!-- block:map-ai-block-section-models-paragraph-02 -->
This document is a cross-sectoral profile of and companion resource for the AI Risk Management Framework (AI RMF 1.0) for Generative AI, pursuant to President Biden’s Executive Order (EO) 14110 on Safe, Secure, and Trustworthy Artificial Intelligence. The AI RMF was released in January 2023, and is intended for voluntary use and to improve the ability of organizations to incorporate trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems.

<!-- block:map-ai-block-section-models-paragraph-03 -->
A profile is an implementation of the AI RMF functions, categories, and subcategories for a specific setting, application, or technology, in this case, Generative AI (GAI), based on the requirements, risk tolerance, and resources of the Framework user. AI RMF profiles assist organizations in deciding how to best manage AI risks in a manner that is well-aligned with their goals, considers legal/regulatory requirements and best practices, and reflects risk management priorities. Consistent with other AI RMF profiles, this profile offers insights into how risk can be managed across various stages of the AI lifecycle and for GAI as a technology.

<!-- block:map-ai-block-section-models-paragraph-04 -->
As GAI covers risks of models or applications that can be used across use cases or sectors, this document is an AI RMF cross-sectoral profile. Cross-sectoral profiles can be used to govern, map, measure, and manage risks associated with activities or business processes common across sectors, such as the use of large language models (LLMs), cloud-based services, or acquisition.

<!-- block:map-ai-block-section-models-paragraph-05 -->
This document defines risks that are novel to or exacerbated by the use of GAI. After introducing and describing these risks, the document provides a set of suggested actions to help organizations govern, map, measure, and manage these risks.

<!-- block:map-ai-block-section-models-paragraph-06 -->
EO 14110 defines Generative AI as “the class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content. This can include images, videos, audio, text, and other digital content.” While not all GAI is derived from foundation models, for purposes of this document, GAI generally refers to generative foundation models. The foundation model subcategory of “dual-use foundation models” is defined by EO 14110 as “an AI model that is trained on broad data; generally uses self-supervision; contains at least tens of billions of parameters; is applicable across a wide range of contexts.”

## L5 · Inference

<!-- block:map-ai-block-section-inference-paragraph-01 -->
Source: [National Institute of Standards and Technology, *NIST SP 800-239: Security and Trust Considerations for AI Data Centers*](https://doi.org/10.6028/NIST.SP.800-239.ipd).

<!-- block:map-ai-block-section-inference-paragraph-02 -->
An AI data center is specifically optimized for the training and inference of AI models. This optimization is evident in its hardware infrastructure, software stacks, workflows, and data storage systems.

<!-- block:map-ai-block-section-inference-paragraph-03 -->
The Computing Zone provides key services for preprocessing the training dataset, training AI models, and providing inference and other AI services. Compute nodes in an AI data center are often equipped with accelerators to accelerate AI model training and inference. These are often used in conjunction with High-Bandwidth Memory (HBM) for local data caching, key-value (KV) caching, model parameter caching, and related operations. High-speed networks are used for both scale-up and scale-out to build a unified AI supercomputing system. For instance, NVLink can connect multiple accelerators within the same rack to scale up computing, while an InfiniBand-type high-speed network connects computing nodes across different racks to scale out computing resources. This software stack, including AI libraries, AI software development kits, and tools, enables efficient AI model training and inference.

<!-- block:map-ai-block-section-inference-paragraph-04 -->
The Data Storage Zone provides a pivotal service for storing and accessing data since AI workloads consume and generate tremendous amounts of data. Like an HPC system, an AI data center storage system is multi-tiered to meet the extreme performance, capacity, and latency requirements of modern AI training and inference.

<!-- block:map-ai-block-section-inference-paragraph-05 -->
Data storage systems typically consist of local disks, block storage, object storage, and/or file storage. Databases and high-performance parallel file systems run on top of them to provide fast data retrieval services. In addition to structured data, AI model training also consumes large amounts of unstructured data, such as video, audio, and PDF files. Cost-effective object-based storage is often adopted to store vast amounts of raw data in its native format.

<!-- block:map-ai-block-section-inference-paragraph-06 -->
An AI data center storage system often acquires data from external storage systems for training and inference purposes and may share AI models and tokens with external parties. For example, retrieval-augmented generation (RAG) is an AI inference framework that improves the accuracy of large language models (LLMs) by retrieving data from external sources, such as company documents and live databases, in real time before generating a response. In the context of agentic AI, data storage systems may also retrieve and share data with external systems.

## L6 · Routing

<!-- block:map-ai-block-section-routing-paragraph-01 -->
Source: [National Institute of Standards and Technology, *NIST SP 800-239: Security and Trust Considerations for AI Data Centers*](https://doi.org/10.6028/NIST.SP.800-239.ipd).

<!-- block:map-ai-block-section-routing-paragraph-02 -->
The Access Zone in an AI data center has significantly greater responsibilities than in an HPC system. In addition to the services provided by an HPC system, the Access Zone in an AI data center hosts an AI gateway to handle incoming AI inference and application requests. The AI gateway routes those requests to the appropriate servers, which promptly generate responses. Monitoring and logging user requests and AI-generated responses are often required for security and auditing purposes. The Access Zone provides an interface for data storage systems to retrieve external datasets and/or share internal datasets. All of these functions expose the Access Zone to greater threats and risks.

<!-- block:map-ai-block-section-routing-paragraph-03 -->
The Management Zone in an AI data center operates similarly to that of an HPC cluster. However, the Management Zone in an AI data center has greater responsibility for monitoring, logging, anomaly detection, and compliance checks. This is due to more frequent and diverse user access, increased data acquisition and sharing, more capable AI models, and heightened regulatory and compliance requirements for data and AI models. As a result, extensive monitoring, logging, and compliance checks become necessary. These functions are typically managed within the Security Operations Center (SOC), which is essential for protecting high-value assets, monitoring unique AI-related attack surfaces, and ensuring regulatory compliance. Overall, the Management Zone plays a critical role in maintaining the security posture of an AI data center.

<!-- block:map-ai-block-section-routing-paragraph-04 -->
While the composition and services offered by these zones resemble those found in an HPC system, AI data centers are specifically designed to support the unique requirements of AI workloads.

### L6 routing, worked

<!-- block:map-ai-block-routing-sim-frame -->
A router compares capability, latency, price, availability, and policy for each task. The deciding factor changes with the work and with current conditions. Route A: Low latency, text input, standard context. Route B: Long context and stronger planning, with a higher call cost. Route C: Image input, strict schemas, and approved customer-document handling. The routes and cases are fictional examples of the decision process.

<!-- block:map-ai-block-routing-sim-cases -->
Pull action items from a non-confidential meeting transcript while the attendee waits. Routed to Route A; latency decides this case. Route A returns the list within the response-time budget and has enough extraction accuracy for a result the attendee can check against the transcript. Under the hood: This prompt repeats often, so cache state can change first-token latency. A route switch may start with a cold cache. Draft a migration plan from a repository and incident notes for a service with sparse documentation. Routed to Route B; price decides this case. Route B costs more per call and produces a reviewable plan in one pass. This job runs once, and an engineer reads the result tomorrow, so repair time carries the most weight. Under the hood: The routing budget includes the engineer's review time. A thin plan can make a cheap call expensive to finish. Turn scanned customer invoices into a fixed JSON schema during an approved overnight batch. Routed to Route C; capability decides this case. Route C accepts images, supports the required schema, and is approved for customer documents. Those requirements determine the destination before the batch starts. Under the hood: Schema adapters and tokenizers vary across model families. The router must select the matching adapter with the destination. Summarise a customer document covered by a data-residency agreement. Routed to Route B; policy decides this case. The customer's contract names the allowed processing region. That clause narrows the pool first, and the router chooses among the approved destinations. Under the hood: Availability changes throughout the day. Region and contract filters apply before the router checks live capacity.

## L7 · Agent control

<!-- block:map-ai-block-section-harness-paragraph-01 -->
Source: [National Institute of Standards and Technology, “Announcing the AI Agent Standards Initiative for Interoperable and Secure Innovation”](https://www.nist.gov/news-events/news/2026/02/announcing-ai-agent-standards-initiative-interoperable-and-secure).

<!-- block:map-ai-block-section-harness-paragraph-02 -->
The Center for AI Standards and Innovation (CAISI) at NIST announced the launch of the AI Agent Standards Initiative. The Initiative will ensure that the next generation of AI, AI agents capable of autonomous actions, is widely adopted with confidence, can function securely on behalf of its users, and can interoperate smoothly across the digital ecosystem. Working in coordination with other federal partners, including the Information Technology Laboratory (ITL) at NIST, CAISI aims to foster the emerging ecosystem of industry-led AI standards and protocols.

<!-- block:map-ai-block-section-harness-paragraph-03 -->
AI agents can now work autonomously for hours, write and debug code, manage emails and calendars, and shop for goods, among other emerging use cases. While the productivity promise is enticing, the real-world utility of agents is constrained by their ability to interact with external systems and internal data. Absent confidence in the reliability of AI agents and interoperability among agents and digital resources, innovators may face a fragmented ecosystem and stunted adoption. To address this concern, NIST, including CAISI, aims to foster industry-led technical standards and protocols that build public trust in AI agents, catalyze an interoperable agent ecosystem, and diffuse their benefits to all Americans and across the world.

<!-- block:map-ai-block-section-harness-paragraph-04 -->
CAISI, with ITL at NIST, will collaborate with the National Science Foundation and other interagency partners to advance the Initiative along three pillars:

<!-- block:map-ai-block-section-harness-paragraph-05 -->
1. Facilitating industry-led development of agent standards and U.S. leadership in international standards bodies. 2. Fostering community-led open source protocol development and maintenance for agents. 3. Advancing research in areas of AI agent security and identity to enable new use cases and to promote trusted adoption across sectors of the economy.

<!-- block:map-ai-block-section-harness-paragraph-06 -->
NIST will announce research, guidelines, and further deliverables for the AI Agent Standards Initiative. To support the interoperable and secure adoption of AI agents, NIST will leverage a full toolbox for public input, including convenings, requests for information, listening sessions, and other approaches.

## L8 · Applications

<!-- block:map-ai-block-section-surfaces-paragraph-01 -->
Source: [Ram D. Sriram, National Institute of Standards and Technology, “AI in the Doctor’s Office: How Standards Can Support Trustworthiness”](https://www.nist.gov/blogs/taking-measure/ai-doctors-office-how-standards-can-support-trustworthiness).

<!-- block:map-ai-block-section-surfaces-paragraph-02 -->
When you go to a medical appointment, does the doctor look at you while you talk? Or are they busy typing everything you say into a computer? If it’s the latter, you may find it will change soon, thanks to artificial intelligence (AI).

<!-- block:map-ai-block-section-surfaces-paragraph-03 -->
Some doctors’ offices are using AI transcription services to transcribe your discussion with the doctor and automatically enter the results into your electronic medical records.

<!-- block:map-ai-block-section-surfaces-paragraph-04 -->
That’s a time-saver for doctors, who often spend hours filling out their patients’ records. It also allows them to look at the patient rather than their computer screen.

<!-- block:map-ai-block-section-surfaces-paragraph-05 -->
You may have also noticed AI chatbots asking if they can help you when visiting a company’s website. These chatbots are not as common in health care yet, but it’s possible they could assist you with basic medical questions in the future. This could free up the doctor’s time for more complex concerns.

<!-- block:map-ai-block-section-surfaces-paragraph-06 -->
These are just two ways AI may impact your future health care. But given the high stakes, it must be done with thoughtful standards.

<!-- block:map-ai-block-section-surfaces-paragraph-07 -->
If AI will work in the medical field, or any other field it's used in, we need to develop specific and useful standards. These will need to include characteristics that can be used to judge an AI model on its reliability and trustworthiness.

<!-- block:map-ai-block-section-surfaces-paragraph-08 -->
One way AI can prove its trustworthiness is by demonstrating its correctness. If you’ve ever had a generative AI tool confidently give you the wrong answer to a question, you probably appreciate why this is important. If an AI tool says a patient has cancer, the doctor and patient need to know the odds that the AI is right or wrong.

<!-- block:map-ai-block-section-surfaces-paragraph-09 -->
Another issue is reliability, particularly of the datasets AI tools rely on for information. Just as a hacker can inject a virus into a computer network, someone could intentionally infect an AI dataset to make it work nefariously. In many AI systems, which use large datasets to learn, people can introduce Trojans, similar to computer viruses. This can alter the AI system’s reasoning. This can be done at the level of the input (dataset), the model (the thinking) or the AI’s environment and how it interacts with the world.

<!-- block:map-ai-block-section-surfaces-paragraph-10 -->
For example, researchers introduced a Trojan by placing a sticker on a stop sign. This Trojan made the self-driving car run through a stop sign because it thought it was a speed limit sign. So, there are dangers we’ll have to face if AI is unreliable. My NIST colleagues are doing considerable work to help detect Trojans, which I hope will make AI more reliable.

## Rebuild the stack

<!-- block:map-ai-block-layer-reconstruction -->
Eight layers, shuffled. Put them in dependency order, with the foundation first and the surface a person uses last. Your order disappears when you leave. There is no score, and you can reveal the answer whenever you want.

**Show me the order.** The order, atoms to bits:

1. Energy & grid: a transistor switches nothing without electricity, and whether that electricity can be bought at all is settled years earlier, in interconnection queues and turbine order books.
2. Chips & fabs: everything above it is a program, and a program has to run on a transistor somewhere.
3. Datacenters: a chip in a box does nothing until a building gives it power, cooling, and a network.
4. Models: weights are what a long training run leaves behind, and that run is months of those buildings and that hardware.
5. Inference & serving: weights sit inert until a serving system loads them and turns them back into tokens.
6. Routing & gateways: routing begins once several models are available to receive a request.
7. Agent control: deciding what happens next presumes a model call you can already make.
8. Surfaces: this is where finished work reaches a person, so it rests on all seven beneath it.

## Research links

As of 2026-08-12 · newest first

I kept the most useful research links below. The claims above use their own sources.

<!-- block:map-ai-block-research-link-01 -->
- [European Commission: AI Act enforcement from August 2, 2026](https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august) · 2026-07-31 · regulator · enforcement and transparency dates, with exceptions and transitions
<!-- block:map-ai-block-research-link-02 -->
- [FTC: large AI partnerships and investments](https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2025/01/behind-ftcs-6b-report-large-ai-partnerships-investments) · 2025-01-17 · regulator study · compute access, switching costs, talent, and information access
<!-- block:map-ai-block-research-link-03 -->
- [Berkeley Lab: U.S. Data Center Energy Usage Report](https://eta.lbl.gov/publications/united-states-data-center-energy-2025) · 2025 update · independent report · U.S. electricity and water scenarios, not a universal forecast
<!-- block:map-ai-block-research-link-04 -->
- [TSMC: CoWoS advanced packaging](https://3dfabric.tsmc.com/english/dedicatedFoundry/technology/cowos.htm) · accessed 2026-08-12 · company documentation · logic and HBM integration; not comparative capacity evidence
<!-- block:map-ai-block-research-link-05 -->
- [NIST: AI Agent Standards Initiative](https://www.nist.gov/artificial-intelligence/ai-agent-standards-initiative) · 2026 · standards · interoperability and security work, not adoption evidence
<!-- block:map-ai-block-research-link-06 -->
- [OWASP: AI Agent Security Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/AI_Agent_Security_Cheat_Sheet.html) · accessed 2026-08-12 · security guidance · agent-specific risks and controls
<!-- block:map-ai-block-research-link-07 -->
- [OSI: Open Source AI Definition 1.0](https://opensource.org/ai/open-source-ai-definition) · accessed 2026-08-12 · definition · distinguishes system freedoms from access to weights
<!-- block:map-ai-block-research-link-08 -->
- [MLCommons: Endpoints benchmark](https://mlcommons.org/benchmarks/endpoints/) · accessed 2026-08-12 · benchmark method · throughput, interactivity, latency, and concurrency
<!-- block:map-ai-block-research-link-09 -->
- [European Commission: Digital Markets Act](https://digital-markets-act.ec.europa.eu/index_en) · accessed 2026-08-12 · distribution governance · gatekeeper framing for core platform services
<!-- block:map-ai-block-research-link-10 -->
- [Ramp Router](https://ramp.com/router) · accessed 2026-07-26 · correction · Ramp was placed in L6 routing on this map and should not have been: a company routing its own traffic is not a peer of the gateway vendors beside it. It is opening the router by request, which is worth watching and is why the link stays
<!-- block:map-ai-block-research-link-11 -->
- [Stratechery: Who's Afraid of Chinese Models?](https://stratechery.com/2026/whos-afraid-of-chinese-models/) · 2026-07-20 · analysis · its COGS versus R&D section is the framing behind the serving-cost argument; subscriber content
<!-- block:map-ai-block-research-link-12 -->
- [Artificial Analysis: Model leaderboard](https://artificialanalysis.ai/leaderboards/models) · accessed 2026-07-23 · independent benchmark snapshot · dated capability comparison; English text-only weighting, not adoption
<!-- block:map-ai-block-research-link-13 -->
- [Hugging Face: State of Open Source on the Hub, Spring 2026](https://huggingface.co/blog/huggingface/state-of-os-hf-spring-2026) · Spring 2026 · platform analysis · open-model distribution and ecosystem activity, not active users
<!-- block:map-ai-block-research-link-14 -->
- [NIST: additional TSMC U.S. investment](https://www.nist.gov/news-events/news/2026/07/trump-administration-secures-additional-100-billion-us-semiconductor) · 2026-07-16 · government announcement · planned U.S. investment raised to $265 billion
<!-- block:map-ai-block-research-link-15 -->
- [Narayanan and Kapur: Up the Stack](https://www.normaltech.ai/p/up-the-stack-how-ais-escape-from) · 2026-07-09 · essay · historical commodity-and-complements argument
<!-- block:map-ai-block-research-link-16 -->
- [Mozilla.ai: The Control Layer](https://blog.mozilla.ai/the-control-layer-why-the-next-era-of-ai-is-about-infrastructure-not-just-models/) · 2026-07-07 · essay · public argument for the control-layer framing
<!-- block:map-ai-block-research-link-17 -->
- [EO 14409: Promoting Advanced AI Innovation and Security](https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/) · 2026-06-02 · primary · opt-in early-access review for frontier models
<!-- block:map-ai-block-research-link-18 -->
- [Anthropic and Amazon compute agreement](https://www.anthropic.com/news/anthropic-amazon-compute) · 2026-04-20 · company announcement · contracted maxima and company-reported Trainium deployment
<!-- block:map-ai-block-research-link-19 -->
- [Linux Foundation: Agentic AI Foundation](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation) · 2025-12-09 · standards governance · MCP contribution and AAIF formation
<!-- block:map-ai-block-research-link-20 -->
- [Stanford HAI: AI Index Report 2026](https://hai.stanford.edu/ai-index/2026-ai-index-report/research-and-development) · 2026 · independent report · estimates of AI compute and semiconductor concentration
<!-- block:map-ai-block-research-link-21 -->
- [IEA: Electricity 2026, Grids](https://www.iea.org/reports/electricity-2026/grids) · 2026 · intergovernmental analysis · grid queues and connection constraints
<!-- block:map-ai-block-research-link-22 -->
- [CoreWeave 2025 Form 10-K](https://www.sec.gov/Archives/edgar/data/1769628/000176962826000104/crwv-20251231.htm) · 2026-03-02 · company filing · contracts, financing, risks, and useful lives
<!-- block:map-ai-block-research-link-23 -->
- [Epoch AI: How much does it cost to train frontier models?](https://epoch.ai/publications/how-much-does-it-cost-to-train-frontier-ai-models) · 2024-06-03 · independent estimate · training-run cost methodology and uncertainty
<!-- block:map-ai-block-research-link-24 -->
- [European Commission: General-purpose AI obligations](https://digital-strategy.ec.europa.eu/en/factpages/general-purpose-ai-obligations-under-ai-act) · current at cutoff · regulator guidance · applicable dates and obligations