Artificial intelligence is a technology, infrastructure, business and capital-markets ecosystem — from the chips and data centers that train and run models to the software, agents and public companies built on top of them.
This is SaGeminieTech’s map of that ecosystem: how the layers fit together, how the economics work, where the risks sit, and where to find deeper research on each part.
Overview
What SaGeminieTech’s Artificial Intelligence Vertical Covers
The Artificial Intelligence vertical treats AI as a connected system rather than a single product category. A model is only one layer: it depends on specialized chips, data-center capacity and data, and it creates value only when it is embedded in applications and workflows. Public and private companies, investors and regulators each interact with a different part of that system. This page organizes the territory at a broad level and points to specialist research for each part.
Models, training and inference
How modern AI systems are built, adapted and run — from foundation models to retrieval, inference and agents.
Chips, data centers and cloud
The semiconductors, accelerators, networking, cloud platforms and power that make large-scale AI possible.
Applications and economics
How organizations adopt AI, what it costs to build and operate, and where the revenue is earned.
Companies and risk
How public companies participate at different layers, and how disclosures and regulation shape the picture.
Informational architecture only: nothing on this page is investment advice, a ranking of companies or a forecast. Facts that are likely to change are shown in labelled “dated data” boxes with their source and review date; the rest of the page is written to stay valid as the field evolves.
Concepts
Understanding Artificial Intelligence
Three terms are often used as if they meant the same thing. They are nested: machine learning is one way of building AI, and generative AI is one kind of machine learning system.
Artificial intelligence
The OECD's widely used definition centers on inference: an AI system takes input and works out how to produce outputs — predictions, content, recommendations or decisions — that can influence physical or virtual environments, with differing degrees of autonomy and adaptiveness after deployment.
Source: OECD definition of an AI system (revised 2023).
Machine learning
Instead of hand-writing rules, developers train a model on data so that it learns statistical patterns, then apply it to new inputs. Much of modern AI is built this way, which is why data, compute and evaluation matter as much as algorithms.
Conceptual summary; see the foundation-model and scaling-law sources below.
Generative AI
NIST describes generative AI as models that emulate the structure and characteristics of their input data in order to generate derived synthetic content — for example text, images, audio or code.
Source: NIST AI 600-1, Generative AI Profile (July 2024).
Foundation models and modern AI systems
A foundation model, in the term's original definition, is a model trained on broad data at scale that can be adapted to a wide range of downstream tasks (Bommasani et al., Stanford CRFM, 2021). Many of today's large language models build on the Transformer architecture, introduced in 2017 as a design based solely on attention mechanisms that its authors found to be more parallelizable, and to need significantly less time to train, than the recurrent and convolutional designs they compared it with (Vaswani et al.).
Two research results help explain why demand for AI infrastructure has grown so quickly. Empirical scaling laws show that language-model loss falls predictably as model size, data and training compute increase (Kaplan et al., 2020). Later work found that, for a fixed compute budget, model size and training data should be scaled together rather than model size alone (Hoffmann et al., 2022).
What a model is
A trained set of parameters that maps inputs to outputs. On its own it has no access to your data, no memory of previous sessions and no ability to take action.
What an AI system adds
A production system wraps a model with data retrieval, tools, guardrails, monitoring and an interface. Retrieval-augmented generation, for example, combines a model's built-in knowledge with an external index so answers can draw on up-to-date sources (Lewis et al., 2020). Many of the practical questions about value and risk sit in this system layer.
Architecture
The AI Technology Stack
The layers below depend on each other from the bottom up: an application is only as capable, fast and affordable as the layers beneath it. The diagram is drawn with layer 1 at the base; on a screen reader the list reads in numbered order.
Read from the bottom up ↑
Semiconductors and compute
Chips built for the parallel matrix arithmetic that neural networks depend on: GPUs, custom accelerators such as Google's Tensor Processing Units, networking and memory silicon, and the manufacturing equipment behind them.
Data centers and cloud infrastructure
The facilities and cloud platforms that house accelerators, networking, storage and power — and rent that capacity as a service. Electricity is a binding input: see the dated IEA data in the next section.
Foundation models
Large models trained on broad data that can be adapted to many tasks. Some developers keep their model weights proprietary; others release open models, and some do both.
Data and retrieval
The pipelines, databases and search indexes that let a system use an organization's own information at the moment of a request. Retrieval-augmented generation is the best-known pattern (Lewis et al.); vector and embedding stores are part of the security surface (OWASP LLM08).
Inference
Running a trained model to produce outputs from new inputs — the step that happens every time an AI feature is used. MLCommons benchmarks it separately from training because it measures how fast systems process inputs using a trained model (MLPerf Inference).
Applications
Software products that put models to work for a user or a business process: assistants, coding tools, customer-service systems, analytics and industry-specific tools.
Agents and automation
Systems that use a model to plan, call tools, observe results and continue across multiple steps toward a goal — extending applications from answering to acting. See the dedicated section below.
Real companies span several layers, and layer boundaries move over time; the stack is a way to ask where a technology or company creates and captures value, not a fixed classification.
Build and run
AI Models and Infrastructure
Modern AI is limited as much by physical infrastructure as by algorithms. This section covers the concepts; the economics follow in the next section.
Training versus inference
| Aspect | Training | Inference |
|---|---|---|
| What it does | Builds or adapts a model by adjusting its parameters on data. | Uses a trained model to produce outputs from new inputs (MLCommons). |
| When it happens | During development, and again when a model is updated or fine-tuned. | Every time the model is used, so total work grows with usage. |
| Typical resource pattern | Large accelerator clusters running a job for an extended period. | Serving capacity sized for demand, judged on speed (latency and throughput) as well as cost. |
| Cost pattern | Large, front-loaded spend per model version. | Recurring operating cost that scales with requests. |
| Standard benchmarks | MLPerf Training (MLCommons). | MLPerf Inference (MLCommons). |
Compute requirements
Performance improves with model size, data and compute (Kaplan et al.), and compute-optimal training scales data along with parameters (Hoffmann et al.); together these results help explain why demand for accelerators and data-center capacity has grown. Memory capacity, memory bandwidth and the speed of the network linking accelerators matter alongside raw arithmetic capability.
GPUs and accelerators
Neural networks spend most of their time on large matrix operations, which specialized processors handle efficiently. Google describes its TPUs as custom application-specific integrated circuits built to accelerate machine-learning workloads (Google Cloud). NVIDIA's annual report describes a software stack built around its CUDA platform that runs on its GPUs for workloads including AI model training and inference (Form 10-K, fiscal 2026), and Broadcom's lists custom accelerators (XPUs) among its AI semiconductor solutions (Form 10-K, fiscal 2025). The lesson for readers: accelerators are a hardware-plus-software ecosystem, not just chips. See Semiconductors.
Cloud infrastructure
Organizations can rent AI capacity instead of building it. Microsoft's annual report describes supercomputing power for AI at scale as part of its Azure offerings (Form 10-K, fiscal 2026); Alphabet says it offers customers a range of accelerator options, including GPUs and its own TPUs (Form 10-K, fiscal 2025); and Amazon lists AI and machine-learning services within AWS (Form 10-K, fiscal 2025). See Cloud & Infrastructure.
Model deployment
Deploying a model means choosing how it will be served: through a managed model service, on infrastructure the organization operates, or as an open-weight model it runs itself. Each choice trades off control over data and latency against operating effort and cost, and the right answer differs by workload. For how those choices affect cost per request, see AI Inference Costs: The Economics of Running AI at Scale.
Data-center electricity demand
Source: IEA, Energy and AI (2025).
| Measure | Figure | As of |
|---|---|---|
| Global data-center electricity consumption | about 415 TWh, or about 1.5% of world electricity | 2024 (IEA estimate) |
| IEA projection for 2030 | more than double, to around 945 TWh | Projection published 2025 |
These are estimates and projections from a 2025 report and will be superseded by later IEA editions.
Economics
The Economics of Artificial Intelligence
AI economics follows the structure of the stack: large fixed investments at the bottom, recurring operating costs in the middle and revenue models at the top. Understanding which cost a given business bears helps explain differences in margins and risk. This section summarizes the structure; specialist articles own the detail.
Training economics
Building a frontier model is a front-loaded investment: compute, data and engineering are paid for before any customer uses the result. Scaling research helps explain why budgets rise with ambition — performance improves with compute, data and parameters, and compute-optimal training balances them.
Inference economics
Every request consumes compute, so inference is a recurring operating cost that grows with usage. A falling price per unit of AI can coexist with a rising total bill when usage and task complexity grow faster than unit prices fall.
Infrastructure utilization
Accelerators are costly assets that earn a return only when kept busy. Unused capacity, demand that arrives in bursts, and the choice between rented and owned capacity are central to whether an AI service is profitable.
Scaling and operating costs
Beyond chips: power, cooling, networking, storage, software, staff, evaluation and monitoring, and the cost of keeping systems secure and compliant. Costs are workload-specific rather than universal.
Enterprise AI spending
Organizations spend through cloud consumption, software subscriptions and add-ons, integration services and data preparation. Adoption is tracked by government surveys such as the U.S. Census Bureau's Business Trends and Outlook Survey; measures differ over time, so compare like with like.
Where revenue is earned
Hardware vendors sell products; cloud providers sell metered capacity; model developers sell access or embed models in their own products; software companies add AI to subscriptions. The companies section below maps these models to the layers.
Go deeper: the specialist article on inference economics. Tokens, model choice, utilization, throughput, caching and the way AI agents complicate cost accounting are analyzed in AI Inference Costs: The Economics of Running AI at Scale. This page deliberately does not repeat that analysis.
Related: Why cloud costs are rising for AI startups · Why AI startups burn cash fast
Corporate AI investment
Source: Stanford HAI, AI Index Report 2026 (published April 13, 2026).
| Measure | Figure | As of |
|---|---|---|
| Global corporate AI investment | $581.7 billion, up 130% from the prior year | Calendar year 2025 |
| of which United States | $285.9 billion | Calendar year 2025 |
Figures are as reported by the AI Index and follow its definitions of corporate investment; they are not a forecast.
Applications
Enterprise AI and Real-World Applications
The application categories below are stable even as individual products change. In each, AI is often embedded in an existing workflow rather than sold as a stand-alone product, and the practical questions are the same: what data does it use, who reviews its output, and what happens when it is wrong?
Software development
Assistants that draft, explain, review and test code, and tools that help maintain large codebases. Human review remains essential: model output that is used without validation can introduce defects or vulnerabilities (see OWASP LLM05, improper output handling).
SaaS & Software →Customer service
Chat and voice assistants, ticket triage and agent-assist tools, often grounded in a company knowledge base through retrieval. The main design question is when the system must hand off to a person.
AI chatbot software →Finance
Forecasting, fraud and anomaly detection, document analysis and compliance support. Financial institutions apply established model-risk practice to quantitative models — for example the Federal Reserve's SR 11-7 guidance on model risk management (April 2011).
AI in finance →Healthcare
Imaging analysis, clinical documentation support and workflow tools. In the United States the FDA maintains a public list of AI-enabled medical devices authorized for marketing, which it notes is not comprehensive.
Healthcare sector →Cybersecurity
Machine learning supports detection and response, while AI systems themselves become targets. NIST publishes a taxonomy of attacks on machine-learning systems (NIST AI 100-2 E2025).
Cybersecurity →Analytics and decision support
Natural-language access to data, summarization of reports and forecasting support. Results depend on data quality, access controls and clear definitions — a model cannot fix inconsistent data.
AI forecasting tools →Productivity
Drafting, summarization, enterprise search and meeting support embedded in the software people already use — for example Microsoft describes Microsoft 365 Copilot and agents in its annual report (Form 10-K, fiscal 2026).
Enterprise software →Automation and AI agents
Multi-step workflows across systems — from routing documents to completing tasks — where a model chooses steps instead of following a fixed script. See the agents section below.
AI agents →Markets
AI Companies and the Capital-Markets Ecosystem
Public companies do not participate in AI in one way. Some sell the chips, some rent the infrastructure, some build models, and many embed AI in products they already sell. Knowing which layer a company operates in — and how it earns money there — is usually more informative about its margins, customers and risks than the label “AI company”.
| Layer | How companies participate | Common revenue models | What to read in company filings |
|---|---|---|---|
| Semiconductors and equipment | Design and sell accelerators, networking and memory silicon; supply equipment used to manufacture chips. | Product sales, often with attached software or licensing. | Customer concentration, supply-chain dependencies, export-control exposure, segment disclosures. |
| Cloud and platform | Build data centers and rent compute, storage and AI services; some also design their own chips. | Metered, consumption-based services and committed contracts. | Capital expenditure, capacity commitments, segment results for cloud. |
| Model developers | Train foundation models — as a division of a large company or as a stand-alone lab. Some labs are privately held. | Vary: access to the model, subscriptions, embedding models into other products. | For public parents: R&D and infrastructure spend; for private labs, funding disclosures elsewhere. |
| Enterprise software | Embed AI in applications and workflow platforms. | Subscriptions and usage-based add-ons. | Retention, pricing changes, AI-related product disclosures, risk factors on competition. |
| Data, network and security infrastructure | Provide the networking, observability, security and data platforms AI workloads depend on. | Product and subscription revenue. | How AI is described in products versus results; exposure to AI-related demand. |
Company profiles by layer
The profiles below are examples of participation, not a ranking, recommendation or complete list. Role descriptions come from each company's own annual report and are attributed accordingly; many companies operate in more than one layer.
NVIDIA
Describes itself as a data-center-scale AI infrastructure company; its CUDA software stack runs on its GPUs for workloads including AI training and inference.
Role as described in the company's Form 10-K, fiscal 2026.
View company profile →AMD
Reports data-center AI accelerator products (AMD Instinct GPUs) alongside EPYC server processors.
Role as described in the company's Form 10-K, fiscal 2025.
View company profile →Broadcom
Designs semiconductor and infrastructure-software solutions; its AI semiconductor solutions include custom accelerators, Ethernet switching silicon and optical components.
Role as described in the company's Form 10-K, fiscal 2025.
View company profile →ASML
Supplies lithography, metrology and inspection systems to chipmakers and says AI-related investment by Logic and Memory customers supported 2025 demand.
Role as described in the company's Form 20-F, fiscal 2025.
View company profile →Microsoft
Offers Azure AI services and supercomputing capacity for AI, and Microsoft 365 Copilot and agents.
Role as described in the company's Form 10-K, fiscal 2026.
View company profile →Amazon
Reports AWS as one of three segments; AWS offers compute, storage, database, analytics and AI and machine-learning services.
Role as described in the company's Form 10-K, fiscal 2025.
View company profile →Alphabet
Reports Google Cloud as a segment, offers customers GPUs and its own TPUs, and develops frontier AI models (Gemini) through centralized AI research.
Role as described in the company's Form 10-K, fiscal 2025.
View company profile →Meta Platforms
Builds AI into its Family of Apps and has released Llama foundation models; says it expects to train a combination of open and closed models.
Role as described in the company's Form 10-K, fiscal 2025.
View company profile →Salesforce
Sells cloud-based customer-relationship and related AI offerings (Agentforce) primarily on a subscription basis.
Role as described in the company's Form 10-K, fiscal 2026.
View company profile →Cisco
Designs and sells networking, security, collaboration and observability products and says it is incorporating AI across them.
Role as described in the company's Form 10-K, fiscal 2026.
View company profile →CrowdStrike
Describes its Falcon cybersecurity platform as AI-native.
Role as described in the company's Form 10-K, fiscal 2026.
View company profile →Reading AI claims critically
Because “AI” attracts attention, statements about it deserve the same scrutiny as any other business claim. In March 2024 the SEC charged two investment advisers over false and misleading statements about their use of artificial intelligence (SEC press release). Useful habits: identify the layer, check what the filings actually say about AI, and separate demonstrated revenue from aspiration. Questions about valuation and price behavior are covered in dedicated market articles, linked in the research map below, not here.
Next application layer
AI Agents and the Next Application Layer
An AI agent uses a model to decide what to do next, take actions through tools, observe the results and continue until a goal is met or a person is needed. NIST's National Cybersecurity Center of Excellence describes AI agents as software systems that use data and algorithms to autonomously perform tasks (concept paper, February 2026). What separates an agent from a chatbot is not intelligence but the ability to act.
Tool use
Research on interleaving reasoning with actions showed that a language model can plan, call external tools and update its plan from what they return (ReAct, Yao et al.). Tools may be search, databases, business applications or code execution.
Standard interfaces
The open Model Context Protocol defines a standard way for applications to give models context and expose tools; its specification is explicit that tools represent arbitrary code execution and require user consent and careful handling (MCP specification).
Retrieval
Agents often need current, permissioned information. Retrieval connects them to an organization's documents and systems so decisions are grounded in data rather than the model's memory alone.
Multi-step workflows
A task such as processing an invoice or resolving a support case involves several decisions across systems. Design choices include how much autonomy to grant, where a person approves, and how failures are detected and reversed.
Automation versus agents
Traditional automation follows fixed rules and is predictable. Agentic systems choose steps at run time, which makes them more flexible but harder to test and to cost, because one task can trigger many model calls.
Identity and authorization
Giving an agent access to data and tools raises questions of identity, permission and auditing. NIST's concept paper seeks input on identification, authorization, auditing and prompt-injection controls for AI agents; OWASP lists excessive agency among its top LLM application risks (OWASP Top 10 for LLM Applications).
Agents also change the cost profile of AI, because a single task can involve many model calls; that is analyzed in AI Inference Costs. Related coverage: Cybersecurity.
Constraints
AI Risks, Constraints and Governance
AI adoption carries real risks alongside its potential. Risk is typically managed not by a single control but by treating AI like any other consequential system: define who is accountable, test before and after deployment, limit what the system can access, and keep humans involved where errors are costly. Distinguishing durable AI value from short-term hype means looking at demonstrated outcomes rather than announcements.
| Risk | What it means | Primary references |
|---|---|---|
| Reliability and hallucinations | Models can produce confidently stated but false or erroneous content (NIST calls this confabulation), which can mislead users. | NIST AI 600-1 |
| Privacy | Leakage or unauthorized use of personal or sensitive data through training data, prompts or outputs. | NIST AI 600-1; OWASP LLM02 |
| Cybersecurity | Attacks on AI systems (such as prompt injection and data poisoning) and the use of AI to lower barriers for offensive cyber activity. | NIST AI 100-2 E2025; NIST AI 600-1; OWASP LLM01 |
| Model and data risk | Defects in training data or models propagate to every application built on them; supply-chain and poisoning risks apply to models, data and embeddings. | Bommasani et al.; OWASP LLM03–LLM04 |
| Infrastructure constraints | Power availability, data-center capacity and chip supply limit how fast AI can scale. | IEA, Energy and AI |
| Regulation and governance | Voluntary frameworks and binding rules define how organizations should identify, measure and manage AI risk. | NIST AI RMF; EU AI Act |
Governance frameworks
The NIST AI Risk Management Framework (AI RMF 1.0, released January 26, 2023) is intended for voluntary use and is organized around four functions — Govern, Map, Measure and Manage; NIST published a Generative AI Profile in July 2024 (NIST AI RMF). The EU AI Act is a binding regulation whose obligations phase in over several years (dated box below). Organizations in regulated sectors such as finance and healthcare may face additional expectations for testing, monitoring and explanation.
EU AI Act — application timeline
Source: European Commission, AI Act page (page last updated August 3, 2026). The Digital Omnibus amendment entered into force on July 27, 2026.
| Milestone | Date |
|---|---|
| AI Act entered into force | August 1, 2024 |
| Prohibited practices apply | From February 2, 2025 |
| Obligations for general-purpose AI models apply | From August 2, 2025 |
| General application of the Act | August 2, 2026 |
| High-risk systems in sensitive areas apply | From December 2, 2027 |
| High-risk systems in regulated products (extended transition) | Until August 2, 2028 |
Regulatory timetables change; confirm dates on the Commission's page before relying on them.
Research map
Explore SaGeminieTech’s Artificial Intelligence Research
Existing SaGeminieTech research on artificial intelligence, organized by subtopic. Each group states what belongs there; specialist articles own their narrower topics and this page links to them rather than repeating them. The map grows as new research is published.
AI Infrastructure · 7 resources
Chips, accelerators, data centers, cloud capacity and the power that sustains them.
Hubs
- SemiconductorsThe chip and manufacturing ecosystem behind AI compute.
- Cloud & InfrastructureCloud platforms and infrastructure economics.
Research and profiles
- Why Cloud Costs Are Rising for AI StartupsCost drivers for AI workloads in the cloud.
- AI Ecosystem — AI and semiconductor sector analysisSector-level view of the AI–semiconductor relationship.
- NVIDIA company profileGPU and data-center computing platforms.
- ASML company profileLithography systems for chip manufacturing.
- Broadcom company profileSemiconductor and infrastructure software.
AI Economics · 3 resources
What it costs to build, run and scale AI — and how those costs shape business models.
Models & Generative AI · 5 resources
Foundation models and the generative AI products built on them. These are product profiles and reviews of specific tools.
- Google Gemini Product ProfileFeatures, pricing, pros and cons.
- ChatGPT Product ProfileFeatures, pricing, strengths, weaknesses and alternatives.
- ChatGPT SaaS ReviewPricing, features and security.
- Claude AI ReviewFeatures, pricing and alternatives.
- Jasper AIAI writing and content-generation platform.
Enterprise AI · 9 resources
How organizations apply AI in software, customer service, finance, healthcare and operations.
Customer, finance and operations
- Best AI Chatbot Software for Customer ServiceCustomer-service AI tools compared.
- AI in Customer AcquisitionStrategies, tools and use cases.
- How AI Will Disrupt FinanceForecasting, compliance and decision-making.
- AI Tools for Financial ForecastingSolutions and use cases.
- FinTech & Financial IntelligenceSector hub for financial technology.
- Healthcare Sector InsightsSector view including healthcare innovation.
Software and automation
AI Agents · 8 resources
Systems that plan and act through tools. Dedicated agent research is still being built out; these are the closest existing resources.
SaGeminieTech
- AI Inference CostsIncludes how agents complicate cost accounting.
- Salesforce company profileEnterprise software company describing agent-based (Agentforce) offerings.
- How to Automate Your StartupPractical automation guide.
- Manufacturing Automation SoftwareAutomation software trends and tools.
- CybersecurityHub covering security topics relevant to agents.
Primary references
AI Companies & Markets · 12 resources
How public and private companies participate in AI, and how markets have treated the theme. Existing market articles are listed for context; this page does not rank or recommend securities.
Company directories
Sector and market context
- AI Ecosystem — AI and semiconductor sector analysisSector overview.
- Why Are AI Stocks Volatile?What drives price swings in AI-related shares.
- AI Stock BubbleValuation risks and growth expectations.
- AI Investment OpportunitiesSectors, stocks, risks and trends.
- Tesla AI and RoboticsAI and robotics at one company.
Private markets
AI Risk & Governance · 6 resources
Reliability, privacy, security, model and data risk, infrastructure limits and regulation. Dedicated SaGeminieTech coverage is still being built out; the closest existing hub and the primary references are below.
SaGeminieTech
- CybersecurityHub for security topics, including AI-related risk.
- AI Stock BubbleValuation and expectations risk in markets.
Primary references
Ongoing coverage
Artificial Intelligence at SaGeminieTech
This vertical will continue to track the technology, economics, infrastructure and business evolution of artificial intelligence. New specialist articles and company profiles are added to the research map above so this page stays the stable entry point, while the detail lives in the pieces that own it.
Evergreen material — concepts, the stack, the structure of costs and risks — changes slowly. Time-sensitive facts appear only in the labelled dated-data boxes, each with its source and review date, and are refreshed when the underlying source is updated: a new edition of the IEA or AI Index reports, EU AI Act milestones or amendments, new NIST publications, or annual-report changes affecting the company descriptions. Last reviewed: 2026-09-21.
Related technology and market hubs
Sources and references · primary and authoritative sources used on this page
Company descriptions reflect each company's own annual report and are attributed as such. Definitions are paraphrased from the sources shown. Figures in dated-data boxes are quoted with their as-of dates.
Definitions and concepts
- OECD — definition of an AI system
- NIST AI 600-1 — Generative AI Profile (July 2024)
- Bommasani et al. — On the Opportunities and Risks of Foundation Models (2021)
- Vaswani et al. — Attention Is All You Need (2017)
- Kaplan et al. — Scaling Laws for Neural Language Models (2020)
- Hoffmann et al. — Training Compute-Optimal Large Language Models (2022)
- Lewis et al. — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020)
Infrastructure and energy
- MLCommons — MLPerf Inference: Datacenter
- Google Cloud — Introduction to Cloud TPU
- IEA — Energy and AI (2025)
Agents, security and governance
- Yao et al. — ReAct (2022)
- Model Context Protocol — specification
- NIST NCCoE — AI agent identity and authorization concept paper (February 2026)
- NIST AI 100-2 E2025 — Adversarial Machine Learning taxonomy (March 2025)
- OWASP — Top 10 for LLM Applications (2025)
- NIST — AI Risk Management Framework
- European Commission — AI Act
- Federal Reserve — SR 11-7, Guidance on Model Risk Management (2011)
- FDA — AI-enabled medical devices list
- SEC — press release 2024-36
- U.S. Census Bureau — Business Trends and Outlook Survey
Data and market context
Company annual reports (SEC)
- NVIDIA Form 10-K, fiscal 2026
- AMD Form 10-K, fiscal 2025
- Broadcom Form 10-K, fiscal 2025
- ASML Form 20-F, fiscal 2025
- Microsoft Form 10-K, fiscal 2026
- Amazon Form 10-K, fiscal 2025
- Alphabet Form 10-K, fiscal 2025
- Meta Platforms Form 10-K, fiscal 2025
- Salesforce Form 10-K, fiscal 2026
- Cisco Form 10-K, fiscal 2026
- CrowdStrike Form 10-K, fiscal 2026
This page is informational and is not investment, legal or professional advice, a recommendation to buy, sell or hold any security, or a prediction of future results. See SaGeminieTech's Disclaimer.
Continue Exploring Artificial Intelligence
Browse AI company profiles, in-depth articles, and the wider technology and public-markets ecosystem on SaGeminieTech.