Chatbot referral leader
—
—
Loading the saved source checks…
SOURCED MARKET INTELLIGENCE
Understand the business
behind AI.
Follow the money, understand the technology and see what the leading companies are building next. Clear charts. Dated evidence. Sources you can check.
Business adoption leader
—
—
Enterprise LLM spend leader
—
—
Cloud infrastructure leader
—
—
UNDERSTAND THE INDUSTRY
From chips to useful work
Follow the numbered steps. Each layer supplies something the next layer needs. A company can operate in several layers.
Highlight a company:
- 1
Chips & systems
The hardware that performs the calculations.
Supplies computing power - 2
Cloud & data centers
Power, cooling and hosted computing capacity.
Runs training and inference - 3
Models
Systems that learn patterns and generate or reason over content.
Provides model capabilities - 4
Apps & agents
Interfaces and tools that turn capabilities into tasks.
Delivers help and actions - 5
People & businesses
The value comes from work completed successfully.
- Write and check code
- Research and summarize
- Support customers
- Analyze documents
Useful outcomes
How to read this map and check the examples
Training develops a model. Inference means running it to answer a request. Agents combine models with tools and a repeated plan–act–check process. Products can use several models or clouds. Highlighting indicates example activities, not market share.
Company examples: Google’s AI stack · Microsoft’s platform · OpenAI’s infrastructure · Anthropic’s products · NVIDIA’s platform. Editorial review: 22 September 2026.
FOLLOW THE MONEY
What is the AI build-out costing?
Large infrastructure bills sit behind the apps. Here is what companies actually reported, with the accounting basis available for each company.
Research reviewed 22 September 2026. These figures and company plans require editorial review; the refresh button updates market shares and news.
Company capital investment
Quarter ended 30 June 2026 · USD billions · company-wide, including non-AI activity
Capital expenditure buys long-lived assets such as equipment and data centers. These disclosures use different lease treatments, so the bars show scale rather than an exact like-for-like ranking of AI spending.
SELECT A COLUMN TO SEE WHAT IS INCLUDED
Amazon · $53.1B
Cash capital expenditures
Alphabet / Google · $44.92B
Purchases of property and equipment
Microsoft · $41B
Capital expenditures including finance leases
Meta · $31.08B
Capital expenditures including finance-lease principal payments
Microsoft investment dollars went to CPUs, GPUs and other short-lived assets
≈ two-thirds short-lived assets · ≈ one-third long-lived assets
Share of its reported $41B capital expenditures in Apr–Jun 2026. The remainder was long-lived assets. This is an approximate company disclosure.
Source · 29 Jul 2026 ↗NVIDIA spent on research and development in one quarter
Reported GAAP R&D expense for the quarter ended 26 July 2026 (FY2027 Q2). Operating expense, not capital expenditure or an AI-only budget.
Source · 26 Aug 2026 ↗Anthropic committed to AWS technologies over ten years
Announced commitment for up to 5 GW of new capacity. This is a multi-year agreement, not money already spent or currently operating capacity.
Source · 20 Apr 2026 ↗RESEARCH & DIRECTION
What are the leaders building toward?
Select a company to see its disclosed investment priorities, research focus and stated view of the future.
Company plans are ambitions, not guaranteed outcomes. Public disclosures rarely reveal a complete AI budget, so “largest spending category” is only stated where supported.
COMPANY BRIEFING
Reviewed 22 September 2026OpenAI
Models, assistants and agents
01 / MONEY
Where investment is going
Compute and energy infrastructure are a major disclosed commitment. OpenAI announced a $500M investment in SB Energy; this is one deal, not its total budget. It does not publish a complete spending breakdown in these sources.
Financial / investment source · 9 Jan 2026 ↗02 / RESEARCH
What it is working on
Larger-scale training, more efficient inference and broader access to capable models. Stargate brings data centers, chips and energy partners together to support that work.
Company source · 29 Apr 2026 ↗03 / FUTURE
Its stated direction
OpenAI presents more abundant compute as a way to deliver stronger models and make advanced AI useful to more people and businesses.
Company outlook · 29 Apr 2026 ↗What to watch · editorial analysisHow much contracted capacity becomes operational, and whether reliability and the cost of using models improve.
COMPANY BRIEFING
Reviewed 22 September 2026Anthropic
Claude, coding and work agents
01 / MONEY
Where investment is going
A disclosed commitment of more than $100B to AWS technologies over ten years supports training and running Claude. It is not an annual expense; no complete spending split is disclosed here.
Financial / investment source · 20 Apr 2026 ↗02 / RESEARCH
What it is working on
Safety and interpretability: understanding model behavior, alongside more compute and improvements to Claude Code and Cowork.
Company source · 28 May 2026 ↗03 / FUTURE
Its stated direction
Anthropic says it wants Claude to become more useful across everyday and enterprise work, supported by greater capacity and adaptable products.
Company outlook · 28 May 2026 ↗What to watch · editorial analysisWhether agents complete longer tasks reliably and whether safety findings translate into measurable safeguards.
COMPANY BRIEFING
Reviewed 22 September 2026Google / DeepMind
Chips, cloud, models and consumer products
01 / MONEY
Where investment is going
Alphabet reported $44.92B of property-and-equipment purchases in Apr–Jun 2026. Google describes large infrastructure and custom-chip investment; the figure includes the whole company.
Financial / investment source · 22 Jul 2026 ↗02 / RESEARCH
What it is working on
World models, multimodal generation, long-running agents, scientific tools and specialized chips for training and inference.
Company source · 19 May 2026 ↗03 / FUTURE
Its stated direction
Sundar Pichai describes AI moving into background tasks and everyday products, with agents helping people act and tools accelerating scientific discovery.
Company outlook · 19 May 2026 ↗What to watch · editorial analysisWhether demonstrations become broadly available products, and how much useful work they complete without correction.
COMPANY BRIEFING
Reviewed 22 September 2026Microsoft
Azure infrastructure and workplace AI
01 / MONEY
Where investment is going
Reported $41B of capital expenditures in Apr–Jun 2026, including finance leases. About two-thirds funded short-lived assets, mainly CPUs and GPUs; the rest funded long-lived assets.
Financial / investment source · 29 Jul 2026 ↗02 / RESEARCH
What it is working on
Enterprise agents connected to business data, persistent memory, tools, evaluations and governance; ongoing investment in research compute, talent and data.
Company source · 29 Jul 2026 ↗03 / FUTURE
Its stated direction
Microsoft describes Copilot evolving toward multi-step and long-running work, with security and management systems for the agents companies deploy.
Company outlook · 29 Jul 2026 ↗What to watch · editorial analysisCustomer retention, task success and infrastructure efficiency, rather than counting agent registrations alone.
COMPANY BRIEFING
Reviewed 22 September 2026Amazon / AWS
Cloud infrastructure, chips and AI services
01 / MONEY
Where investment is going
$53.1B in cash capital expenditures in Apr–Jun 2026, primarily technology infrastructure, mostly for AWS, and additional fulfillment capacity. Amazon is not reporting an AI-only number.
Financial / investment source · 30 Jul 2026 ↗02 / RESEARCH
What it is working on
Custom Trainium silicon, AI infrastructure and services that support training and deploying models. The company reports growing adoption of its chips by AI labs and other customers.
Company source · 30 Jul 2026 ↗03 / FUTURE
Its stated direction
Andy Jassy emphasizes growth in AWS, AI and chips. The strategy described is to provide capacity and tools for customers building AI applications.
Company outlook · 30 Jul 2026 ↗What to watch · editorial analysisWhen new capacity earns revenue and whether custom chips improve customers’ cost and performance.
COMPANY BRIEFING
Reviewed 22 September 2026Meta
Social products and personal AI
01 / MONEY
Where investment is going
$31.08B of capital expenditures in Apr–Jun 2026, including finance-lease principal payments. Its 2026 capital-spending guidance was $130–145B; guidance is a plan, not realized spending.
Financial / investment source · 29 Jul 2026 ↗02 / RESEARCH
What it is working on
Personal agents, advanced models and open models that can help people and small businesses pursue their goals.
Company source · 10 Aug 2026 ↗03 / FUTURE
Its stated direction
Mark Zuckerberg argues for highly capable personal AI that understands individuals and makes advanced capabilities widely accessible. This is Meta’s stated ambition.
Company outlook · 10 Aug 2026 ↗What to watch · editorial analysisUseful everyday adoption, privacy controls and which models and capabilities are actually released openly.
COMPANY BRIEFING
Reviewed 22 September 2026NVIDIA
AI chips, systems and software
01 / MONEY
Where investment is going
Reported $7.05B in research-and-development expense in FY2027 Q2. This is the largest named operating-expense line, not its biggest cost overall; manufacturing-related cost of revenue is separate.
Financial / investment source · 26 Aug 2026 ↗02 / RESEARCH
What it is working on
Integrated CPUs, GPUs, networking and software for agentic AI; scientific computing and physical AI are additional areas highlighted in its announcements.
Company source · 26 Aug 2026 ↗03 / FUTURE
Its stated direction
Jensen Huang describes a broader ecosystem of AI labs and physical AI creating demand for computing. This is the company’s outlook, not a guaranteed demand forecast.
Company outlook · 26 Aug 2026 ↗What to watch · editorial analysisReal customer deployments, performance per watt and the cost of running useful workloads.
Four questions shaping the next phase
Editorial synthesis of the briefings above. These are themes, not a ranking of research budgets.
Can agents finish the job?
Moving from answers to reliable, multi-step work.
Microsoft · Google · AnthropicCan compute get cheaper?
More capacity, specialized chips and better efficiency.
NVIDIA · AWS · OpenAICan AI understand the world?
Multimodal systems, simulation and scientific discovery.
Google DeepMind · NVIDIACan we trust personal AI?
Useful context, understandable behavior and human control.
Anthropic · Meta · MicrosoftTHE MARKET IN NUMBERS
Four measures. Four different questions.
Compare providers, explore spending and see how concentrated the cloud market is. Exact values stay visible alongside every visual.
01
Consumer AI assistants
Share of measured referrals from AI chatbots to websites worldwide.
—
Which chatbot sends the most referrals?
Share of measured AI chatbot referrals to websites
Each dot marks a share on the same 0–100% scale. These are website referrals, not total chatbot users.
COMPARE TWO PROVIDERS
—
—
Open source ↗02
Business AI adoption
How many businesses are actually paying the major model providers?
—
Which providers do businesses pay for?
Share of businesses in Ramp’s dataset; providers overlap
Each grid represents 100%. One square = one percentage point; a partly filled square shows a fraction. This illustrates the rate, not a sample of 100 individual businesses. The same business can appear in both providers’ rates.
—
overall paid AI adoption in the stated Ramp reporting period · source
—
These numbers can exceed 100% in total because one business may purchase from multiple AI providers.
Open source ↗
03
Enterprise LLM API spending
Where enterprise model/API dollars are estimated to flow.
—
Where do enterprise API dollars go?
Estimated share of enterprise LLM API spending
Area represents each provider’s share of the reported total. Select a block to explore it.
EXPLORE AN ILLUSTRATIVE BUDGET
If your budget followed this market split, how much would go to each provider? A hypothetical illustration in USD, not a price quote or spending recommendation.
—enterprise GenAI spend
—AI application spend
—foundation-model API spend
—
Open source ↗04
AI agents & digital workforce
An assistant answers a request. An agent can use tools and work through several steps toward a goal.
NO RELIABLE SHARE YET
Capability map, not market shareNo comparable cross-vendor agent market-share dataset is connected.
- 1
Give a goal
“Research suppliers and draft a comparison.”
- 2
Make a plan
Break the work into steps and decide which tools to use.
- 3
Use tools
Search, read documents or run code, within the permissions given.
- 4
Check the result
Review evidence and ask a person to approve consequential actions.
Conceptual workflow: real agents can repeat steps, ask for help or stop when a task fails. Autonomy and reliability vary by product.
Coding
Edit code, run checks and explain changes.
Examples: Codex · Claude CodeKnowledge work
Research information and prepare documents.
Examples: Claude Cowork · Microsoft CopilotBusiness workflows
Connect company data and tools to complete tasks.
Examples: Microsoft Foundry · Google agentsIllustrative categories; see the dated company briefings and sources for context.
05
AI-enabling cloud infrastructure
The cloud layer underneath a large part of the AI economy.
—
How concentrated is the cloud market?
Share of worldwide cloud infrastructure service revenue
Select the big three or one provider. The coloured arc shows the selection; the dark arc is the rest of the market. This covers cloud infrastructure overall, not just AI.
—quarterly cloud market
—year-on-year growth
—
Open source ↗NEWS DESK
Latest from the publishers
Official company announcements from selected RSS feeds, newest first. Coverage is selective and reflects each publisher's perspective.
News feed status and last successful checks
METHODOLOGY
How to read this dashboard
Referrals ≠ all chatbot usageStatcounter measures referrals to websites, rather than total chatbot users, conversations or revenue.
Adoption ≠ spending shareA company can buy several providers, so provider adoption totals may exceed 100%.
Cloud share ≠ AI shareAWS/Azure/GCP market share covers broader cloud services, even though AI drives much of current growth.
Agent share is immatureUntil comparable agent revenue/usage data exists, a vendor market-share pie would create false precision.