People ask “How many AIs are there?” — and it’s a trick question. There’s no single registry. Below I break down types of AI, show a practical counting method, and give a realistic estimate you can explain to friends.
Why counting “AI” is not like counting cars
When people say “How many AIs exist?” they usually mean one of three things: how many AI products (services), how many distinct AI models, or how many devices/systems that use AI. Each question gives a very different number.
So the first rule: define your unit — are you counting models, services, or deployments? I’ll show all three and explain why the numbers vary wildly.
AI categories — the way I break the world down
To be useful, we split “AI” into clear groups. Each group is easier to estimate:
- Foundation models — large models like GPT, Gemini, Llama derivatives (distinct trained models).
- AI services & apps — web services, chatbots, image generators, search assistants.
- Enterprise deployments — custom AI systems inside companies (HR bots, fraud detectors).
- Edge & consumer devices — smartphones, cameras, cars that run AI features.
- Open-source & hobby models — forks, small models, research prototypes.
Each category grows at a different pace; foundation models are tens–hundreds, while AI features embedded in phones are in the hundreds of millions.
My practical 2025 estimate (one-page summary)
Below is a simple, conservative estimate based on public data, market reports, open-source listings, and reasonable extrapolation.
Estimated counts (conservative ranges)
- Foundation models: ~ 50 – 300 distinct large models (major vendors + community forks)
- AI services & apps (public): ~ 5,000 – 20,000 (chatbots, SaaS, APIs)
- Enterprise deployments: ~ 50,000 – 500,000 (company-level AI systems)
- Edge & consumer devices with AI features: ~ 1,000,000,000+ (phones, cameras, smart devices)
- Open-source & hobby models: ~ 10,000 – 100,000 (smaller models, research)
Bottom line: if you count “AI systems” narrowly (models/services), you're in the thousands. If you count every device running AI features, the number is in the billions.
Concrete examples (what counts as “one AI”?)
Here are real-world items you might call “an AI”:
- OpenAI GPT-4 — a foundation model (counts as 1 model)
- ChatGPT.com — a public AI service (counts as 1 service)
- Company fraud detection system — an enterprise AI (counts as 1 deployment)
- Smartphone camera face recognition — each device is a deployment (millions)
So the phrase “How many AIs” must be clarified — that's the key takeaway.
Quick comparison table
| Category | What it is | Estimated count (2025) | Why the range |
|---|---|---|---|
| Foundation models | Large trained models (GPT, Gemini, Llama forks) | 50 – 300 | Major labs + open source forks |
| Public AI services | APIs, chatbots, image sites, assistants | 5,000 – 20,000 | Many startups + products launch weekly |
| Enterprise deployments | Company-specific AI systems (internal) | 50,000 – 500,000 | Large variety; many private systems not public |
| Devices with AI | Phones, cameras, cars with on-device AI | 1,000,000,000+ | Every smartphone + smart camera counts |
| Research & hobby models | Small open models, experiments | 10,000 – 100,000 | Many tiny projects on GitHub and Hugging Face |
How I arrived at these estimates (short methodology)
I used a simple, conservative approach:
- Count public foundation models listed by major hubs (OpenAI, Google, Meta, Hugging Face).
- Survey product marketplaces and directories for public AI services (several thousand entries exist on app lists and product directories).
- Estimate enterprise deployments by extrapolating from survey data (many companies run multiple AI systems internally).
- Devices: extrapolate from global smartphone and IoT shipment numbers — most modern devices include at least one AI feature.
- Add a buffer for hobby/research models (GitHub repos, Hugging Face models, private forks).
This is intentionally conservative — many private and internal systems are invisible to public counts.
Why this number matters (and why it doesn't)
Counting AIs helps understand scale — how many systems might affect jobs, privacy, and society. But an exact count is less useful than understanding distribution: how many powerful models vs. how many simple on-device features exist.
For example, ten foundation models could power thousands of services — so policy and ethics should focus on influence, not just raw counts.
How to ask the question better (so you get a useful answer)
If you want a number that helps, try asking:
- “How many public AI models are there?”
- “How many AI services (chatbots/APIs) are publicly available?”
- “How many devices run AI on the edge?”
Each version yields a different, useful answer.
FAQ
Can we ever count all AIs exactly?
Not realistically — private, offline, embedded, and transient AI systems are difficult to discover. We can make good estimates but not a single authoritative count.
Does each smartphone count as “one AI”?
It depends. If you count devices with any AI feature, yes. But if you count distinct AI systems or models, no — many phones run the same model version.
Are open-source models included?
Yes — they make up a large and fast-growing part of the ecosystem (tens of thousands of small models and forks).
Short history of AI (how we reached billions)
AI didn’t suddenly appear in 2025 — it evolved in waves. Each wave massively increased the number of AIs in the world.
- 1950s–1980s: Early symbolic AI, rule-based systems. Very few AIs existed.
- 1990s: Machine learning boom. AI started appearing in research labs and select industries.
- 2012: Deep learning revolution (ImageNet). AI systems exploded in capability and quantity.
- 2018–2020: Transformer models — the birth of modern foundation models.
- 2023–2025: Generative AI era — chatbots, copilots, image models, agents, and AI in almost every device.
These waves explain why counting AI today is so complicated — different generations still coexist.
Future: How many AIs will exist by 2030?
Based on current growth, the number of AI systems will multiply rapidly. By 2030, we can expect:
- Foundation models: 300–1,000+
- Smart devices with AI: 3–5 billion (almost every phone, car, and appliance)
- AI-enabled software tools: 50,000–200,000+
- Autonomous agents: millions (AI bots performing tasks independently)
The growth is exponential — not linear.
What are the benefits of having so many AIs?
More AI systems in the world means more capabilities for society:
- Faster productivity: AI automates boring tasks so people can focus on creativity.
- Better healthcare: detection systems, drug discovery, patient monitoring.
- Safer transportation: AI in cars, traffic systems, navigation.
- Personal assistance: AI companions, writing help, coding help, design help.
- Global access: people in remote areas get digital support instantly.
What are the risks of having so many AIs?
Large numbers of AI systems also bring challenges:
- Privacy issues: more data collection by apps and devices.
- Misinformation: more AI-generated fake content if not regulated.
- Job displacement: routine work may reduce.
- Security risks: hacked AIs can cause large-scale problems.
- Bias: poorly-trained AI can give unfair outcomes.
Understanding the risks helps us build safer AI systems.
Where is AI used the most?
- Chatbots & writing tools: ChatGPT, Gemini, Claude.
- Image & video generation: Midjourney, DALL·E.
- Search & discovery: AI-powered search engines.
- Education: homework help, tutoring, explanations.
- Business tasks: emails, marketing, analytics.
- Healthcare: diagnosis assistance, scans.
- Coding & automation: code generation, debugging.
Conclusion
There is no fixed number of AIs in the world. It depends on what we count — models, applications, or devices.
The realistic picture is this:
- Thousands of AI models and services exist.
- Millions of businesses use AI internally.
- Billions of devices run AI features daily.
Instead of asking “How many AIs exist?” a better question is “How powerful, accessible, and safe are the AI systems we use?”