Who Builds These AIs — The Companies Behind ChatGPT, Claude, and the Rest

In brief

You use ChatGPT, Claude, or Gemini? Behind each assistant, there’s a company that built the AI model (the “brain” of the software). There are only a handful of them in the world, but they are engaged in fierce competition — through funding, through technology, and sometimes through geopolitics. This article tours the landscape: who they are, where they come from, and what exactly they do.

Picture the car-maker market in the 1920s: a dozen brands shared the world, each with its own strategy — luxury, popularity, sport, reliability. The AI model market in 2026 looks like that, except the playing field is global and the factories cost several billion dollars.

The names you already know

If you’ve ever used an AI assistant, you’ve certainly come across one of these three.

OpenAI — the company behind ChatGPT

OpenAI, based in San Francisco, is the one that started the wave. ChatGPT, released at the end of 2022, was the first AI assistant to reach the general public at scale. Their latest model is called GPT-5. In 2025, ChatGPT counted approximately 800 million weekly users. It is by far the best known, though not necessarily the most used among professionals.

Anthropic — the company behind Claude

Anthropic was founded by former members of OpenAI. Its assistant is called Claude. The company has built a reputation on the reliability and safety of its models, which has allowed it to capture a significant share of the professional market — particularly in software development. Anthropic is also based in the United States and is one of the best-funded companies in the sector, with billions of dollars in investment.

Google DeepMind — the company behind Gemini

Google has an advantage that no one else has: distribution. Gemini, its AI assistant, is integrated directly into Google Search, Gmail, and Android. The result: hundreds of millions of users access it without even looking for it. Google also designs its own specialized chips (TPUs) to run its models, making it less dependent on external suppliers.

In short: OpenAI, Anthropic, and Google form the American “top 3.” OpenAI has the consumer mass, Anthropic the trust of pros, Google the distribution baked into Search/Gmail/Android. Three different positions, three different targets.

The ones you may not know yet

Beyond the three giants, other players matter — and some are changing the game.

Meta — the giant that gives everything away for free

Meta (the parent company of Facebook and Instagram) builds AI models called Llama. The distinctive feature: Meta publishes them for free, in open access. Anyone can download and use them. Why? Meta doesn’t sell the AI itself — the company wants to prevent a competitor (like OpenAI or Google) from locking up the market with a closed system. It’s a strategy of influence as much as technology.

Mistral — the French champion

Mistral AI is based in Paris. With fewer than 300 people, it’s a small outfit compared to the American giants. But it has managed to position itself in the European professional market, betting on data sovereignty (a strong argument for companies that don’t want to entrust their information to American servers). The company also publishes some of its models as open access.

DeepSeek — the Chinese surprise

At the start of 2025, a startup from Hangzhou (China) with around 200 employees published a model called DeepSeek R1. Surprise: on mathematical reasoning and programming benchmarks, it rivaled the best OpenAI models — at a fraction of the cost. The announcement rattled financial markets and challenged a widespread belief: that only players with colossal budgets could build state-of-the-art models.

A global competition

What’s at stake here goes beyond companies: it is also a competition between nations. The United States dominates with OpenAI, Anthropic, Google, and Meta. China is rising with DeepSeek, Alibaba (which publishes the Qwen model series), and Baidu. Europe is trying to hold its place with Mistral. Governments are involved: the United States, for example, has restricted the export of certain chips to China to slow its progress. A paradoxical result: deprived of these chips, Chinese teams were pushed to innovate differently, making their software more efficient.

Open or closed — two philosophies

You’ll often hear about “open” models (open source, meaning the code or parameters are accessible to all) and “closed” models (accessible only through the company’s service).

• Closed: ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google). You access them through their website or app, but you cannot see how the model works in detail.
• Open: Llama (Meta), some Mistral models, DeepSeek R1. Anyone can download, study, and modify them.

Both approaches have strengths. Closed models are often easier to use and come with better support. Open models allow more control and transparency. The boundary between the two is blurring: open models are quickly closing the performance gap with closed ones.

In short: choosing an open or closed model is a tradeoff between ease of use and control. For an individual, a closed model via an app just works. For a company that must keep its data in-house, an open model hosted internally becomes the option to consider.

Key takeaways

• OpenAI (ChatGPT), Anthropic (Claude), and Google (Gemini) are the three dominant players, all American.
• Meta publishes its models freely in open access — a strategy to prevent a closed monopoly.
• Mistral (France) and DeepSeek (China) show that competing is possible with more modest resources and from other continents.
• The competition is also geopolitical: governments invest and regulate to avoid being left behind.
• Open and closed models coexist, and the performance gap between the two is narrowing rapidly.