The Major Players

The Major Players

The AI landscape is dominated by a small number of organisations with the compute, data and talent to train frontier models — but the ecosystem is much broader. Understanding who does what helps you navigate tool choices, API decisions, and the broader strategic picture.

The Frontier Labs

These organisations train the most capable general-purpose AI models:

Anthropic: Founded 2021 by former OpenAI researchers. Builds Claude. Focus on safety and interpretability research. Backed by Google and Amazon.

OpenAI: Founded 2015. Builds ChatGPT and the GPT model family. The company that kicked off the current wave of mainstream adoption in late 2022. Backed by Microsoft (major Azure integration).

Google DeepMind: Merged Google Brain and DeepMind in 2023. Builds Gemini. Has unique advantages in TPU compute and search integration.

xAI: Founded 2023 by Elon Musk. Builds Grok, integrated into the X platform. Rapidly built out large-scale GPU clusters to compete at the frontier.

Meta AI: Historically built the open-weight Llama series; now also ships closed models (see the open-source section below).

Mistral: Paris-based, founded 2023. Builds efficient, high-performance models. Mix of open-weight and commercial offerings.

DeepSeek, Alibaba (Qwen) and Moonshot (Kimi): Chinese labs that have become major forces at the frontier, particularly in open-weight models — see the open-source section below.

As of October 2026: OpenAI's early "o-series" reasoning models (o1, o3) are now legacy — reasoning is a built-in, adjustable capability of its current GPT-family models, not a separate product line.

The Cloud Hyperscalers

Microsoft Azure: Deep OpenAI partnership — Azure OpenAI Service is the enterprise on-ramp for GPT models — but Microsoft also now trains its own in-house models (see callout). Also offers a wide range of open-weight models via Azure AI.

Amazon AWS: Invested in Anthropic. Offers Claude, Llama, Mistral, and others via Amazon Bedrock, alongside its own model family. Strong enterprise reach.

Google Cloud: Offers Gemini and other models via Vertex AI. Google is unusual among the hyperscalers in being both a cloud provider and a frontier lab in its own right (Google DeepMind, above), with tight integration between the two.

It's no longer accurate to say the hyperscalers don't train frontier models themselves: Google always did, and Microsoft and Amazon have since followed with model families of their own. Enterprises often still choose a cloud provider's AI offering for security, compliance, and billing simplicity, regardless of who actually trained the underlying model.

As of October 2026: Microsoft's MAI-Thinking-1 (June 2026), from its in-house Microsoft AI unit, is its first self-trained frontier reasoning model. Amazon's Nova model family powers Bedrock and Alexa+.

The Open Source Ecosystem

Meta built the Llama series, which for several years was the most widely used open-weight model family — a strategic choice to commoditise the model layer while competing on applications and data.

Chinese labs — including DeepSeek, Alibaba's Qwen and Moonshot's Kimi — now lead much of the open-weight frontier, often releasing models that are competitive with the best closed models within weeks of them shipping.

Hugging Face: The central hub for open-source AI. Hosts models, datasets, and the Transformers library. The GitHub of AI.

Ollama: Makes running open models locally trivially easy — a single command downloads and runs a model (see callout for a current example).

Community fine-tunes: Open-weight models are constantly fine-tuned by the community — for coding, roleplay, instruction following, specific languages. The Hugging Face model hub has hundreds of thousands of these.

Open-weight models are closing the gap with proprietary ones, especially in coding and instruction-following tasks.

As of October 2026: Meta's Llama 4 (April 2025) was the last Llama release — Meta's Muse Spark (April 2026) was the company's first closed-weights model. Current open-weight leaders include DeepSeek, Qwen, Kimi and GLM, alongside Gemma, Mistral and OpenAI's own open-weight gpt-oss models. A current example: ollama run qwen3.