The frontier
The labs
Ten labs, 133 releases, read against the durability curve. Each page carries the same three things: where the site reads their value position, what would change that read, and the substrate underneath it.
Reading onto the architecture layer
The durable position is a substrate other work sits on. The model underneath still commoditises.
- Anthropic Anthropic is climbing the stack toward the layer where models do real work. Claude Code and the agent tooling are its durable position; the raw model underneath commoditises like everyone else’s.
- Google Google’s durable position is the distribution it already owns: Search, Android, Workspace, and the TPU stack the models run on. Gemini compounds because it ships into surfaces a billion people open every day.
Reading mixed
Some of the value sits in a substrate, some in the model itself. The split decides what survives.
- Alibaba (Qwen) Qwen’s durable position is ubiquity: the most downloaded open-weight family, the default base thousands of teams fine-tune, wired into Alibaba Cloud’s distribution across Asia. The models commoditise; the ecosystem around them is the moat.
- Zhipu (Z.ai) Zhipu’s durable position is the open GLM ecosystem plus the enterprise and state relationships of the first publicly listed Chinese AI lab. Public-market funding buys the runway; the open weights buy the adoption that compounds into it.
- xAI xAI’s models live on the commoditising output layer, so its edge has to come from elsewhere. The durable bets are distribution through X and compute at Colossus scale.
- Meta Meta’s strength is three billion users and a strategy of commoditising rivals’ moats with open weights. The 2026 pivot to closed frontier models bets that distribution plus a private model beats giving the weights away.
Reading onto the output layer
Value is captured per model. That is the layer the curve says thins fastest.
- OpenAI OpenAI’s scarcity sits on the output layer, the part of the stack that commoditises fastest. Its durable bet is ChatGPT as a daily habit and the platform that routes work to whichever model wins. The frontier model itself is the commodity.
- Mistral AI Mistral ships open weights, the fastest-commoditising part of the stack. The durable position is European sovereignty, on-prem enterprise trust, and Le Chat as a distribution surface.
- DeepSeek DeepSeek lives at the sharp end of the output layer: open weights, frontier-adjacent quality, trained for a fraction of the cost. The durable asset is the efficiency itself, the ability to keep matching the frontier cheaply enough to reset everyone’s price expectations.
- Moonshot AI Moonshot’s models are open and frontier-adjacent, which is the fastest-commoditising place to stand. The durable bet is Kimi as a consumer habit in China and the agentic, long-context niche it reached first.