aldena learnupdated

who are the major developers of ai?

the major model developers are openai, anthropic, google deepmind, meta, mistral, alibaba, and deepseek. microsoft and amazon are mostly distributors with their own smaller models. nvidia sits underneath all of them as the hardware supplier, which arguably makes it the most consequential company in the list.

the model developers

organisationbest known forweights
openaigpt series, chatgpt, codexclosed
anthropicclaude series, claude codeclosed
google deepmindgemini, alphafoldmostly closed, gemma open
metallama seriesopen weights
mistralefficient european modelsmixed
alibabaqwen seriesopen weights
deepseekstrong reasoning models at low training costopen weights
xaigrokmixed

The open-weight column matters more than the ranking. It decides whether you can run a model on your own hardware, which decides whether a data-residency conversation is a conversation or a wall.

the layer underneath

Nvidia designs the accelerators nearly all of the above train and serve on, plus cuda, which is the software moat that keeps competitors from being drop-in replacements. Amd and google's own tpus are the credible alternatives, and both are gaining, slowly.

Tsmc manufactures the chips, and asml makes the machines tsmc uses. Three companies deep, and you reach a supply chain with no second source. Anyone assessing concentration risk in ai should be looking here rather than at model vendors.

the distributors

Microsoft, amazon, and salesforce build comparatively little frontier capability and control a great deal of how it reaches people, through azure, bedrock, office, and the crm most sales teams live in. In enterprise software, distribution has beaten capability every previous time, and there is no obvious reason this round is different.

why the list keeps being read wrong

Because "major developer of ai" gets used for three different things: who trains frontier models, who ships ai products at scale, and who supplies the compute. Those are three different lists with three different leaders, and coverage mixes them constantly.

The practical version: the model developers decide what is possible, the hardware suppliers decide what it costs, and the distributors decide what most people actually use.

how this works in aldena

Aldena is on none of these lists and deliberately so. It does not train models. It gives them a workplace: eleven prebuilt roles working in an isolated room with its own server, arranged in an org chart so managers delegate and a reviewer checks the result.

Which developer's model runs each agent is a per-agent setting from the published catalog, and you can mix providers in one room. That keeps the question above a choice you revisit rather than a bet you made once, which is the sensible posture when the ranking changes every few months.

ready when you are

spin up your first room.

one room per client, project, or product, staffed with a project manager, an analyst, engineers and a reviewer.