AI Market Share

How it was born

I published it on June 22, 2026, on the same day as StoryMaker, as the first two entries of what was then called the vibecoding page. My brief was a single sentence, who is winning, and the first thing I settled was that the answer changes every eighteen months and that a static chart would be a lie by the time it loaded.

So the chart moves. A bar chart race, one frame per period, with the bars reordering as players rise and fall. Pause it on any year and the picture of that year is the picture that was true then, not the one that is true now.

What it does

• Five boards, one per tab: AI assistants and large language models, image generation, video generation, 3D generation, and the creative platforms that bundle the models rather than train them. Sixty-one series in all, from the research era of BERT and StyleGAN to the current crop.
• The model behind each row, named as it ships: the enterprise board walks from Claude 1 to Fable 5.1 and from GPT-3.5 to GPT-6 Astra, the 3D board from Tripo to its P2.0 preview. Pause on a date and the names are the names of that date.
• Two lenses on the assistants: consumer usage, and enterprise measured as API spend. They do not rank the same players in the same order, which is the point of showing both.
• Pause, replay, reorder on or off, and every tab and year has its own link, so a view can be sent as it is.
• Export to MP4, timestamped, for the deck where the chart will end up anyway.

What the numbers are

Estimates, and the app says so under the chart. The series are illustrative trend estimates, not official market data, normalised to 100 percent per period. The early years mostly reflect research-lab prominence, GPT-1, BERT, GPT-3; the consumer market tips with ChatGPT, then Gemini, Claude, Grok and DeepSeek arrive. The September 2026 points are anchored to August web-traffic measurements, ChatGPT between 51 and 56 percent, Gemini around 26, Claude around 10, from Similarweb and First Page Sage. The other boards lean on what companies publish: Seedance passes a billion renminbi a month in June 2026 against a little over 850 million for Kling over the whole quarter, which is why ByteDance leads the video board; Higgsfield reaches 700 million dollars annualised in July 2026, fal about 400 million in February. Every source is linked under the chart it feeds.

That is the caveat, and it is not small. A chart of estimates that moves is more honest than a chart of estimates that stands still, because motion is what the estimates are actually about. It is still a chart of estimates. Read it as a shape, not as a figure.