Ramp's July AI index: Anthropic's market share hit 43.5%, widening its lead over OpenAI; Fable 5 is only 6% of tokens businesses bought, likely due to high cost (Ara Kharazian/Ramp)
Dear Colleagues: Today's letter includes my monthly update of Ramp AI Index, our flagship research using spend data from Ramp to track how American businesses are using AI.
In July, Anthropic extended its lead in business AI adoption, with 43.5% of U.S. businesses paying for Anthropic subscriptions or tokens, up 1.1 percentage points from the previous month. XAI saw the fastest growth in adoption, rising to 4% of businesses. OpenAI, however, underperformed with a 0.23 percentage point increase, reaching 39.7% of businesses.
The share of businesses using model serving platforms, which provide access to open source models and some Chinese-developed models, rose again to 6.1%, an increase of 0.2 percentage points from the previous month. This growth has yet to significantly impact spending on OpenAI and Anthropic. However, adoption of OpenAI and Anthropic has slowed recently.
In July, Anthropic released Fable 5, the most performant model on the market but also the most expensive at around $10 per 1M tokens. Despite being the most expensive model, Fable 5 accounted for only 6% of tokens purchased from Anthropic, and 11.4% of dollars spent on Anthropic models. This suggests that businesses are not finding enough value in paying for Fable 5, making it less popular than OpenAI's GPT-5.6 Sol, which made up 25% of OpenAI token spend and 23% of its spend.
The slower adoption of Fable 5 raises concerns about the future of AI trade, as the performance of newer models may not justify their higher costs. To encourage businesses to adopt the latest models, companies will need to demonstrate performance improvements beyond what Fable 5 offers and ensure that competitors do not catch up quickly. Open source models are also gaining ground, closely following the advancements of these new models.
Written by urgent.news from Techmeme's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.