Card data, not vendor decks: the enterprise AI split is unsettled, not settled

Ramp, the corporate card and expense management company, has published share figures indicating that OpenAI has started gaining on Anthropic among US business customers ^[1]. For anyone mid-way through a model vendor evaluation, the useful signal is not who leads but how fast the order changes, because neither lab has published financials and will not until they are close to their planned IPOs ^[19].

The sequence matters. OpenAI lost the lead among Ramp's paying business users in May, when Anthropic reached 41% share against OpenAI's 39% ^[2], and has not regained it since ^[3]. As of July, Anthropic held nearly 44% to OpenAI's nearly 40% ^[4]. Read the snapshots alone and Anthropic is doing better, not worse: the gap went from about two percentage points in May to about four in July, roughly double ^[5]. The narrowing story rests on something else, which is Ramp economist Ara Kharazian's statement that OpenAI is currently growing faster in this segment in Q3 to date than Anthropic ^[7]. Ramp declined to release dollar figures, sharing only percentages ^[11], so the growth-rate claim cannot be checked against spend. TechCrunch notes there was still a month left in the quarter when the data came out, and that the trend could shift again ^[8].

Know what the sample is before quoting it in a procurement memo. It covers more than 70,000 American businesses that spend billions through Ramp's bill pay and corporate card products ^[9], skewed toward the tech industry even though the customer base spans sectors ^[10]. It is not the total market and excludes large enterprises that run spend management through providers such as American Express ^[12]. The two labs together account for roughly 84% of this measured segment, leaving about 16% for everyone else ^[6].

Kharazian's explanation, posted on X, is that "GPT-5.6 Sol is really good, increasingly the choice for developers" while "Fable 5, meanwhile, disappointed both in adoption and real-world application given price + data retention requirements imposed by regulators" ^[13][14]. TechCrunch pushes back that this oversimplifies: Fable is Anthropic's higher-end tier, expensive but aimed at a narrower set of use cases than a general chatbot ^[15]. It also notes Anthropic drew criticism when it told Fable users it must retain their data for 30 days ^[16].

The other line in the data cuts against zero-sum framing. The share of Ramp customer companies paying for AI at all topped 50% in March and reached nearly 56% by July ^[17], about six points in four months ^[18], which Ramp's data suggests should let both vendors grow business revenue even while fighting over share ^[21]. TechCrunch's read is that buyers flop back and forth as each lab ships, volatility that should make investors question how sticky enterprise AI spending really is ^[20].

Watch whether OpenAI's faster Q3-to-date growth ^[7] actually closes any of the roughly four-point July gap ^[5] when the quarter ends, and whether Ramp ever attaches dollars to the percentages ^[11]. Until then, treat "Claude for work, GPT for consumers" as a snapshot with a two-month half-life, not an architecture decision.