For most of the last two years, the story about business AI spending was simple: OpenAI led, Anthropic chased. That flipped in May, and it has not flipped back.
New numbers from Ramp, the corporate card and expense company, track spending across more than 70,000 US businesses that use its cards and bill pay tools. As of July, close to 44% of these businesses paid for Anthropic's Claude models, compared to about 40% for OpenAI's ChatGPT and API products. Anthropic has held the top spot every month since May.
But the more useful number is the trend line, not the snapshot. OpenAI is growing faster than Anthropic again this quarter, according to Ramp economist Ara Kharazian. The reason traces back to one business decision Anthropic made in June, and it is a decision every company buying AI tools should pay attention to.
When Anthropic launched its top model, Fable 5, it also introduced a new rule: any business using its most capable models must let Anthropic store prompts and responses for 30 days, with no way to turn it off. Anthropic says this data only helps catch security attacks and is never used to train new models. That explanation did not stop the backlash. Companies in healthcare, finance, and legal services had built entire AI compliance programs around a promise that nothing gets stored. Microsoft restricted its own employees from using Fable 5 internally because the rule conflicted with its own privacy commitments, even while still offering the model to outside customers who opt in.
Meanwhile, OpenAI released GPT-5.6 Sol, a model built for coding and multi-step tasks, and cut prices on its cheaper tiers by as much as 80%. Developers responded. That is the kind of shift that moves market share fast in a market this young.
The bigger lesson here is not about which chatbot is better. It is that enterprise AI spending is not sticky yet. Businesses are willing to move their budgets between vendors within a single quarter based on price changes, privacy terms, and how well a model handles a specific task like writing code. That is unusual for enterprise software, where switching costs and long contracts normally keep customers locked in for years.
For any business currently picking an AI vendor, this is useful information twice over. First, do not assume the vendor you pick today will still be the best value in six months. Second, before adopting a company's most powerful or most expensive model, read the data handling terms closely. A model's benchmark score means nothing if its data policy creates a compliance problem for your industry.
Both companies are also heading toward public listings, with Anthropic reportedly targeting a debut as early as October and OpenAI possibly waiting until 2027. Investors evaluating those listings should treat this kind of month-to-month volatility as a real risk factor, not noise. A business that can lose a chunk of market share over a single policy change does not have the loyal customer base that usually justifies a trillion dollar valuation.