OpenAI may be regaining ground against Anthropic among paying business customers, according to new purchasing data from corporate expense platform Ramp. The figures suggest that enterprise adoption remains fluid as companies compare rapidly changing models, prices and data policies.
The dataset is useful but limited. Ramp tracks spending by more than 70,000 US businesses, with a customer base that leans toward technology companies. Its percentages show whether those businesses paid a provider, not how much they spent or how intensively they used the service. Large enterprises using other payment methods may also be absent.
Anthropic leads while OpenAI accelerates
Anthropic overtook OpenAI among AI vendors paid by Ramp customers in May, reaching 41% compared with OpenAI’s 39%. By July, Anthropic was used by nearly 44% of paying AI customers in the dataset, while OpenAI was near 40%.
That still leaves Anthropic ahead on the reported measure. However, Ramp’s data indicates that OpenAI has been growing faster during the third quarter to date, with one month remaining in the quarter when TechCrunch published its analysis. The trend may reflect renewed interest following a model release, but the source does not establish a single cause.
Ramp economist Ara Kharazian pointed to OpenAI’s GPT-5.6 Sol as a strong release for developers. He also said Anthropic’s Fable 5 had disappointed some users because of pricing and data-retention concerns. TechCrunch cautions that this interpretation may be too simple, as businesses weigh many factors and the dataset cannot reveal every purchasing decision.
The numbers therefore support a modest conclusion: more Ramp customers were paying Anthropic in July, while OpenAI’s share was rising faster recently. They do not prove that OpenAI has retaken the broader enterprise market or that model quality alone explains the movement.
Business buyers appear willing to switch
The close competition suggests limited lock-in at this stage of the market. Companies can test a new model, add another provider or move a workload when performance and economics change. That makes each major release a potential customer-acquisition event rather than a feature update that affects only existing users.
Switching is not frictionless. Enterprise teams still need to evaluate security, retention rules, integration effort and the reliability of model outputs. Yet many AI tools are accessed through APIs or general-purpose applications, which can make comparisons easier than replacing a deeply embedded business system.
Ramp’s metric can also count a company as a customer of more than one provider. The percentages should not be treated as shares of a fixed market that must total 100%. A business may pay OpenAI and Anthropic simultaneously for different teams or workloads, particularly while it evaluates new models.
That overlap matters when interpreting competitive gains. OpenAI’s increase does not necessarily mean Anthropic lost the same customers, and Anthropic’s lead does not imply exclusive relationships. The data is better read as evidence of expanding and overlapping adoption than as a winner-takes-all ranking.
The overall paid AI market is still expanding
Both companies can grow because more businesses are paying for AI. Among Ramp customers, paid AI penetration rose above 50% in March and reached nearly 56% in July. This broadening base gives providers room to add customers even while their relative positions change.
The sample’s technology bias may make that adoption rate higher than it is across the entire economy. Companies using American Express or other procurement systems may also have different vendor preferences. Ramp’s view is consequently a signal from a large cohort, not a complete census of enterprise AI spending.
Even with those caveats, the data challenges the idea that an early lead guarantees durable dominance. Business users are still experimenting, multiple providers can coexist inside one organisation, and model releases can shift purchasing momentum within months. Anthropic currently leads this particular measure, while OpenAI’s recent acceleration keeps the contest open.
