OpenAI Closing the Gap on Anthropic for Business Spending

Corporate expense platform Ramp has released data showing that Anthropic now holds nearly 44% of AI spending share among U.S. businesses on its platform, compared to OpenAI's nearly 40% as of July 2026. OpenAI lost its long-held lead in May, when Anthropic first edged ahead. The dataset covers more than 70,000 American companies. OpenAI is currently growing faster in the third quarter, but with a month still remaining, the gap could swing either way again.

What happened

Data point	Detail
Source	Ramp corporate card and bill pay platform
Businesses covered	More than 70,000 U.S. companies
Anthropic share (July 2026)	Nearly 44%
OpenAI share (July 2026)	Nearly 40%
When OpenAI lost the lead	May 2026 (Anthropic 41%, OpenAI 39%)
AI adoption among Ramp customers	Topped 50% in March, reached ~56% by July

Ramp, a corporate credit card and expense management company popular in Silicon Valley, publishes periodic snapshots of what its customers are actually paying for. Its latest data, shared by Ramp economist Ara Kharazian, shows OpenAI held the top spot among business AI spenders for a long time before Anthropic overtook it in May 2026. As of July, OpenAI has not closed the gap in overall share, but Kharazian says OpenAI is growing faster within this segment in Q3 so far.

One caveat worth noting: Ramp’s numbers skew toward tech-industry companies and exclude large enterprises that use spend-management tools from providers like American Express. The figures represent percentage of customers, not total dollars spent. Ramp declined to share the actual dollar amounts.

What is driving the shift between providers?

Kharazian pointed to two specific model releases on X. He credited OpenAI’s GPT-5.6 Sol as “really good” and “increasingly the choice for developers.” On the Anthropic side, he said Fable 5, Anthropic’s higher-end model tier, “disappointed both in adoption and real-world application” because of its price and a data retention requirement imposed by regulators.

Anthropic warned Fable users that it must retain their data for 30 days, which caused notable backlash. Fable is priced at a premium and targets specific professional use cases rather than general chat, so the data retention policy hit a user base that tends to be more sensitive about confidentiality. The article in our news section covering OpenAI’s private safety processing versus Anthropic’s data retention approach covers the background on these competing policies in more detail.

Why it matters

The headline story here is not which AI lab is “winning.” It is how easily businesses are switching. A single model release, or a single policy change like a data retention requirement, is enough to move market share by several percentage points within weeks. That kind of volatility is a real concern for investors in both companies ahead of their anticipated IPOs.

The more encouraging finding is the overall growth in paid AI adoption. Among Ramp’s 70,000-plus customers, the share paying for AI crossed 50% in March 2026 and climbed to nearly 56% by July. That means both OpenAI and Anthropic can grow revenue even while fighting over the same pool of customers, because the pool itself keeps expanding. Businesses that were not paying for any AI six months ago are now signing up.

For teams evaluating AI integration for their own workflows, this data reinforces something practitioners already know: model quality and data handling policies matter more than brand loyalty. The business that had Claude locked in as their standard tool in April may have quietly started routing developer tasks to GPT-5.6 Sol by August.

Our take

Ramp’s data is useful precisely because it is transactional, not a survey. People voted with their company cards. That said, the sample has real limits: tech-heavy, Ramp-using companies are not the median American business. A law firm in Ohio or a manufacturing shop in Tennessee is not in this dataset.

Still, the switching behavior is the real signal here. Enterprise software is supposed to be sticky because of integrations, training, and procurement inertia. AI subscriptions, at least at this stage, are not behaving that way. Teams are treating frontier models more like SaaS tools with monthly billing and less like infrastructure contracts. That is good for buyers, because competition keeps providers honest, and bad for anyone projecting stable ARR multiples ahead of an IPO.

If you are a business operator, the practical implication is: do not over-engineer your AI stack around a single provider. Keep your prompts and workflows portable. The model you standardize on today may be the runner-up by next quarter. We have seen exactly this pattern across client projects we have shipped, where a provider chosen for one capability becomes a bottleneck when another lab releases something better at half the cost.

What to do about it

1. Audit which AI providers your team is currently paying for and whether usage justifies each subscription.
2. Store your system prompts and workflows in a format that is not locked to one provider’s API (plain text or a version-controlled repo works fine).
3. Check Anthropic’s data retention terms for Fable before routing any client or sensitive data through it.
4. Set a quarterly review date to compare model performance on your specific tasks, not just benchmark scores.

The market is still expanding fast. Pick tools that let you move when it does.

Source: TechCrunch · AI