OpenAI Claws Back Ground Against Anthropic in Corporate Spending Race

The Swing Vote Problem

OpenAI has begun closing the gap with Anthropic among American business buyers, according to payment data from Ramp, the corporate card and expense platform. The shift marks a modest comeback for the ChatGPT maker, which lost its lead in this segment earlier this year and has yet to reclaim the top spot - but appears to be gaining momentum.

At DailyTechWire, we've tracked the enterprise AI spend story closely, and the latest numbers underscore a deeper truth: businesses remain remarkably willing to switch providers, a pattern that complicates the narrative around defensibility and customer retention in frontier models.

Ramp's data covers more than 70,000 U.S. companies spending billions through its bill-pay and card products. The customer base skews toward tech and growth-stage firms, the exact cohort racing to integrate large language models into their operations. In May, Anthropic captured 41% of spend share among these users, edging past OpenAI's 39%. By July, Anthropic had widened its advantage to nearly 44%, while OpenAI held close to 40%.

Yet the most recent weeks tell a different story. According to Ramp economist Ara Kharazian, OpenAI is now growing faster than Anthropic in the third quarter to date - though a full month remains, and momentum can reverse quickly in a market that moves at this velocity.

Why Enterprises Keep Switching

The churn between the two labs reflects a combination of model performance, pricing pressure, and regulatory friction. OpenAI's GPT-5.6 Sol release has drawn developer interest, particularly for tasks that demand lower latency and broader applicability. Anthropic's Fable 5, by contrast, targets narrower, high-stakes use cases and comes with a steeper price tag - alongside a 30-day data retention mandate that triggered pushback from privacy-conscious enterprises.

That retention requirement stems from regulatory frameworks in certain jurisdictions, but it has forced Anthropic into an awkward position: offering a premium-tier model that imposes constraints many businesses would rather avoid. The result has been softer adoption than the company likely anticipated, even as Fable delivers technical improvements over earlier iterations.

Still, the volatility cuts both ways. The same businesses that migrated to Anthropic in the spring are now drifting back toward OpenAI as new releases shift the performance calculus. This suggests that lock-in remains weak, a concern for investors betting on recurring enterprise revenue and expanding gross margins.

Market Expansion Masks the Churn

One encouraging signal for both companies: the total addressable market among Ramp's customer base continues to expand. The share of firms paying for AI services crossed 50% in March and reached nearly 56% by July. That growth means OpenAI and Anthropic can both increase absolute revenue even as they trade market share points.

But the ease with which customers move between platforms raises questions about the durability of enterprise relationships in this category. Traditional SaaS vendors have long relied on integration depth, workflow embeddedness, and switching costs to retain customers. Frontier model providers, by contrast, often compete on API compatibility and benchmark performance - attributes that make it trivially easy for a developer to swap one model endpoint for another.

The implication is that both labs will need to invest heavily in tooling, customer success infrastructure, and vertical-specific solutions if they hope to build the kind of stickiness that commands premium multiples at IPO. Raw model capability, while necessary, may not be sufficient to sustain leadership in a market where competitors release new versions every few months.

What the Data Doesn't Capture

Ramp's figures offer a valuable window into mid-market and growth-stage behavior, but they exclude large enterprises that use spend-management platforms from providers like American Express or Coupa. Those buyers - Fortune 500 firms, global banks, healthcare systems - often negotiate custom contracts with model providers, and their spending patterns may diverge significantly from the Ramp cohort.

Additionally, Ramp shared only percentage data, not absolute dollar figures, which limits our ability to assess revenue scale or average contract value. A shift of a few percentage points could represent millions of dollars in monthly spend, or it could reflect a handful of mid-sized customers reallocating budgets. Without granular breakdowns, the data serves best as a directional indicator rather than a precise financial measure.

Still, the trend is clear enough to matter. If businesses continue to treat frontier models as interchangeable commodities, both OpenAI and Anthropic will face margin pressure and heightened customer acquisition costs. The winner in that scenario may be the lab that can move fastest up the stack - building application-layer products, vertical AI agents, or embedded solutions that make switching painful.

The IPO Clock and the Loyalty Question

Both companies are widely expected to pursue public listings in the next 18 to 24 months, and prospective investors will scrutinize customer retention metrics closely. The back-and-forth swings visible in Ramp's data will likely surface in S-1 filings as net revenue retention rates, logo churn, and cohort analysis - metrics that reveal how well each company retains and expands within its customer base.

For now, the data suggests that neither lab has solved the loyalty problem. Enterprises are opportunistic, responsive to new releases, and sensitive to pricing and compliance friction. That dynamism benefits the market as a whole - competition drives innovation - but it complicates the path to predictable, high-margin growth.

At DailyTechWire, we expect this pattern to persist until one of two things happens: either a breakout application layer emerges that locks users into a specific model provider, or the performance gap between frontier labs narrows to the point where differentiation becomes difficult. In either scenario, the current volatility will give way to a more stable competitive structure - but we're not there yet.