Anthropic Overtakes OpenAI: RAMP Data Reveals Structural Shifts in the Enterprise AI Market and the Rise of Math AI Startups

In the spring of 2026, the dynamics of the enterprise AI market are shifting quietly but decisively. Ara Karzai, Chief Economist at the corporate payment and spend management platform RAMP, appeared on CNBC to explain the AI market landscape based on real-time corporate spending data. Also appearing on the program was 24-year-old Karina Hong, a Stanford PhD dropout who founded the math-specialized AI startup Axiom. She discussed the journey to raising $200 million in a Series A round at a $1.6 billion valuation, as well as the significance of their approach to "solving math with AI."

1. The Rise of Anthropic as Shown by RAMP Data—The Reality of "One in Four Companies"

1-1. A Sudden Shift: 25x Adoption Rate in One Year

According to the RAMP AI Index, as of 2026, one in four companies on the RAMP platform is paying for Anthropic.A year ago, that figure was one in 25. Karzai points out that the speed of change indicated by this number is almost unthinkable in the typical enterprise software market.

Throughout 2025, the number of AI-adopting companies grew steadily, and currently, nearly half of all U.S. companies pay for some form of AI service. However, more than 50% have yet to adopt it, and "who captures that market" will be the focus of future competition.

1-2. Anthropic Wins 70% in Head-to-Head Battles—The Turning Point Was January 2026

The most notable data point is that in head-to-head matchups between Anthropic and OpenAI, 70% of companies are currently choosing Anthropic.This reversal occurred in January 2026. The primary drivers are considered to be the release of Claude Code and Anthropic's expansion from "developers and engineers" to "non-technical business users."

Karzai analyzed, "Through the agent Claude Cowork, Anthropic has enabled general business users who cannot write code to interact with PC files and business systems. This has accelerated enterprise adoption."

2. OpenAI's "Worst Month"—The Meaning of a 1.5% Market Share Drop

2-1. Defeated in the Battle for New Customers

In the RAMP Index, OpenAI recorded its largest share decline (a 1.5% drop) during this reporting period.However, Karzai analyzes this not as "existing customer churn" but as "weakness in new customer acquisition." The structure is not that existing OpenAI users are switching en masse, but that the proportion of companies adopting AI for the first time that choose Anthropic is surging.

Anthropic's models are, on average, more expensive than OpenAI's, and there are rate limits due to excess demand. The fact that they are still being chosen demonstrates a clear preference among engineers and users for model quality over price competition.

2-2. The Brand Effect Created by the Department of Defense's "Ban"

It is worth noting the paradoxical development where the U.S. Department of Defense's attempt to effectively blacklist Anthropic as a "supply chain risk" actually increased Anthropic's brand awareness.Following this controversy, Anthropic's Claude reached #1 on the App Store, and paid subscribers doubled.

Karzai maintains a cautious stance, stating, "In the data, it is still too early to judge whether this event is a net positive or negative." On the other hand, she noted, "Engineers and early adopters within companies have a strong preference for Anthropic's tools. Because they act as advocates (evangelists) within their companies, it is not easy to cancel contracts at the enterprise level."

3. Anthropic's Next Growth Trigger—Collaboration with Private Equity

3-1. The Significance of Reported Blackstone Partnership

The partnership negotiations between Blackstone and Anthropic, reported by The Information, are not yet an official announcement, but if realized, they could change the distribution structure of the AI market.

Karzai pointed out, "AI adoption rates already differ between VC-backed companies and PE-backed companies. If PE firms push their portfolio companies to adopt Anthropic, this becomes a powerful distribution channel that drives adoption from the top down." Even now, AI adoption rates at PE-owned companies are higher than at general companies, which could lead to penetration into industries where AI has traditionally been difficult to enter, such as large retail chains and hospital networks.

3-2. A Turning Point Where "Economic Productivity Gains" Become Visible

Karzai is focused on whether the productivity effects of AI will appear in economic statistics. "The current benefits are mostly concentrated in software engineering workflows. However, many workers are not in tech companies, but in industries like retail, healthcare, and manufacturing. Only when AI penetrates those areas will we see macro productivity improvements." His view is that the catalyst for this will be the top-down deployment by PE firms.

4. The "Open Claw" Phenomenon—Enterprise Adoption of AI Agents Has Only Just Begun

4-1. Many companies have tried and stopped

The program also mentioned "Open Claw"-style agent utilization based on Anthropic's MCP (Model Context Protocol), where AI accesses internal Slack, Google Docs, and files to process tasks. According to Karzai, many companies have tried this but have often stopped quickly. The reason is the fear of the "risk of making scalable mistakes."

"If AI takes over entire customer service operations, a single mistake could ripple across thousands of interactions. We haven't fully guaranteed the reliability of that yet," Karzai stated. This "trust problem" has become one of the biggest barriers to the full-scale deployment of enterprise AI.

5. Axiom—A 24-year-old challenges major labs with mathematical AI

5-1. "ChatGPT can provide the answer to a math problem, but it cannot prove why it is correct"

Karina Hong (24), who dropped out of her PhD program at Stanford to found Axiom, entered the AI market with a unique approach focused on the "rigor of mathematical reasoning." She raised $200 million in a Series A round, reaching a valuation of $1.6 billion. This is an exceptional speed, occurring just 7 to 8 months after founding.

The problem Axiom is trying to solve is simple. Current large language models can provide "answers" to math problems, but they cannot "prove" that those answers are correct. They are attempting to solve this by combining AI with a classical technique called Formal Verification.

5-2. Why is mathematical proof important?

As the amount of code generated by AI surges, the inability to prove that the code is "truly correct" is becoming a serious risk. Hong explains, "The more complex software becomes, the more systems there are that cannot be handled by statistical code generation alone. Formal verification fills that gap."

Formal verification is already used in Nvidia's GPU design, AWS, Boeing, Airbus, and European public transportation systems, but Axiom's goal is to automate and democratize this using AI. Competitors include mathematical AI projects from major labs like DeepMind and OpenAI, as well as Harmonic, which also specializes in mathematical AI.

5-3. Use of the $200 million and future product development

The $200 million raised will be used for compute costs and hiring. The most recent product is "Axel (Axiom Lean Engine)," which is provided for free to the community as a tool for verifying and manipulating formal proofs. Full-scale commercial products are planned for deployment to companies in hardware design, software verification, and financial systems as code verification tools.

Regarding an IPO, Hong stated clearly, "I'm not thinking about that yet. Right now is the building phase." She says that keeping the feedback loop between research and product running is the top priority at this moment.

Conclusion

The rise of Anthropic indicated by RAMP data shows that the preference for AI models has shifted from "default" to "explicit choice." The pattern where Anthropic is chosen despite higher prices and usage limits is creating a new trend where strong brand preference in the engineering community drives enterprise adoption. Meanwhile, the emergence of startups like Axiom that pursue "AI reliability and provability" is one of the signs that AI is moving to the next stage from "generation" to "verification." For investors, it is worth maintaining a perspective on companies building AI reliability infrastructure, alongside investing in companies that use AI.

Recommended Articles

Next Big Wave—Growth Stocks, Seeds of Ideas, and Deep Dives into Trends