The Moment the AI Market Shifted
For the first time since large language models entered the corporate mainstream, more American businesses are paying for Anthropic’s Claude than for OpenAI’s ChatGPT. The Ramp AI Index, released in May 2026, captured a watershed moment: Anthropic’s business adoption climbed 3.8 percentage points to 34.4 percent of U.S. companies, while OpenAI slipped 2.9 points to 32.3 percent. Overall AI adoption among businesses inched up just 0.2 points to reach 50.6 percent.
Ramp, a corporate card and finance automation platform, tracks spending across more than 50,000 businesses. Its data offers a rare real-time window into which AI companies are actually converting interest into invoices. The crossover was not a fluke. It capped a yearlong surge that few industry watchers predicted. Anthropic has quadrupled its anthropic business adoption over the past twelve months. OpenAI grew its business adoption by only 0.3 percent during the same stretch.
The milestone carries symbolic weight, but the same report that crowns a new market leader also warns of fragility. Anthropic’s position faces three serious threats that could reshape the competitive landscape again.
How Anthropic Overtook a Dominant Rival
To appreciate what happened, consider where the two companies stood just one year earlier. In April 2025, OpenAI commanded roughly 32 percent of business AI adoption according to Ramp’s underlying data. Anthropic sat at under 8 percent. OpenAI had built an early, commanding lead as the consumer default. ChatGPT was where most people first encountered generative AI, and that familiarity carried into corporate purchasing decisions.
Anthropic took a different path. The company resonated early with engineers, AI evangelists, and the technical vanguard inside organizations. As Ramp lead economist Ara Kharazian noted in the March 2026 edition of the index, Anthropic leveraged that early-adopter base to go mainstream. By February 2026, Anthropic was winning about 70 percent of head-to-head matchups against OpenAI among businesses purchasing AI services for the first time. That represented a complete reversal from 2025 trends.
The trajectory shows up clearly in Ramp’s figures. Anthropic climbed from 0.03 percent of businesses in June 2023 to 7.94 percent by April 2025, then rocketed to 34.44 percent by April 2026. OpenAI peaked near 36.5 percent in mid-2025 and has been slowly declining since.
The engine behind much of this growth is a single product: Claude Code, the company’s agentic AI coding tool. It has become the fastest-growing product in Anthropic’s history. A recent analysis estimated that 4 percent of all GitHub public commits were authored by Claude Code, double the percentage from just one month prior. For context, that means roughly one in every twenty-five public commits on GitHub originates from an Anthropic product.
Business Insider reported in April 2026 that the crossover was imminent. A Ramp spokesperson told the outlet that “at the current pace, Anthropic is on track to surpass OpenAI within the next two months,” noting that it already led “among early adopters, including VC-backed companies, and in key sectors like software, finance, and professional services.” That prediction proved accurate almost to the day.
Separate survey data underscores how deeply AI has embedded itself into American economic life. For the first time in Gallup’s measurement, half of employed American adults say they use AI in their role at least a few times a year, up from 46 percent the prior quarter. Frequent use is also rising, with 13 percent of employees now using AI daily and 28 percent reporting they use it a few times a week or more. Those numbers come from a February 2026 Gallup survey of 23,717 U.S. employees.
3 Threats Ahead for Anthropic
But the Ramp report that celebrates Anthropic’s lead also flags three significant headwinds. Each could slow or reverse the company’s momentum, and each is baked into the business model that fueled its rise.
1. Escalating Compute Costs
Running large language models at scale requires immense computing infrastructure. Anthropic trains and deploys frontier models on vast clusters of specialized hardware, largely supplied by partners and cloud providers. These costs are not fixed. They rise with every new model release, every expansion in user count, and every increase in context window size. Anthropic’s rapid anthropic business adoption means more users, more API calls, and more inference compute. That virtuous cycle of adoption driving revenue also drives spending on GPUs, networking, data center power, and cooling.
Industry analysts estimate that training a single frontier model can cost hundreds of millions of dollars when factoring in compute, data acquisition, and engineering salaries. Inference costs for heavily used models can exceed training costs within months. For Anthropic, whose business user base has quadrupled in a year, the inference bill has likely ballooned faster than revenue from those users. Margins tighten as compute demand outpaces pricing adjustments. If Anthropic cannot negotiate favorable compute pricing or achieve efficiencies that reduce per-token costs, the economics of serving its growing customer base will erode.
The company has explored custom silicon and optimized inference stacks, but it operates at a structural disadvantage compared to competitors like Google, which manufactures its own TPUs, or Microsoft, which has deep alliances with GPU suppliers. Anthropic must buy compute on the open market or through partnership agreements, leaving it exposed to supply constraints and price increases. Any disruption in chip supply, whether from geopolitical tensions, manufacturing delays, or allocation shifts by cloud providers, could directly throttle Anthropic’s ability to serve existing customers or onboard new ones.
Furthermore, the competitive pressure to release more capable models compels Anthropic to invest in ever-larger training runs. Each generation of model demands more FLOPs than the last. If compute costs continue to rise faster than per-customer revenue, Anthropic may need to raise prices, which could slow adoption among price-sensitive businesses. The very growth that made Anthropic the market leader could become the thing that makes it harder to sustain that lead.
2. Compute Constraints and Capacity Bottlenecks
Beyond pure cost lies a more urgent problem: availability. The global market for high-end AI accelerators remains strained. Demand from every major AI lab, cloud provider, and enterprise buyer far outstrips supply. Even well-capitalized companies face allocation limits and wait times for hardware. Anthropic, as a relatively younger company without a captive supply chain, must compete for access alongside rivals that have deeper pockets or vertical integration.
This constraint manifests in several ways. First, it limits how quickly Anthropic can scale training infrastructure. If the company wants to train its next-generation model on a larger cluster, it must secure those GPUs months or years in advance, often paying deposits before receiving hardware. Second, compute constraints affect inference capacity during peak usage. When a popular product like Claude Code drives a surge in API calls, Anthropic must either throttle access, degrade performance, or invest in additional inference infrastructure that may not be immediately available. Any degradation in latency or reliability risks alienating the business customers that just made Anthropic the market leader.
Ramp’s data shows that Anthropic’s adoption is concentrated among early-adopter companies and VC-backed startups in software, finance, and professional services. These are precisely the customers with the lowest tolerance for slowdowns or service interruptions. A software startup that integrates Claude Code into its CI/CD pipeline expects consistent response times. A financial services firm using Claude for document analysis cannot afford processing delays during market hours.
You may also enjoy reading: Duracell Taps Driivz to Power Fast EV Charger Network.
The constraint also creates strategic vulnerability. If Anthropic cannot secure enough compute to support rapid growth, it may need to prioritize certain customers or use cases, effectively rationing capacity. That could slow the pace of anthropic business adoption and create openings for competitors. OpenAI, with its deep partnership with Microsoft and access to Azure’s sprawling infrastructure, may have more flexibility to absorb demand spikes. Google, which builds its own chips and operates its own cloud, has even more control. Anthropic, by contrast, depends on external providers for the majority of its compute. Any bottleneck at those providers becomes Anthropic’s bottleneck.
Mitigation strategies exist. Anthropic could invest in chip design, form exclusive capacity reservation agreements, or build redundancy across multiple cloud regions. But these moves require significant capital and long lead times. In the short to medium term, compute constraints remain a ceiling on how fast Anthropic can grow its business user base.
3. Token-Based Pricing Vulnerabilities
Anthropic, like most LLM providers, charges customers based on token consumption. A token is roughly four characters of text in English. The company sets per-token rates for input and output, with output tokens typically costing more. This model has fueled Anthropic’s revenue growth because it aligns pricing directly with usage. As businesses adopt Claude for more tasks, they consume more tokens and generate more revenue. During the past year of explosive anthropic business adoption, token-based pricing worked beautifully for Anthropic’s top line.
But the same model creates structural vulnerabilities. Token-based pricing is opaque to customers. Most business buyers do not know how many tokens their workflows consume or how changes in prompt length, context window, or output verbosity affect their bills. Surprise charges erode trust. As businesses mature in their AI usage, they begin to scrutinize costs more carefully. Finance teams that approved small-scale experiments last year are now seeing monthly AI line items in the thousands or tens of thousands of dollars. That scrutiny often leads to requests for flat-rate pricing, usage caps, or hybrid models that offer predictability.
Anthropic competes for enterprise accounts against providers that offer alternative pricing structures. OpenAI has experimented with consumption tiers, subscription models, and capacity reservations. Google offers per-seat pricing for some products. If Anthropic sticks rigidly to pure token-based pricing, it may lose deals to competitors that offer cost predictability. The company has introduced some options, such as batch API pricing and committed throughput plans, but its default model remains consumption-based.
Another risk is that token-based pricing incentivizes Anthropic to maximize usage, which aligns with revenue but may not align with customer value. When a model becomes more efficient, producing the same output with fewer tokens, Anthropic’s revenue from that output drops unless usage volume increases to compensate. The company therefore has a financial incentive to encourage longer outputs, larger context windows, and more frequent API calls. Customers, meanwhile, want concise, accurate answers delivered cheaply. This tension grows as businesses adopt AI for high-volume, cost-sensitive applications like customer support, content generation, and data processing.
Furthermore, competitors that offer specialized models with higher efficiency per token could undercut Anthropic on price. If a rival delivers similar accuracy at half the token cost, businesses will switch for economic reasons alone. Anthropic must continuously improve model efficiency to maintain its pricing competitiveness. That requires ongoing research investment, which adds to the compute cost problem described earlier.
Finally, token-based pricing creates unpredictable revenue for businesses that budget quarterly. Finance teams dislike uncertainty. As AI spending grows, CFOs will push for fixed-cost arrangements. Anthropic’s ability to offer those arrangements while maintaining profitability depends on the company’s confidence in its own cost structure. If compute costs are volatile, offering fixed pricing becomes risky. The company may need to build sophisticated financial hedging models or sacrifice margin to retain accounts.
A New Leader, but an Uncertain Throne
The Ramp AI Index offers one of the clearest available snapshots of real business AI adoption. Its methodology counts corporate card and invoice payments for AI products, meaning it measures actual purchasing behavior rather than survey sentiment. That makes the crossover a concrete signal. Anthropic now leads in paid business usage. Claude Code is the fastest-growing product in Anthropic’s history, driving adoption among software developers and technical teams who in turn influence broader organizational purchasing.
Yet the Gallup data provides a sobering counterpoint. Only about one in ten employees in AI-adopting organizations strongly agree that AI has transformed how work gets done. CEO studies across the United States, the United Kingdom, Germany, and Australia show minimal broad productivity effects from AI over three years. Adoption is up, but transformation remains elusive. That gap between deployment and impact creates risk for every AI vendor, Anthropic included. If businesses fail to realize meaningful returns on their AI investments, they may cut spending or consolidate vendors.
Anthropic has accomplished something remarkable. It went from zero measurable business adoption in mid-2023 to the market leader by early 2026. Its products solve real problems for developers, analysts, and knowledge workers. But the three threats outlined here — escalating compute costs, compute constraints, and token-based pricing vulnerabilities — are not hypothetical. They are structural features of the business model that produced this success. How Anthropic navigates them will determine whether its lead is a lasting shift or a temporary peak in a still-early market.
