A Price War, Not Benchmarks, Flipped Enterprise AI Spending Back to OpenAI
OpenRouter data shows OpenAI's share of enterprise AI spending among dual-vendor businesses jumped from under 25% to nearly 50% in September 2026.
OpenAI has recovered about half the enterprise model spend it ceded to Anthropic this year, according to routing data from OpenRouter. Among the roughly 120,000 businesses that buy from both labs, OpenAI's share of enterprise AI spending climbed from under 25% at the start of 2026 to nearly 50% in September. The last time OpenAI held the weekly spending lead was the week of 26 February 2024.
The flip landed in the week of 7 to 13 September, when OpenRouter users sent more dollars to OpenAI models than to Anthropic's for the first time in more than two and a half years. Anthropic had controlled about three-quarters of the combined spend in January; by September the two labs were effectively level.
What moved the wallets was price and packaging rather than benchmark position. OpenAI's recovery tracks its pricing moves and the GPT-6 generation closely enough that the scoreboard between the two labs is now denominated in dollars spent. That reframes both companies' path to public markets as a contest over spend share.
How Enterprise AI Spending Flipped
OpenRouter routes developer and enterprise traffic across competing models, which makes its spend totals a live proxy for where API budgets land. Its weekly figures for 7 to 13 September put OpenAI just above half of the combined OpenAI-Anthropic total, the first weekly lead since early 2024.
The annual arc is steeper than the weekly one. Anthropic opened 2026 with roughly 75% of combined spend while OpenAI held less than a quarter. Nine months on, among companies that buy from both, the split is close to even.
| Metric | Start of 2026 | September 2026 |
|---|---|---|
| OpenAI share of combined spend | Under 25% | Nearly 50% |
| Anthropic share of combined spend | About 75% | Roughly even |
| Businesses buying from both labs | ~120,000 | ~120,000 |
A single week of spend can swing on one launch, one promotional rate or one large customer shifting a workload, so the monthly split is the more durable reading. OpenRouter publishes two views of the same market. Spend totals track dollars flowing to each lab; the platform's model ranking tracks how much traffic each model attracts. Those measures can point in different directions at the same moment, which is why a weekly crossover is a signal to verify rather than a settled verdict.
The two headline figures together describe the year: OpenAI erased a three-to-one deficit inside three quarters. The accounts driving that swing are the dual-sourcing customers, the ones large enough to run both vendors side by side and reallocate traffic as prices change.
How Anthropic Built the Lead in the First Place
The gap OpenAI just closed was recent, not structural. Ramp's business-spend tracking put only about 9% of US companies paying for Anthropic products in May 2025. A year later Anthropic had overtaken OpenAI on that measure for the first time, at 34.4% against 32.3%.
Coding and productivity workloads carried that rise. Anthropic built its enterprise position on developer tools where its models held an edge, and those are the same workloads where OpenRouter data now shows OpenAI making up ground. The lead has changed hands quickly in both directions.
What Actually Moved the Money
Discounting did most of the work. Anthropic's enterprise API contracts carry volume discounts of 30% to 40% below list pricing, which had let it undercut OpenAI on comparable capability. OpenAI's answer, tied to the GPT-6 family, reset the effective cost of a premium workload.
Two models carry the shift. GPT-6 Astra is winning premium workloads while a cheaper tier, Luna, drives volume. Ramp's corporate expense data, drawn from more than 70,000 US businesses, puts Astra at about 13% of tracked enterprise AI spending against roughly 8% for Anthropic's Claude Fable.
Subscription counts at Ramp move more slowly. Anthropic still leads there, with 43.5% of companies buying its models against OpenAI's 39.7%, based on Ramp's August figures. The gap between the two datasets carries information: Anthropic holds the broader customer base while OpenAI pulls more dollars per account from its heaviest users.
| Ramp metric (US businesses) | Anthropic | OpenAI |
|---|---|---|
| Companies buying models | 43.5% | 39.7% |
| Share of tracked spend, flagship model | ~8% (Claude Fable) | ~13% (GPT-6 Astra) |
Those are different measures of one market. OpenRouter counts API consumption by developers and product teams worldwide; Ramp counts card spend and subscriptions at US companies. Neither settles the question alone, and the two disagree on magnitude even where they agree on direction.
Anthropic is not standing still on the commercial side either. The company is weighing a new model to answer the GPT-6 generation, which would put the two labs back into a capability race on top of the price contest already running.
The Cheaper-Model Squeeze Underneath Both
OpenAI and Anthropic are competing hard for a premium tier that is itself under pressure. Open-weight models, many of them Chinese, run 60% to 90% cheaper than the leading systems from either lab, with self-hosted inference priced between $0.17 and $1.00 per million tokens against $5 to $15 for commercial APIs.
Token share shows the drift. In June 2025, models from Google, OpenAI and Anthropic together held about 70% of token share on OpenRouter; by June 2026 that had fallen to roughly 30%. A price war between two US labs is partly a defensive response to a cheaper tier neither one controls.
That context makes OpenAI's gain narrower than the headline suggests. Winning the larger half of a two-way split is a real milestone, but the split covers a smaller slice of total model consumption than it did a year earlier.
Why the IPO Frame Now Hinges on Spend Share
Both labs are moving toward public listings, and revenue durability will carry more weight than capability claims in a prospectus. Spend share among dual-sourcing enterprises is the best available churn proxy for both companies, because it shows whether a customer keeps paying after a model release stops being new.
OpenAI's swing from under 25% to nearly 50% is the strongest evidence yet that price decides which lab keeps an enterprise account. Anthropic's 30% to 40% enterprise discounts, and its decision to weigh a new model against the GPT-6 generation, point the same way: the response is commercial as much as technical.
Margins are where this gets uncomfortable. Deep enterprise discounts on both sides compress the revenue each lab books per token, and a roughly even split means neither can claim the pricing power that would justify premium multiples at listing.
The verdict for buyers is leverage. Two vendors trading the lead on price give procurement teams a reason to keep both contracts live and to renegotiate at renewal, since switching costs are lower than the headline gap between the labs suggests.
Why this matters
Enterprise AI spending is consolidating around a price contest between two labs whose models are close enough on capability that cost decides the routing. That is good news for buyers negotiating renewals and harder news for both vendors' margins as they approach public markets. The number to watch is whether OpenAI holds its half of the split once the current discounts stop being the differentiator.
Photo by Brecht Corbeel on Unsplash
Related Articles
- Chinese AI Models Now Handle 46% of US Enterprise Tokens on OpenRouter as Cost Gap Drives Migration
- OpenAI enterprise revenue overtakes consumer, ROI data lags
- DeepSeek V4 Pro Adoption Surges as US Firms Pivot to Low-Cost AI Models
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.