What the samples actually measure
OpenRouter's data is the most cited and the least representative at the same time. The company says its own customer base leans toward startups that build on AI as their core business, mixed with larger tech companies and some big enterprises. That is a real economy, just not the whole economy. Anthropic's early-year 75 percent was measured inside that same skewed frame, which makes the swing from 75 to 50 a genuine trend inside OpenRouter's world, not a census of global enterprise spend.
Ramp's data cuts the other way. Its customers include Fortune 500 firms, which is closer to the procurement offices that actually matter. But Ramp's headline, the first OpenAI lead since December, had a half-life of days. A lead measured in card swipes across a single week is a momentum signal, not a position.
Read them together and the honest picture is this: OpenAI closed a massive gap in nine months, and the gap is still closing, but “neck and neck” overstates it the way “dominant” once overstated Anthropic's lead.
Price did the work
Nothing in either dataset suggests companies switched because GPT-5.6 out-reasons Claude. The trigger was arithmetic. An 80 percent cut on the cheapest tier changed the bid sheet for every company running inference at scale, and OpenAI kept shipping follow-up versions tuned for price-to-performance. Claude 5.5's late-September launch was Anthropic acknowledging the battlefield had moved from capability to cost.
The secondary drivers matter too. The Journal reports that Fable5's launch, bundled with a 30-day user-data retention policy for trust and safety, actively pushed companies toward OpenAI. Retention policy is an underrated churn driver in enterprise: it turns procurement lawyers into the decision-makers, and they do not negotiate on default data retention.
The practical read
David Joo, CEO of AI sales platform Reevo, told the Journal his company is moving workloads from Anthropic to OpenAI to cut costs and reduce dependence on a single vendor, adding that the preferred model “could change again.”
That is the paragraph worth keeping. Enterprise buyers are now multi-sourcing by default and re-benchmarking quarterly, which means the enterprise lead is a rented position, not an owned one. The buying question has shifted from which model is most capable to which delivers the most performance per dollar at the volume you actually run.
Why this matters beyond the two labs
Anthropic's IPO case, if the reported November timing holds, rests on enterprise revenue durability. OpenAI's next-year case rests on showing its enterprise business is more than a consumer halo effect. Both need this number. A spend-share flip that took nine months on price alone tells IPO investors that the enterprise AI market has no incumbent protection: every contract is contestable, and the contest is fought with the discount schedule.
One caution from the fine print. A share measured among companies that use both labs is a share of the engaged middle of the market. The companies using only one lab, or neither, are not in the denominator. When the S-1s arrive, compare these percentages against total cloud AI spend before deciding who won the year.
Sources
- [1] DigitalToday — “OpenAI vs Anthropic price war intensifies, emerging as major wild card for enterprise AI” (Oct 10, 2026), reporting the Wall Street JournalRead source
- [2] Gizmodo — “Report Suggests OpenAI Has Clawed Tons of Market Share Back From Anthropic in 2026” (Oct 8, 2026), on OpenRouter data seen by the Wall Street JournalRead source
- [3] JKN — “Companies Grapple with Rising Costs as OpenAI's Aggressive Pricing Challenges Anthropic's Dominance” (Oct 7, 2026), on the Wall Street Journal reportRead source