OpenAI's 90 Percent Discounts Are Pulling Anthropic's Biggest Customers

OpenAI is offering discounts of up to 90 percent to poach Anthropic's largest customers, and the strategy is working, according to a report by The Information. Companies including Rippling, Replit and DataCamp are shifting more of their AI spending to OpenAI, in a price war that reveals how aggressively the market leader will trade margin for market share.

How much are customers actually switching?

The most concrete number comes from Rippling. CEO Parker Conrad said his company boosted OpenAI's share of its coding workload from 16 percent to 52 percent. Over the same period, Anthropic's share fell from 57 percent to 18 percent.

That is a near-total inversion of a major customer's vendor mix in a segment — coding assistance — where Anthropic had built a strong position. Rippling, which sells human resources and IT software, treats AI coding workloads as a significant and measurable share of its internal operations, which makes Conrad's figures a useful proxy for what other enterprise buyers may be doing.

Replit made its move for a different reason: price. The software company switched after OpenAI offered more than 30 percent off a contract worth over $100 million a year. A 30 percent cut on a nine-figure annual commitment represents tens of millions of dollars in annual savings for a single customer, and it signals the scale of the deals OpenAI is now willing to discount.

DataCamp's experience points to an even deeper cut. CEO Jonathan Cornelissen said OpenAI's pricing reduced his costs by 80 to 90 percent. For an education technology company that builds AI learning products, that magnitude of reduction changes the unit economics of what DataCamp can afford to ship.

What did the bank deal look like?

The Information's report also details a deal with a US bank holding more than $7 billion in assets. The bank had its ChatGPT seats doubled to 500 employees at no extra charge.

The per-user math is stark. The bank now pays under $7 per user instead of the usual $20 or more — a discount of roughly two-thirds on a standard seat price, delivered simply by doubling the licensed headcount. For OpenAI, the giveaway costs little marginal revenue if those seats would otherwise have gone unused, while locking a regulated financial institution deeper into its ecosystem.

Enterprise seat pricing has become a key battleground in AI sales. A bank is a sticky customer: procurement, security review and compliance processes make switching vendors expensive, so winning 500 seats today has value well beyond the discounted first year.

Why can OpenAI afford this?

Two forces are at work, and only one of them is a subsidy.

The first is straightforward share buying. OpenAI is partly purchasing market share at the expense of its own margins, as The Information reports. The discounts to Replit, DataCamp and the bank are loss-leader economics: accept lower or negative margin on marquee accounts now to deny a fast-growing rival its reference customers and its revenue base.

The second force is real cost reduction. OpenAI has achieved genuine price cuts through its Sol and Luna model lines, thanks to more efficient models and more efficient infrastructure. This distinction matters for the market. If discounts were pure subsidy, competitors could wait them out. If the underlying cost of serving each token is falling, then list prices across the industry have room to fall permanently, and every vendor's revenue projections shrink accordingly.

Sol and Luna are OpenAI's efficiency-focused model families, and their role in this story suggests OpenAI intends to convert architectural and infrastructural efficiency into list-price pressure on rivals.

What is happening to Anthropic?

The report's most consequential datapoint for Anthropic is its growth trajectory. According to an investor memo cited by The Information, Anthropic's growth slowed from the second quarter to the third quarter.

A slowdown is not a decline, but timing is everything. Anthropic had been one of the fastest-growing AI companies, and coding — the very workload where Rippling cut Anthropic's share from 57 percent to 18 percent — has been a core strength. Losing share in your flagship category while your growth rate decelerates is precisely the combination investors punish.

Anthropic now faces pressure on two fronts simultaneously:

• OpenAI's pricing offensive, which is converting Anthropic's established enterprise accounts with discounts of 30 to 90 percent.
• Chinese model pricing, which resets the price floor for everyone.

How does China change the equation?

Both OpenAI and Anthropic face growing price pressure from China, where new models by DeepSeek and Kimi deliver comparable performance for a fraction of the cost, according to the report.

This is the structural backdrop that makes the discount war more than a two-company skirmish. If a Chinese lab can match frontier-adjacent performance at a fraction of the serving cost, then Western vendors cannot sustain premium pricing indefinitely. OpenAI's aggressive discounts can be read as an attempt to consolidate enterprise relationships before cheaper alternatives — whether from Beijing or from open-weight models — erode the pricing power of every incumbent.

Why does this matter for the AI market?

The stakes are straightforward. Enterprise AI spending is consolidating around a small number of vendors, and price is proving to be a decisive weapon — decisively more decisive, in these cases, than model preference. Rippling did not switch because OpenAI's models became categorically better; it switched because the economics changed, and Conrad's 16-to-52 percent figure shows how fast a vendor mix can flip when it does.

The margin question now hangs over OpenAI's economics. Buying share at the expense of margins is sustainable only while capital markets fund it, and only while real efficiency gains from the Sol and Luna lines continue to arrive fast enough to narrow the gap between discounted prices and true costs.

For buyers, the immediate effect is favorable: a US bank pays under $7 per seat, DataCamp pays 80 to 90 percent less, and a $100 million-plus annual contract gets cut by more than 30 percent. For vendors, the pressure compounds — especially for Anthropic, whose investor memo already records decelerating growth while its share of marquee coding customers collapses.

The next signal to watch is whether Anthropic responds with its own discounting, and whether OpenAI's Sol and Luna efficiency gains are large enough to let it hold these prices once the share-buying phase ends. Until then, every large Anthropic customer holds a bargaining chip that did not exist a year ago.