The news: OpenAI is discounting enterprise ChatGPT subscriptions—but only if customers agree to buy more AI products. Microsoft is unhappy, as it rarely offers discounts for its competing services aimed at enterprise users, per The Information.
Discounts on ChatGPT can range between 10% to 20% for clients that agree to buy tools like Deep Research or the Codex coding agent or commit to higher API access quotas.
Partners now in a price war: Microsoft has lost deals because it didn’t match OpenAI’s prices and leaned instead on Azure’s security and compliance as differentiators to ChatGPT’s service.
The Information cited a software company in talks to spend $8 million in three years for OpenAI models through Microsoft; however, OpenAI offered a 20% discount for the same models and reduced the company’s cost by $1 million over three years.
Direct competitors: OpenAI taking business from Microsoft—most recently through a defense contract—is the latest in a series of conflicts accelerating the fallout between the two companies.
- On Tuesday, the Department of Defense (DoD) granted OpenAI a $200 million contract for frontier models designed to “address critical national security challenges in both warfighting and enterprise domains.”
- A report from TechCrunch says that the DoD, which engaged Microsoft for Azure OpenAI Service in April, is now going directly to the source, effectively cutting out the middle man.
Things could get worse: OpenAI’s executives are weighing antitrust charges against Microsoft while seeking federal oversight of their contract terms in which Microsoft owns a 49% stake in OpenAI, per Reuters.
Key takeaway: Business leaders should anticipate potential shifts in AI pricing resulting from fraying alliances and increased competition between OpenAI and Microsoft. Lock in longer-term pricing and negotiate for essential services while pushing back on expensive add-ons.
Diversifying AI vendors and solutions reduces reliance on single entities and provides opportunities for testing of models from different sources, some of which may be more easily customized for specific use cases.