Microsoft Expresses Concern Over OpenAI Amazon Talks

Court filings from the ongoing Musk v. Altman trial, as reported by The Verge, reveal early-stage communications between Microsoft and OpenAI leadership about funding and cloud compute for OpenAI's gaming experiments. According to the reporting, OpenAI CEO Sam Altman proposed roughly "Probably something like $300 million at Azure list prices" to scale a Dota 2 project, and Microsoft Azure executive Jason Zander cautioned in an August 2017 email that the arrangement would need to drive "significant incremental revenue directly due to the deal ($500 million+)", per The Verge. The filings also record Microsoft executives saying OpenAI might "storm off to Amazon" and "shit-talk" Microsoft, and describe an alternative proposal from Altman to create an Xbox-facing partnership and to share IP, according to The Verge.

The documents focus on compute scale and commercial terms rather than model architecture. The specific request reported by The Verge frames the ask as the equivalent of $300 million in list-price Azure consumption, indicating early OpenAI requirements for large-scale cloud GPUs to support gaming-related ML workloads. The filings, as reported, do not detail exact instance types, model sizes, or training recipes tied to that dollar estimate.

Editorial analysis: Companies negotiating exclusive or preferred cloud arrangements during rapid model scale-up frequently confront tensions between short-term compute discounts and long-term commercial control. Public reporting on historical negotiations, such as these court filings, highlights how cloud economics and exclusivity concerns have long shaped relationships between startups building large-scale models and hyperscale providers.

Editorial analysis: For practitioners, these disclosures are primarily historical, but they illuminate persistent dynamics: early compute needs can be large enough to drive bespoke commercial proposals, and cloud-provider competition affects where large ML workloads run. The episode also shows executives explicitly discussing reputational risks between partners, a factor that can influence contract language and go-to-market arrangements.

Editorial analysis: Observers should monitor whether future litigation or disclosures produce additional contract-level details such as minimum-commitment clauses, noncompete-like language for cloud usage, or specific account-level revenue thresholds. Those contract elements, when they appear in public filings, offer concrete signals about how compute supply and commercial terms evolved as model scale increased.