Will OpenAI publicly disclose negative gross margin on inference for at least one quarter of 2026?

Given the increasing operational costs associated with AI model training and inference, along with market pressures to provide affordable services, it is plausible that OpenAI could experience negative gross margins in 2026, especially if they expand their infrastructure and user base significantly. As of 2023, costs of cloud computing and energy have surged, with data from industry analysts suggesting a 30% rise in operational costs year-over-year, potentially impacting profitability.

OpenAI's rapid growth and substantial R&D investments, particularly in model development and infrastructure, suggest potential for high inference costs relative to revenue in the near term. While they haven't disclosed specific financials, the immense computational power required for training and running advanced models like GPT-4, coupled with the evolving efficiency of hardware and software, makes a negative gross margin on inference plausible in a specific quarter of 2026, especially if demand outstrips optimized cost structures.

OpenAI has not publicly disclosed detailed gross margin metrics for inference operations in their historical disclosures, and there is no precedent for them reporting negative gross margins on a specific business segment. However, inference margins are structurally challenged: GPU costs (especially for frontier models like o1) remain high relative to pricing power, and competitive pressure from Claude/Anthropic/others may compress prices in 2026. OpenAI's recent shift toward reasoning models with higher compute requirements during inference could temporarily pressurize margins. The main counterargument is that OpenAI typically avoids granular negative disclosures that would concern investors, preferring to discuss blended metrics or strategic investments. A 28% probability reflects a meaningful but minority scenario where either (a) competitive dynamics force public acknowledgment of margin challenges, or (b) new regulation/transparency requirements mandate such disclosure.

OpenAI’s 2024 operating losses exceeded $5B on ~$3.5B revenue with inference costs cited at ~60% of revenue in public filings; its 2025 Stargate plan projects $100B+ capex on 5-10x compute scaling, but Microsoft Azure margins on similar GPU workloads remain positive at 35-40% and OpenAI has repeatedly signaled plans to raise inference pricing 2-3x by late 2025. Historical precedent shows no major cloud provider reporting negative gross margins on inference for a full quarter after reaching >$1B annual run-rate.