OpenAI's Massive Compute Expenses

What happened

OpenAI committed an estimated $1.4 trillion over eight years to compute infrastructure, significantly exceeding its current $13 billion annual revenue. This investment covers AI chips and servers for model training and response generation. In 2024, compute spending was $5 billion for R&D, primarily on experimental training runs, and $2 billion for inference. To manage costs, OpenAI is expanding paid ChatGPT, offering data centre access, and developing hardware. This involves substantial agreements with vendors including Oracle, which receives $60 billion annually for five years for cloud infrastructure. Projected gross profit margins are 48% in 2025, rising to 70% by 2029.

Why it matters

The substantial, long-term financial commitments to multiple external compute infrastructure vendors introduce a significant operational constraint, tightening dependencies on supplier performance and pricing. This increases exposure to vendor lock-in and supply chain vulnerabilities, placing a higher burden on procurement and finance teams to manage these extensive contracts and associated financial risks. Platform operators face increased due diligence requirements for optimising resource utilisation across diverse cloud environments to meet ambitious profit margin targets.