OpenAI: What Is OpenAI's $600B Compute Spending? Did They Really Cut the $1.4T Investment Commitment?

The Information’s latest report contains this quote: “But it warned it will burn more than twice as much cash through 2030 than previously predicted, as it spends $665 billion on the costs of running and training its AI, according to the financial forecasts.”

In the past two hours, I’ve seen numerous news mentioning that OpenAI has reduced its $1.4 trillion CAPEX commitment to $600B, interpreting this as an extremely negative signal. I feel compelled to write this report to clarify the misconceptions.

Let me state two key conclusions upfront before diving into the details

• Misaligned Metrics

  • The $1.4 trillion refers to CAPEX Commitment, which includes not only OpenAI’s direct investment but also investments from Azure, AWS, ORCL, and CRWV.

  • The $600B refers to Compute Spending/Compute Cost within OPEX. The Information has historically shared excellent reports that you can track - they’re all OPEX metrics. In fact, I understand OpenAI never discusses CAPEX metrics because they’ve had virtually no CAPEX until now.

  • The key difference between CAPEX and OPEX is that this year’s CAPEX gets depreciated over the next 5 years, making CAPEX a leading indicator. During CAPEX upcycles, CAPEX will consistently exceed OPEX.

• Misaligned Time Horizons

  • The $1.4 trillion commitment was first mentioned by Sam in a podcast about the AVGO partnership after AVGO’s earnings, where he referenced the next 8 years. So $1.4 trillion covers investment commitments from 2025 to 2032/2033.

  • The $600B represents Compute Spend (Training Compute + Inference Compute) from 2026 to 2030.