← Back to feed

Testifying for a second day in the federal lawsuit brought by Elon Musk against OpenAI, company president and co-founder Greg Brockman stated that OpenAI expects to spend $50 billion on computing power in 2026. The testimony was delivered May 5 at a federal courthouse in Oakland, California, where Musk is suing the company on the grounds that it improperly converted from a nonprofit to a for-profit entity.

Brockman framed the figure as part of a longer arc: OpenAI’s compute costs have risen from approximately $30 million in 2017 — when the organisation was a small research lab — to tens of billions of dollars today. The 9-year trajectory represents a roughly 1,600x increase in spending.

The $600B Target Through 2030

The $50B annual figure is not the ceiling. A source familiar with the matter told Reuters in February that OpenAI is targeting roughly $600 billion in total compute spending through 2030 — implying an average of $120B per year over the remainder of the decade, or a continuing ramp well beyond 2026.

For context, AWS’s entire capital expenditure for 2025 — one of the largest annual infrastructure programmes in corporate history — came in at roughly $145 billion. OpenAI’s own 2026 compute budget is thus tracking at more than a third of that.

What the Money Buys

OpenAI’s compute spending finances three things: training runs for frontier models (GPT-5.5, its successors), inference capacity to serve ChatGPT’s global user base, and research compute for alignment, safety, and capability work. The company serves models to hundreds of millions of users and is investing heavily in agentic infrastructure — OpenAI Codex and its autonomous agent products place additional inference load on top of consumer ChatGPT traffic.

The spending is distributed across Microsoft Azure (OpenAI’s primary cloud provider under a multiyear exclusive agreement that has since been partially opened to AWS and Google), as well as OpenAI’s own compute investments under Project Stargate and Colossus.

The Trial Context

Musk is seeking $150 billion in damages, alleging OpenAI abandoned its founding charitable mission when it restructured into a capped-profit and then fully commercial entity. Brockman’s testimony touched on the early dynamics of the organisation: he recounted a 2017 Dota 2 win by an OpenAI system as a trigger point after which Musk pushed to commercialise the lab, and said Musk subsequently demanded a 51% controlling stake in the organisation. Those negotiations reportedly ended without agreement, and Musk departed the board.

Brockman also disclosed under oath that his personal stake in OpenAI is worth approximately $30 billion, placing him among the wealthiest 100 people globally. He confirmed the company is exploring an initial public offering.

The Infrastructure Implication

OpenAI spending $50B on compute in 2026 — while still privately held — reshapes how infrastructure suppliers, cloud providers, and chip makers model demand. That single customer relationship is large enough to move GPU allocation queues, influence TPU roadmap priorities, and anchor multi-year financing deals. The Stargate joint venture with SoftBank, Oracle, and others was partly engineered to give OpenAI capital access to fund exactly this scale of commitment without drawing it entirely from Microsoft Azure agreements.

The $600B through 2030 figure, if realised, would make OpenAI one of the largest infrastructure spenders in history — alongside the hyperscalers whose services it simultaneously relies on and competes with.

Key Numbers

• 2017 compute spend: ~$30 million
• 2026 compute spend: $50 billion (Brockman, under oath)
• Growth multiple: ~1,600x in 9 years
• 2030 target: ~$600 billion cumulative (Reuters, February 2026)
• Musk damages sought: $150 billion
• Brockman personal stake: ~$30 billion