Summary
On May 5, 2026, OpenAI President and co-founder Greg Brockman testified under oath in federal court that OpenAI plans to spend approximately $50 billion on computing power during calendar year 2026. The disclosure — made on the second day of Brockman's testimony in the Musk v. Altman trial before Judge Yvonne Gonzalez Rogers in Oakland, California — is the first time OpenAI has publicly quantified its current-year infrastructure budget at this scale. Brockman placed the figure in historical context: OpenAI's computing costs were approximately $30 million in 2017. Bloomberg and Reuters reported the figure the same day.
What Happened
Greg Brockman took the witness stand as the second major OpenAI executive to testify in the Musk v. Altman trial, following Elon Musk's own three-day testimony during the opening week. Brockman serves as OpenAI's President and is one of its original co-founders; his equity stake in the company, also disclosed during the same trial period, was valued at nearly $30 billion. His testimony addressed the corporate history and structure of OpenAI, contesting Musk's characterization that Brockman and CEO Sam Altman had abandoned the company's charitable mission for personal financial gain.
During direct examination, Brockman stated that OpenAI projects $50 billion in spending on computing power in 2026. He described computing costs as having escalated from approximately $30 million in 2017 — the company's second full operational year — to tens of billions of dollars in the current period. The trajectory represents an increase of roughly 1,666-fold over nine years. Brockman framed the scale of capital required as central to OpenAI's argument for its for-profit conversion: the nonprofit structure, he argued, was incompatible with the financial demands of building and operating frontier AI infrastructure at current scale.
The $50 billion figure, taken at bulk purchasing rates for Nvidia B200 GPUs, corresponds to approximately 1.25 million discrete accelerators. OpenAI's compute commitments extend beyond 2026: the company had previously disclosed, in February 2026, a pledge to investors of roughly $600 billion in spending through 2030. The $50 billion annual rate implies that 2026 represents the current-year execution of that longer commitment, with the company operating inside the Stargate infrastructure partnership with SoftBank, Oracle, and the federal government alongside its separately financed hardware procurement.
The testimony came during the liability phase of the trial, which the court had scheduled to conclude by approximately May 21, 2026. Under cross-examination by Musk's attorneys, Brockman was pressed on whether the scale of his personal financial stake — nearly $30 billion — demonstrated that the nonprofit conversion had enriched OpenAI's principals at the expense of the charitable trust Musk alleged was formed by his approximately $38 million in contributions. The $50 billion compute disclosure was made during this broader testimony about OpenAI's capital structure and financial requirements.
The Musk v. Altman trial had previously produced one other significant financial disclosure: on April 28, Musk acknowledged under oath that his actual contributions to OpenAI totaled approximately $38 million, substantially less than the $1 billion he had publicly announced, citing lost confidence in the team as the reason for the shortfall.
Why It Matters
The $50 billion figure is the first sworn public disclosure of OpenAI's current-year infrastructure budget and anchors the empirical record on AI compute spending in 2026. Prior disclosures — the $600 billion 2030 commitment and participation in the federal Stargate partnership — described multi-year aggregate ambitions. The trial testimony converts the annual rate into a number that has been stated under oath, subject to perjury consequences, rather than offered as a promotional or investor-relations figure. That distinction gives the $50 billion a different epistemological standing than projections disclosed through press releases or earnings calls.
For context on scale: at the $50 billion annual rate, OpenAI alone would represent approximately 5 to 8 percent of total projected US public and private AI infrastructure spending in 2026. The company's computing costs have escalated at a compound annual rate of approximately 100 percent per year over nine years, though the absolute acceleration is more pronounced: the jump from $30 million in 2017 to the current level represents growth that compresses most of the absolute dollar increase into the final two years.
The disclosure is also contextually significant within the trial itself. OpenAI's defense of its for-profit conversion rests substantially on the premise that the nonprofit structure was financially nonviable for an organization competing at the frontier of AI development. The compute spending trajectory — from $30 million to $50 billion in nine years — is the quantitative expression of that premise. How the jury weighs the capital-necessity argument against Musk's claim that the conversion was pretextual enrichment is not yet known; the liability phase of the trial had not concluded as of the date of this disclosure.
What remains unresolved: whether OpenAI's $50 billion in 2026 compute spending is achieved through owned hardware, cloud procurement, or the Stargate partnership's shared infrastructure, and in what proportions; whether the figure will be independently confirmed through financial disclosures in OpenAI's anticipated IPO prospectus; and how the compute spending trajectory affects the company's projected profitability timeline.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.