Goldman Sachs in AI: Research, Compute Finance & Adoption

A 157-year-old bank with a new line of business

Goldman Sachs is a Wall Street investment bank founded in 1869, run today by chairman and chief executive David Solomon. It operates through two main divisions, Global Banking & Markets and Asset & Wealth Management, and reported roughly $58 billion in 2025 revenue against $3.6 trillion in assets under management. None of that is AI infrastructure. Its relevance to the timeline comes from three places: research it publishes on whether AI spending pays off, capital it is starting to put behind AI data centers, and the AI tools it has rolled out to its own employees. [4]

Skeptical research, real money

In June 2024, Goldman Sachs Research published "Gen AI: Too Much Spend, Too Little Benefit?", a Top of Mind report built around interviews including its own head of global equity research, Jim Covello, who argued that the industry's roughly one trillion dollars of planned AI infrastructure spending would need to solve problems more valuable than the ones generative AI had solved so far to pay for itself. The report did not speak for the bank's overall strategy, but it became one of the most cited skeptical documents of the AI buildout, republished and debated well past its release. [3]

Atlas interpretation: That skepticism sat alongside Goldman continuing to work the deal side of the same buildout it was questioning. A research house doubting the return on AI infrastructure and an investment bank arranging capital for AI infrastructure are different parts of the same firm, and both views can be sincerely held at once. [3]

Underwriting the buildout

On August 10, 2026, Nvidia named Goldman Sachs one of six financial partners in a set of memoranda of understanding meant to mobilize more than $500 billion of third-party capital for AI data centers, alongside Apollo, BlackRock, Blackstone, Brookfield, and KKR. The structure treats GPUs themselves as collateral for financing, an approach Nvidia's own announcement describes as opening a new asset class rather than a single transaction. Goldman's role here is arranging and underwriting capital, the business it has run for other infrastructure booms, applied to this one. [1][2]

These were memoranda of understanding, not closed financings, and the announcement did not disclose how much of the $500 billion figure Goldman specifically would underwrite. Coverage of the announcement noted the resemblance to earlier asset-backed lending structures, without settling whether GPU-collateralized debt will perform the way the underlying compute market does. [2]

AI inside the walls

Goldman built its own internal tool, the GS AI Assistant, giving employees firewalled access to large language models including ChatGPT and Gemini for tasks like coding help, translation, and document summarization. It reached about 10,000 employees, roughly a quarter of the firm's global workforce, by January 2025, then rolled out firm-wide in June 2025. Chief information officer Marco Argenti described the goal as productivity gains for staff and downstream benefits for clients. [5][4]

Atlas interpretation: Goldman's AI story is less about building models than about the three roles a large financial institution can play around them: questioning whether the spending is rational, helping finance it anyway, and quietly automating parts of its own workforce with the same technology. [3][1][5]