Latham & Watkins has bought Nvidia GPU servers and begun customising open-weight AI models, creating an in-house technology infrastructure that gives the law firm greater control over sensitive client data and its use of artificial intelligence.
The US law firm, which generated $8.3bn in revenue last year, has purchased several servers in recent years, with each holding multiple GPUs. The investment means Latham assumes responsibility for operating the hardware and protecting it against cyber security risks, while allowing sensitive client information to remain within systems controlled by the firm.
Rene Mendoza, Latham’s chief information officer, said protection of particularly sensitive client information was one reason for retaining an internal option rather than placing all such information with cloud providers. He also pointed to expected AI consumption costs and the value of maintaining flexibility as usage develops.
The model represents a departure from the approach more commonly taken across the legal sector, where firms have tended to subscribe to AI software and cloud infrastructure rather than commit substantial capital to their own computing systems. Large upfront technology expenditure can reduce the profits distributed annually to equity partners, creating a financial constraint on investment decisions.
Latham has been developing its approach for the past few years and continues to use AI products and services supplied by large technology companies alongside its internal systems. The firm has not disclosed how much it has invested or plans to invest.
The potential expense is significant. Buying and operating GPU servers at this scale, together with employing specialists to maintain, run and protect the infrastructure, can cost tens of millions of dollars per year and could rise to hundreds of millions as the technology advances.
Latham leases space in a data-centre facility that is locked and accessible only by its own staff. Its machine learning and software engineers are fine-tuning Nvidia Nemotron 3 open-weight models. Unlike models from OpenAI and Anthropic, open models make their design public and can be downloaded, adapted and operated on users’ own hardware.
The firm has also expanded the workforce supporting its technology strategy. Latham said it employs more than 900 technology specialists. Michael Rubin, a Latham partner and chair of its AI strategy committee, described significant growth over the past few years in professional technology staff, innovation lawyers, machine learning engineers, AI engineers and coding lawyers.
The resulting infrastructure gives Latham lawyers a broader choice of systems beyond ChatGPT, Claude and Gemini. Mendoza said the firm can select between its own AI systems and commercially available products according to the task, while retaining alternatives if providers including OpenAI and Anthropic change their prices or terms.
That flexibility places the economics of AI infrastructure alongside control of sensitive client data and technical capability in Latham’s operating model. Maintaining internal Nvidia hardware requires capital and specialist resources, but it also leaves the firm able to allocate work between its own infrastructure and external AI services according to the task, while retaining alternatives if provider prices or terms change.
