Latham moves down the AI stack with Nvidia servers

On September 11, Legal IT Insider reported that Latham & Watkins had bought Nvidia GPU servers to train and run proprietary AI models, citing a Financial Times report published the previous day. The approach gives the firm more control over sensitive data, open-weight models and consumption costs, while requiring capital, specialist talent and ongoing model operations. For vendors, a firm with Latham’s reported $8.3 billion revenue can become both a buyer and an internal competitor, especially in high-value workflows where infrastructure control supports differentiation. The broader market remains segmented: most firms will still need managed platforms, but large clients may increasingly demand deployment and pricing options that reduce dependence on a single hosted model.