Latham & Watkins buys Nvidia servers to fine-tune open weights in-house
What's the deal? Top US law firm Latham & WatkinsDealroom has a profile for this one. Try Dealroom → is buying Nvidia servers to fine-tune open-weight models in-house — positioning private, on-prem AI as an alternative to frontier APIs from OpenAI and Anthropic, according to the Financial Times .
Why now? The FT reported the build-out on 10 Sep 2026. On 13 Sep, Anand Iyer (@ai) amplified it with a crisp framing that has been widely circulated:
- For Latham: Client work and internal knowledge are among the most confidentiality-sensitive corpora in professional services. Fine-tuning open weights on firm data behind the firewall is a bid for control, auditability, and lower long-run API dependence.
- For Nvidia: Another high-signal enterprise buyer in the “sovereign / private AI” lane — not a hyperscaler, but a white-shoe firm treating GPUs as core infrastructure.
- For OpenAI / Anthropic: The FT casts the move explicitly as an alternative to their hosted models. Even if API spend keeps growing, sophisticated buyers are building parallel stacks for the data they will not send out.
The signal: “Enterprise sovereign AI” is moving from slogan to procurement: open weights + proprietary data + local Nvidia compute. When AmLaw leaders buy servers instead of only seats, the competitive set for frontier labs expands from other labs to every customer that can afford a private cluster.
Read more: Financial Times · @ai · FT on X
