Enterprise AI is becoming hybrid

Enterprise AI is moving closer to enterprise data.

Latham & Watkins, the second-largest US law firm by revenue, has purchased NVIDIA GPU servers to run and customize open-weight AI models within private infrastructure.

The firm operates the servers in a third-party data center accessible only to its employees.

Latham still uses commercial AI platforms, including Harvey.

This is a hybrid strategy, not a rejection of cloud AI.

Different workloads require different environments.

Private AI infrastructure gives large enterprises greater control over:

• Sensitive and proprietary data
• Model selection and customization
• Integration with internal systems
• Security and access policies
• Performance and latency
• Long-term token costs
• Dependence on external providers

The economics become more attractive when demand is large, predictable, and continuous.

Yet owning AI infrastructure also creates new responsibilities:

• GPU utilization
• High-performance networking
• Storage and data architecture
• Power and cooling
• Model lifecycle management
• Cybersecurity and governance
• Specialized technical talent

Latham reportedly employs around 900 technology professionals, including approximately 100 focused on AI. Few organizations have this level of scale and expertise.

The broader lesson is not for every enterprise to buy GPUs.

The lesson is to place each workload where security, performance, control, and economics make the most sense.

Frontier models will remain important. Public cloud will remain important. Private infrastructure and open-weight models will also gain a larger role.

For technology partners, this creates opportunities across AI factories, data-center networking, cybersecurity, integration, model management, and lifecycle services.

My takeawayenterprise AI architecture is shifting from cloud-first toward workload-first.

Does your AI strategy define which workloads belong in the cloud, in a private environment, or across both?