While a company’s current computing infrastructure is the most important factor in
choosing the proper system architecture, there are some challenges that are unique to
deploying large AI models to automate business processes, including computing power
required to run large language models. The competing performance parameters are
data security, model capability, computing power required, initial setup costs, and on-going usage costs.
| Third Party Infrastructure | Cloud-Based Computing | In-House Computing | |
|---|---|---|---|
| Language Models | Third Party Models | Open-Source Models | |
| Model Capability | The Fourth-Generation Models have over 1.7 trillion Parameters | The Largest Open-source Model has 180 billion Parameters, this is Equivalent to Third Generation OpenAI Models. | |
| Model Costs | Pay per Usage | No Fees for Model Usage | |
| Infrastructure | Basic Computing | Cloud-Based GPUs | In-House GPUs |
| Infrastructure Costs | No Additional Costs | Usage per Hour | Maintenance Costs, Initial Setup Costs |
| Security Risks | Potential Third-Party Access to Data | Possible to Prevent Third-Party Access to Data | No Third-Party Access to Data |