AI has taken off as an important, differentiating capability in all industries, and the hardware required to run AI is rapidly evolving. The technology industry is often very focused on the exponential growth in size that the most advanced AI models are going through. The discussions are about tens of billions of parameters, reducing precision, expanding memory, high-performance computing (HPC)–like needs for AI training and inferencing, and racks of accelerated servers. In reality, this extraordinary scale of AI computing is the exception, especially in the enterprise.
Today many businesses are working hard on AI initiatives that do not require a supercomputer. Indeed, a lot of AI development — and increasingly AI deployment, notably at the edge — is taking place on powerful workstations. Workstations have numerous advantages for AI development and deployment. They liberate the AI scientist or developer from having to negotiate server time, they provide GPU acceleration even as server-based GPUs are still not easily available in the data center, they are extremely affordable vis-à-vis servers, and they represent a smaller, one-time expense rather than a rapidly accumulating bill for a cloud instance. That way, they also free the scientists or developers from the anxiety of racking up costs while merely experimenting on AI models.