Still, Enderle sees multiple reasons for companies to buy AI development hardware for on-premises use. “Clients may not have the connectivity they need for an AI project hosted in a cloud,” he explains. “Many developers like their own hardware, especially if they have something special and worry about leaks. Beyond those, healthcare and the legal profession both have compliance requirements that may demand local hardware.”
For resellers who want to build an AI hardware business, “you’ll go down that programming road,” says Bach.
You’ll need to take a deep dive into Linux too, as it’s the OS of choice for AI and ML. Puget Systems uses the CentOS version of Linux.
Don’t expect a huge market, adds Bach. Puget Systems often sells AI devices to existing workstation customers. But while those customers may buy one workstation for each of its engineers, they usually buy only one or at the most two AI development systems. Today the group most interested in hardware for AI and ML systems are the developers.
System builders should know there’s lots of opportunity, but little low-hanging fruit, says Bach. “The good news is that anyone can do it. The bad news is that it takes hard work. You need to work closely [with] the developers to define the hardware, and then all the customer will buy is one or at most two AI workstations.”
“This is an expensive area to get into,” says Enderle. “Look at this market and see if you can fit. The technology has to match what your customers need.”
Image: Courtesy Puget Systems