Positron says inference drives two-thirds of AI workloads

siliconangle.com3 min read

AI infrastructure workloads are shifting significantly toward inference, which now accounts for about two-thirds of operations compared to past dominance by training. To capture this shift, chipmakers like Positron AI Inc. are designing servers around commodity memory instead of scarce high-bandwidth memory to reduce power and cost. According to Positron sales executive Cameron McCaskill, these efficiencies enable rapid deployment in modular data centers. Open article →