INFRASTRUCTURE
“Distributed training hits a wall when the network between nodes can’ t keep up”
Colm MacCárthaigh Vice President and Distinguished Engineer AWS
LiveLink, built as part of the new Leica RTC laser scanner series, for example, creates a real-time connection between field and office teams.”
He adds that, for multi-person field crews, capture progress can be synchronised so everyone has a shared view of what has already been scanned, reducing the risk of duplicated or missed capture.
“ We also use Hexagon Smart Point Cloud( HSPC), a format developed by Hexagon’ s research team that allows metadata to be stored at the point level,” Cormac goes on.“ Rather than loading an entire point cloud, you can select only the information relevant to the task – for example, retaining walls and floors, while excluding vegetation. That principle becomes particularly important for AI. If you can separate useful structural information from noise early in the pipeline, the neural network downstream only has to process the data worth learning from.”
Economics of scale As the pair point out, delivering ROI ultimately comes down to efficiency rather than raw power.
Matching the right workload to the right silicon is paramount to keeping spend grounded, as Colm explains:“ It really comes down to matching the right tool to the job. Custom silicon like Trainium and Inferentia exists because not every workload needs a general-purpose GPU, and the cost savings from that specificity are significant. Beyond hardware, the organisations getting the strongest returns are the ones making smart architectural choices: routing simple tasks to lighter
104 October 2026