Copilot's AI Revolution Will Require Powerful New Hardware
-
Microsoft's definition of an "AI PC" requires at least 40 trillion operations per second (TOPS) to run Copilot locally. Complete local processing not possible yet due to model size.
-
Copilot will likely use a mix of local processing and cloud offloading. Local models could improve latency and privacy.
-
Local processing could act as a fallback if internet connection is disrupted. The mix of local and cloud processing may be dynamic.
-
Current hardware not powerful enough - AMD, Intel and Qualcomm chips fall short of the 40 TOPS requirement. Upcoming Qualcomm Snapdragon X Elite mobile processors will meet specs.
-
As hardware improves, more Copilot functionality expected to shift from cloud to local device processing. But some cloud processing likely always needed due to massive model size.