What decides which computing resource does which job?
By the end of this lesson: Understand what a control plane does; distinguish local orchestration from national scheduling; see why heterogeneous compute makes abstraction valuable.
← Learning pathThe coordination problem
A large AI estate can contain different accelerators, clouds, networks, storage systems and locations. Software must match workloads with compatible resources while balancing performance, availability and cost. This becomes harder as the estate becomes more heterogeneous.
Local and national orchestration are related — but not identical
Inside a tightly coupled AI system, orchestration may happen at microsecond-scale and control memory and processors. Across regions, scheduling operates with very different latency and commercial constraints. Our thesis sees a common optimisation problem at multiple levels without pretending that one control plane literally manages everything.
Where companies enter the picture
Different companies may provide pieces of this abstraction layer through cloud platforms, AI software, heterogeneous-compute management or vertically integrated stacks. Participation does not automatically mean control, and technical relevance does not automatically imply economic value capture.
Evidence, claims and limits
- National integrated computing-power network portal — national platform
- National policy implementation opinion — policy source
Can you apply the idea?
Choose an answer, then reveal the reasoning. This is not scored — the aim is to practise distinguishing what evidence proves from what it merely suggests.