When does the race move to hardware?
The supplied post argues that after years of competition around better foundation models, hardware becomes a strategic battleground. Test the causal idea: usable deployment capacity is constrained by the weaker layer.
Choose a prediction, then run the scenario.
Find the limiting layer.
This is a conceptual capacity model, not a market forecast or a legal account. The post supplies the phase-shift claim; the lab teaches how bottlenecks determine which layer deserves attention.
Worked case: with model capability 78 and hardware capacity 38, deployment is 38. Improving the model alone cannot lift deployment above 38.
Counterexample: once hardware rises to 90 while the model remains 78, the bottleneck moves back to the model. “Hardware matters” is not a universal claim; it depends on which layer constrains the system now.
Transfer task: raise hardware above 78, run the scenario, and explain why the bottleneck changes.