The differences there are that F1 or Scaled will go fairly quickly into vehicle/flight testing and get their data there, also that neither is certifying anything to carry fare-paying passengers.
The light aeroplane world also does relatively little either CFD or wind tunnel work - for that matter they still fairly routinely will do only basic hand calculations or single-operator FEA then build and carry out physical load tests.
It really is an economic decision - you'll always need some combination of analysis, model/sample test, and full scale test on both structures and aerodynamic features. The scale however tends to dictate the proportions.
At the microlight/ homebuilt level, a couple of engineers will probably spend a few daysor weeks with Excel or a basic combined FE/CFD package such as Solidworks working through the main cases, then they'll build two, test one structurally to destruction, and start flight testing the second. At that scale, build and test is usually much cheaper and faster than analysis.
At the airliner level, a massive team of engineers with a cutting edge package such as Catia will analyse every tiny component to every possible case - then the most critical cases will be tested to confirm the analysis is correct. At that large and complex scale the thousands of man-years and millions of pounds on computing time are still far cheaper and faster building and testing.
Structurally also the increasing use of composite structures makes structural strength estimation incredibly tough to get right. In tension it's easy enough, but in compression where the strength is a function of the microscopic-level buckling loads of patterns of fibres that just can't be laid with that sort of microscopic precision in the way that you (relatively) can with metal structures, it is really really hard and you have no choice but to do a lot of testing. Another factor very very hard to model is rate of load onset - so for example undercarriage drop tests are still done at every scale of aeroplane, and probably always will be.
Regarding wind tunnel work specifically, the simple stuff - lift and drag can pretty much be done by analysis. The difficult stuff is behaviour around stalled components where the point of boundary layer separation is incredibly hard to model, anything involving vortex street creation (and thus flutter) which is equally tough to get right because it's so dependent upon very complex shapes and Re/viscosity/Mach interaction effects, and the detail of stability derivatives that lead to prediction of flying qualities - or increasingly determination of control laws for the AFCS.
So in a nutshell, I don't think you'll ever get away from testing - but the more large and complex the project, the more you'll see companies trying. And there will always be a point where the cost of the computing means that it's cheaper to build and test something than "just" do it on computers.
G