How can we design a Datacenter's servers and compute need starting from the application layer ?

Break down the Application to a caluctlated level of processes , then find the required resources (Processor, RAM, HDD, Bandwidth) to run this function multiply by the average numer of cconcurrent sessions connected per time interval as best practice.  

Assess Instructions in the business applications and support applications. Sum all the Instructions and then you can Benchmark the equipments available and identify the CPUs in use.

Each CPU defines a MIPS capacity which is 'million of instructions per second'.

you can then take the first value devise it by the second one and get the number of CPUs needed to run all your applications.

For RAM you can design in proportion with CPU cores as1:1.

1Go of RAM for each CPU core.

You can then find the number of servers needed and the amount of bandwidth to design your Data Center Infrastructure.

If you have the application layer owners with you they can simply provide their needs of servers if they have accurate numbers of users and future requirements from a strategic point of view. A technical workshop can be held between them and the servers provider (internal\\external) to estimate the number of servers\\storage\\critical services ...etc... this should end up with a defined BoQ that will allow the DC designers to calculate the space\\cooling\\power\\connectivity..etc and reflect that on the design and the budget