I've been in this business a long time. When someone asks me if I can design a power supply for them, my first question is usually "have you tried to buy one of these?"

Developing power supplies is expensive, time-consuming, difficult and frustrating, yet the custom power business is alive and well.

Given the modern environment of high performance power, digital control needs, efficiency, etc., why are you contemplating or already designing your own power system?

I have always found that the requirements for the power supply is always pushing some operational boundary. For instance: Why do ASICs need 5 different supply voltages at differing currents, all within a 1.8V range? We power supply designers have to "drop and catch" on every design due to the logic designer's inability to design more robust circuits.

We as power supply engineers, have to get used to being the last to know the true power requirements of the system, once designing a power supply, having the label of "power supply expert" nailed to our heads, and performing miracles on "ground-hugging efficiency" designs (and, of course, at zero cost and space).Only praised as a feature (extended battery life for a portable device, or more efficient than the earlier model).
I applaud your dedication and tenacity. Without us, these systems would not be possible.

The first thing is to review the specification as you know. I have seen so many "boiler plate" specs I can't enumerate them. It's cut and paste from previous designs and not what they really need. If they want to save time and money you have to guide them for needed function for power, again, as you know.

If the form factor is not an issue, then it’s preferable to buy the Power Supply units. That way, the product might be launched quicker.

For high volume products such as Telecom equipment and LED Lighting applications, the cost of the power supply design will be absorbed very quickly. In this case, it’s cheaper than buying the Power Supply units.

For military and aerospace applications, where reliability is very important, usually, the power supply is designed in house.

The reason for the custom supply is often a combination, most designs are trade offs. Space/satellites need radiation tolerance, super high reliability, and often have to fit in strange shapes/spaces. When the company gets rid of power supply designers no one knows how to specify supplies. Many system engineers do not understand power, grounding nor EMI so they need a power supply designer to help and guide them or bail them out and write the requirements.

Depending on product objectives there are different requirements that can be the size and shape, the efficiency of the system etc..
We have different product ranges and usually we start with the search for market products to meet all objectives.
Sometimes we have been forced to design custom models, but the time to market of the products has been associated with the successes and failures in custom design.
Thanks to the experience gained, we try to reason with the customer these difficulties, allowing him to decide about this risk.

Customer expect a continuous support for power supplies ,since off the shelf power supply can offer only full replacement.This approach is expensive and many times frustrating.
Further while power supplies designed to meet customer spec can always be added with additional features,since the main product is far more expensive.

Customers can be frustrating. Unless they buy individual modules for each output V and I Very expensive. It would be be quite impossible to meet the needs of the customers. So, custom designs will always be needed for the large volume products. Size, apace and cost will always be an issue to large companies. There will always be a market for us.

In the automotive world cost is king but in my experience the havoc the vehicle charging systems place on the power supply input are beyond most purchased supplies to reckon with. While 14v is nominal transients conditions like load dump (40v), reverse battery (-14v), cold crank (5v for 100ms), drop out (0v for 10ms) plus all the radiated immunity and emissions tests make in house supplies the only way to go.

It's mostly about performance so that's my vote. In my cases I design a lot of RF and analog signal processing gear and then the switchers have to be designed alongside. COTS solutions mostly don't suffice.

For starters, they can usually not be synchronized. Then EMI is poor, many engineers do not understand that area too well and then it's "Hey, it passed class B, let's ship it". The third reason is reliability. There are manufacturers who have no qualms about placing a 600V FET into a line-side flyback that ships internationally.

Many years ago cost effective units were not available. Since then, I have designed a number of application specific embedded supplies. These were typically compound topology designs with multiple outputs. They were designed to accommodate and take advantage of the load characteristics. In almost every case, there was not a practical off-the-shelf alternative. I did design an 80kV supply whose objective was cost relative to purchased. It worked well and at a fraction of purchased unit cost.

In addition to functional requirements, the need to meet requirement for industrial environment and agency approvals drove us toward in-house designs.