This is best illustrated by one of my most favorite EconTalk episodes ever, “Roberts on Smith, Ricardo, and Trade“. I enjoyed that episode so much that I’ve probably listened to it 3 or 4 times. The point I’m referring to gets made about 20 mins in, although you have to listen through the first 20 mins for the proper buildup.
Smith’s point is that if the extent of the market is large enough, only then does it make sense to invest in capital (technology) that will make it cheaper to produce output. If the extent of the market is not large enough, investing in that capital will actually make the output more expensive.
Or in a more concrete example: If you wanted to make a sandwiches for yourself at home every day for lunch, it wouldn’t make sense to buy a deli meat slicer (the technology) for just yourself. The slicer is expensive and turning it on and cleaning it would be too much hassle for a single sandwich. However if you were making 500 sandwiches per day as a sub shop owner, then having the slicer would allow you to produce sandwiches more quickly and more efficiently (less labor needed per sandwich).
This relates back to the the Amazon video in the examples of innovative datacenter optimizations they’ve created for themselves. Since Amazon is operating at such a massive scale, they’re able to invest resources into optimizing power supplies, BIOSes, UPS firmwares, and custom storage systems in ways that a smaller player would be foolish to spend time investing in improving for their own benefit which is of a much smaller scale.