Amazon would benefit from future citizens that can walk without wobbling.

In the short term maybe. More long term and generally, there are going to be 400 million people living in America, over the lifespan of Amazon the entity, the lifespans of its employees and shareholders, all three will have to interact with those 400 million people, some on the street, a lot (in a small way) in political processes and systems, in art, as future employees, as employees in businesses they partner with, etc. Advertising is mass programming (albeit in a small way) of large swathes of the population.

There's 44 million US Twitch users, so basically you can show a message to 1/10th of the entire youth population of America. In 10 years there is probably some minor difference in society between whether that message was "go drink mountain dew" or "here's how DNA works" or something simpler, or more useful, or different things tailored by age, etc. Now if Amazon makes like 3 cents per ad (a number I made up) * 44 million users compounded by 7% annual interest over 10 years then that's 2.5 million dollars of profit. But money is a tiny slice of the entire pie of what is possible in an economy. If instead Amazon "donated" that 2.5 million dollars (actually more if 3 cents per ad is profit not revenue) by using its ad infrastructure to forcefully cram useful knowledge into the children instead of knowledge about a sugar drink brand, it's possible that everything that is possible in society 10 years from now would essentially make all the other money Amazon makes more valuable, because, for example, doctors that Amazon shareholders would want to go to would charge less because they indoctrinated a few more children into an interest in biology instead of Mountain Dew consumption so there was more doctors and less diabetics.

Will it be enough to offset 2.5+ million dollars? Probably not, but we don't have the technology to express that cost/benefit calculation, because the best we have is spreadsheets and demand/supply curves (if you will permit my derisive oversimplification of what economic tools business people use). We have developed computers and mathematics, as well as a system of assigning token counts to people and groups of people (money) that when people and groups of people work hard on increasing their respective token counts seems to correlate quite well with generally improving things and getting things we want. Numbers (bits really) are highly physically stable states and mathematics is operations we can perform of those states that are themselves very stable in time and predictable. We can measure things in the world and record those measurements into bits and then apply operations on those bits and we find that sometimes we're able to build computations that result in a strong correlation between the resulting state and physical reality. We're able to predict the future. However there's still a pretty huge mismatch with what we're able to easily express with simple mathematics on collections of bits and whatever our brains can do, which we experience as intuition/common sense and empathy.

Basically I'm saying it would take a strong willed person with the intuition to see that "if we use our mass population programming tools to increase the competence of our youth we will be better off than if we increase the consumption of Mountain Dew by 13% (or whatever they estimate)" because it's very difficult to truly express that comparison as a spreadsheet of dollar amounts, which is the primary technology businesses have for making decisions. Because a spreadsheet is a Turing machine and typing in data and programs into it is really time consuming whereas a human brain is a very different type of computer that processes input much more efficiently but is more fallible because other brains can't inspect its thought process, which is why we usually prefer spreadsheets.