I think it would be fairer to compare your $1000/thousands_of_life_years figure against other equally speculative back-of-the-envelope calculations, instead of against AMF. Generally more speculative causes (such as xrisk reduction) can produce better expected returns, but with a greater chance that we're somehow fooling ourselves or ignoring an unknown unknown.
Also, where did the $1000 estimate come from? Does it include the probability of success of cryonics and the probability that we survive for thousands of subjective years without destroying ourselves?
where did the $1000 estimate come from?
It was an estimate from someone at an optimial philanthropy meetup. I don't think it included any improbability discounts.
Imagine you accept the main idea of cryonics, that if we freeze brains future technology is likely to be able to extract the encoded information and revive the person digitally. [1] While this currently costs about $120K/person, if we did it routinely to everyone it could probably get down below $1000/person. Which is interesting: the current cost of averting a death is around $1700, but someone who doesn't die of malaria is still going to die of old age, so you can't really say their death was "averted". While someone who is revived after being frozen wouldn't live eternally, they might get to experience thousands of years of subjective life. In terms of life-years, getting cryonics to be really cheap and paying for people to get it sounds like it beats GiveWell's top charities.
Aside from not agreeing that cryonics is likely to work, however, I don't think the value is actually all that high. A future world which would revive large numbers of people we freeze today would be massively different from the current world economically, but would still have constraints. There would be some number of digital people that could run simultaneously on whatever people-emulating hardware they have. Preserving additional 21st century minds would give future people the option to run revived people instead of new people, but I don't think it affects the overall number emulated.
[1] I don't think this is likely.
I also posted this on my blog.