Saying that you can fund centralized payment systems with Bitcoin isn't an argument for Bitcoin, it's a solution in search of a problem. Even if you think fiat is completely unacceptable you can fund those centralized payment schemes in Africa with gold, or silver, or cowrie shells, or human urea (not making it up; https://twitter.com/UroFoundation), or any of a million other things that don't require billion-dollar foundries to make the mining hardware you need.

You're right that you can't start a new Visa in your garage, but as it turns out even they have competition, and more than "very little". In any case "garage bootstrap ability" is not a value metric that I hear as a serious talking point that often. Normally we bootstrap things in garages and quickly try to scale them up out of garages (Visa itself was probably started in nothing larger than a Bank of America conference room, after all). And good luck fitting the datacenters behind Chinese-based Bitcoin mining operations in a single garage.

As far as competition alone driving fees down, that only works up to a point (which is why no efficient market has managed zero fees, even in high-competition markets where the profit margins are squeezed very low). For Bitcoin as long as miners can turn a profit for the price of electricity they use, they'll continue to come online to buy tickets in the lottery with KW-hr, so the transaction fees will be based on the delta between the price of electricity and the price of Bitcoin itself, and mining itself will migrate to increasingly more efficient (and centralized) operations with plentiful cheap electricity and cooling as less efficient miners are driven out of business. But at 7 tranx/sec even the most efficient miners won't easily be able to pay for that on pennies/transaction.

And since individual miners collaborating in mining pools will probably never go away (since it's not that expensive to "play the lottery" with only a couple of ASICs on your personal power bill), the efficient mining firms will have to eke out a profit on something less than 100% of the blocks out there, causing them to charge even more per transaction.

At the end of the day the "fundamentals" is that Bitcoin in steady-state will convert electrical power into a distributed ledger without trusted third parties (except, of course, for the centralized miners themselves...), and that electrical power has to be paid for by the users of the Bitcoin network itself.

Doing the same thing that Bitcoin does today with distributed networks that are orders of magnitude more power-efficient (even if they did utilize trusted third parties) would result in a much, much cheaper financial network, which would be even that much better for the poor of Africa (and that much easier to drive to near-zero fees by competition).

After all, those transaction fees for Bitcoin don't pay for anything than keeping the ledger up to date, and don't include insurance fees for things such as fraud, scams by the counterparty, and a host of other things that are accounted for in the fees that Visa and its many competitors charges.