The cost of housing and managing a unit of hardware is nonzero. Actuals vary wildly by location, purposes, and sector; but if everyone can get their heads out of the hobbyist-tinkerer mindset for a moment and consider that lifetime TCO is a real and meaningful consideration for businesses, then the component that isn't buying/leasing the server itself is typically several hundred dollars, per physical unit, per annum. Public clouds don't nullify all of it, but they swing the needle dramatically.
Since my duty cycle for a Mac Mini is rather less than 20%, the economics even of on-demand instances immediately make sense, and having it inside the VPC boundary without hybridising is gravy on the meat since I'm consuming several other AWS services besides.
These two factors (lifetime TCO and scale-down) are the economic foundation of the value proposition of utility computing. The "I can build that in my garage for cheaper" crew aren't wrong, but they're missing the point.
Putting on my sometime cloud infrastructure product manager's hat, long-term observers of service pricing may also observe the common enough pattern of starting a new product with a relatively high price, one that selects for early adopters and other price-insensitive (and, you hope, glitch/MVP-tolerant) customer segments, then gradually ratcheting prices down in order to estimate (amongst other things) the price elasticity of demand. It's naturally easier to get cheaper over time, than go the other way. In this AWS's position on compute is congruent to any other commodity merchant⁽¹⁾ with the luxury of being able to withstand potential losses on a single product line. For most EC2 instance types I'd expect they have a very good model of the price sensitivities, but mac1 is undoubtedly a unique platform with elevated uncertainty in the price parameters.
Which is my long-winded way of saying, it'll likely be cheaper in a few months.
-----------------------
⁽¹⁾ yes, yes, just like a fruit shop, very droll.