Thus when you segment the market by "people who can afford $20-100/month", you are also segmenting the market by the correlating variables "people who are mad at their current host" and "people who can spare several days to migrate their infrastructure".
As a member of that class who migrated to Linode, I can attest that the size of that market is nonzero. But I can also attest that a great many projects start and remain on AWS and lowendbox where the barrier to entry is much lower, and even as a satisfied Linode customer I continue to start and continue to maintain many projects in those environments.
One of the biggest costs of running a SaaS business is support. I'd guess that there's not a whole lot of difference for Linode in the cost of supporting a $10/month vs. $320/month box. In fact, there may be negative correlation -- I wouldn't be surprised if the cheaper customers are more expensive.
Now, your argument hinges on there being a high conversion rate from low-end plans to high-end plans. I'd guess that rate is quite low, in fact. Most accounts probably never upgrade and / or cancel after a few months. Now, the canceling in the first few months is a big problem as well, because for most SaaS businesses, your biggest support cost is going to be concentrated in those first months where the customer is getting set up. I wouldn't be surprised if it takes Linode a few months, on average (not median since support isn't evenly distributed), to get into the black on the low-end accounts.
The variables in that equation are really important in deciding if it's worth putting up with cheap, less knowledgable, high churn customers just in the hopes of upselling them over time, or if it's better to slice off the segment from the beginning that finds $20/month to be onerous.
The hourly, broken-up billing is a smokescreen; for exactly what you get out of a (now) Linode 1024, you'd pay $50+/month or so on AWS. Don't forget total bandwidth, an elastic IP, and so on.
The ongoing TCO of AWS is much higher than Linode, of course. But that's precisely my point: you want to capture customers at birth, incubate them into adulthood, and then charge them a lot of money now that they can afford it.
Although if you want to be pedantic, let's say "actually useful, potentially-production barrier to entry". There. Fixed.
I don't find my cheap t1.micro based IRC bouncer / VPN / reverse SSH endpoint / static-content generation box to be an awful joke at all. They're quite useful for many things.
There are $10 VPS solutions out there, but their poor reputations lends credence to the idea that $10 VPS solutions are not where the money is at.
Its not a smokescreen if you are using AWS for a cloud (e.g., dynamically provisioned according to need) server infrastructure rather than as a simple VPS.
Being unprofitable and not being able to spend $240 / year is a different issue.
The first is about the project being unprofitable. The second is about you being broke or not believing on it to spend even 1/150 of your income (I'm speaking of course for US/EU citizens/wages).
There are however other vendors that have even $5/month plans for VPS (and even offer SSD drives). Try Digital Ocean.