Our current Internet lifestyle is unsustainable and will crash sooner rather than later. I say this with no sarcasm: if I had to pay for everything I used online, I would be broke. $5 to HN, $5 to Reddit, $5 to ArsTechnica, $5 for Gmail, $5 for Google Docs, $5 for Google Search, $5 to Github, $5 to Medium, $5 to Wordpress, $5 to StackOverflow, $5 to CNN, $5 to NYT, $5 to Wikipedia, $5 to Weather Channel, $5 to each site I click on from HN or Reddit or Wikipedia, we could be talking hundreds of dollars per month.

You could make your argument about anything. "If Reddit is so important to you, why can't you pay them money?" "If HN is so important to you, why can't you pay them money?" "If Google is so important to you, why can't you pay them money?" You'd run out of money quicker than you'd run out of websites to pay, or you'd seriously scale back what websites you visit.

It's coming crashing down on us that either you have to pay for something or that something goes away. And we can't pay for everything.

Maybe the next Rails or Node can focus on making the cost of hosting infrastructure so inconsequential that sites don't have to make that choice.

> Maybe the next Rails or Node can focus on making the cost of hosting infrastructure so inconsequential that sites don't have to make that choice.

Isn't the much larger cost driver the salaries of the employees than the infrastructure cost (serious question!)?

Depends on what you are hosting.

A text-based site like reddit is almost nothing to host. All of the data is stored as strings, and only references media hosted elsewhere.

HQ Video and Audio files are still expensive to host. Not even just hosting, but bandwidth can be killer. Look at S3 bandwidth costs.

Costs of bandwidth vary a lot worldwide:

> https://blog.cloudflare.com/the-relative-cost-of-bandwidth-a... [2014]

> https://blog.cloudflare.com/bandwidth-costs-around-the-world... [2016]

An alternative direction would be for everything to be decentralized (whether federated or peer-to-peer). That way, if reddit/HN/Google/etc. (or perhaps some new equivalents) are important to you, you can set aside some resources on your computer to let them run.

I would be very careful with this kind of statement: Even if it were technically feasible, this would quasi require trusted computing (https://en.wikipedia.org/wiki/Trusted_Computing), since the company can never be sure that you won't influence some computations. This is something that probably every hacker should oppose.

By "run", I more meant "distribute data" (probably with digital signatures to detect tampering, and probably also with encryption).

Even in the sense of "running" sites in a computational sense, projects like Ethereum and BOINC seem to suggest the feasability of such a distributed model without necessarily requiring TC. If the computation itself is actually performed in parallel by different nodes, then outlier nodes (e.g. ones trying to tamper with computation) can be ignored.

Of course, there's the possibility of an attack where the attacker has control over the majority of the nodes running that particular computation. Hopefully such a system could grow to the point where such an attack is infeasible.

Even with that aside, though, most websites nowadays already run to some extent on untrusted hardware (thanks to JavaScript), thus warranting measures to handle untrustworthy IPC.

All in all, it's a hard problem, but not one that's impossible to solve. We already have quite a few of the tools on hand.

> Even with that aside, though, most websites nowadays already run to some extent on untrusted hardware (thanks to JavaScript), thus warranting measures to handle untrustworthy IPC.

From a security perspective it is a very bad idea to rely on untrusted data (i.e. that only passed "client validation", but is not validated again on the server). In other words: If the client tampers with some of its data, only it itself should be affected from any problems that this causes.

Agreed, hence the clause "thus warranting measures to handle untrustworthy IPC".

So the question here would be one of replacing server-side verification of untrusted data with peer-to-peer distributed verification of untrusted data.

> but its so valuable to them that they can't afford the equivalent of one latte per month?

Ignoring the fact that the prices in the student café in Germany where I get my coffee specialties are much cheaper - which is why I find this comparison unsuitable: All these equivalents of one latte per month sum up when concerning multiple services that think similarly. And these individual "one lattes" sum up over many years when they are recurring.