Breakdown: Fiber circuit, up to 1gpbs, 2-3k/month. Should support up to 500-700 customers at least. Backhaul - $2k-12k each. Usually need about 1 for every 300-500 customers depending on network topology. Access Points - $200-$600 each. One for every 30-50 customers. Switching / routing hardware - $300-$1200 should cover up to 500 ish customers.
Per customer expenses: $200-$350 - customer CPE equipment and install accessories (cat5 cable, etc.) $3-$10/month - billing/customer management software, 3rd party phone support (if you do that), other misc stuff.
Then you need a ladder and tools and a vehicle that can carry them for doing installs and maintenance.
Obviously the risk depends on the competitive landscape. I've seen some people start by selling an apartment complex or office building or similar on an exclusive / semi-exclusive deal to provide service to the whole complex (which you can do considerably cheaper per customer than the above numbers) and use that to cover the basic expenses and expand from there, that's a good way to do it. In most places in the US people are unhappy enough with Comcast et al to give something else a shot. One issue is that people are usually in year contracts so you have to wait that out. If you have an existing consultancy or some community goodwill that you can leverage in to a customer base that decreases the risk, obviously.
How is the connection shared between customers? Can one kid downloading blu ray rips over NNTP eat half of that 1gbps?
What's the price range for the end customer?
4:1 on a 1Gb connection gets you 4Gb/s. Divided into 500 connections is 8Mb/s per connection. I've fought this battle before on HN, but everyone does not need a 150Mb/s connection. Yes, some people do, but most do not.
I haven't checked recently but last time I looked an HD Netflix stream used about 5mb/s of bandwidth. I think your assumption that half of your subscribers would be streaming video is a bit on the high side -- in some communities it may be accurate but not most by my observations. For planning, I would say 10%, maybe 20% tops.
If you're selling a 100Mb/s to residential customers, you could probably even oversell at 8:1 or higher, since their usage compared to total will be much lower.
Your oversell rate will also heavily dependent on how much your base is commercial/business customers. Residential use is peanuts during business hours, business use is peanuts during residential hours. So you can almost oversell by a factor of 2x over your normal rate for business vs residential, and it's also been my experience that as a whole, businesses use far less bandwidth than residential.
There are some outliers that will really skew your numbers. If you sell a 100Mb/s connection to a college that is providing it to their dorms with 1000 students, you can almost guarantee they will be nearly continuously saturating their connection.
Edit: adding answer for 2nd question. For the most part, without any throttling or queuing users downloading will get an equal share of the bandwidth available. 10 simultaneous users would get 100Mb/s, 100 simutaneously downloading users would get 10Mb/s each.
Upload is what will kill your bandwidth. One user saturating your connection with outbound traffic will slow everyone down.
But most ISPs queue/rate limit customer traffic. It's integrated into most commercial wireless equipment and routers. On top of that, depending on what technology you use, CPE(customer premise equipment) generally doesn't have the ability to pull more than 100mb/s. Some are lower, a few can go up. I'm not a Mimosa fan yet, but there is a guy woth a usually well informed blog[1] that sings their praises in providing multi-hundred megabit service to residential customers. My experiences with Mimosa gear have been such that I can't recommend them right now, but they have potential.