Courtesy of Karl Denninger, The Market Ticker
Ok, this is getting tiring.
Look, everyone who follows me knows my view on Netflix. It is my considered opinion that their entire business model is reliant on poaching costs that should be theirs from other people, specifically, ISPs.
Those who wish to argue otherwise cite anecdotes and claims from the company. But none of these so-called "analysts" and media personalities (including Cramer) have bothered to spend a couple of hours running actual figures and facts.
Why not?
It’s not that hard.
Let’s take some facts.
- Each SD-viewing Netflix subscriber requires about 2Mbps of transport bandwidth for the entire time they are viewing the stream. Each HD-viewing Netflix subscriber requires about 4Mbps of transport bandwidth for the entire time they are viewing the stream.
- Each concurrent viewer requires this bandwidth for the entire period they are watching. That is, if a person watches a 2-hour movie, they require that bandwidth for the entire two hours, and anyone else who also watches a movie during that two hour period will add to the total aggregate bandwidth required during that time. So therefore, once the first subscriber starts watching a movie, we must then add to that load all other subscribers who initiate watching a movie during the subsequent period. We’ll call it 2 hours, as that’s about right for a full-length feature film.
There are carrier-class services you can buy that can also be priced. These carrier-class services are exactly the sort that an actual ISP purchases. They’re available from multiple companies and, quoted correctly, separate out all tail and last-mile costs – that is, they’re pure bandwidth buys.
So here’s your assignment, Netflix bulls. You’re going to justify your belief that the company isn’t poaching bandwidth. To do so you’re going to go get the following quote from no fewer than three national carriers with gigabit or better national networks. You will qualify them through the following RFQ requirements, and will publish each of the quotes you receive along with any terms and conditions received. There are about a dozen national providers who can meet these requirements. Pick at least three. In addition, pick three major cities (defined: population > 1m) in which your hypothetical service will be installed. The RFQ you issue will contain the following elements:
- One clear-channel DS-3 port, 44.7Mbps full-duplex. You will bring your own tail circuit from the carrier’s specified location to your location at your cost and expense. If the carrier has a requirement for a given tail circuit provider they are excluded (due to potential kickback concerns); there must be multiple CLECs permitted entry. If there are, and there’s a difference in cost, the highest cost controls (again due to kickback concerns which are rampant.)
- Of your three named large cities for installation, if any quote is more than 5% different in any of the three cities or pricing relies on more than one buy the carrier is excluded from consideration. When you issue the RFQ you will specify that you may buy any, all, or none of the three cities. This requirement excludes regional carriers that can’t actually provide the service you’re quoting (but who will try to claim they can if you don’t include a sufficiently-strong set of filtering requirements.)
- You are buying full, unrestricted, resale-permitted, clear-channel DS-3 transport and Internet routing. No ATM encapsulation, no games, no CIRs. Clear-channel means clear-channel – 44.7Mbps. You may do anything legal with that bit transport service you are purchasing, including reselling it. In fact your RFQ should state that you actively intend to resell it.
- You may, if desired, receive and forward full BGP routing information if you wish to apply for and use your own ASN to achieve diversity routing. The carrier in question will accept and propagate CIDR routes of /17 or larger. IP addresses will be assigned through ARIN and SWIP’d to you subject only to ARIN’s requirements and without passthrough cost. (This is a declaration of your intent to serve commercial customers, multi-home and be a reseller – that is, an actual ISP. It will eliminate the "yahoo" companies who intend to play games with you – they won’t agree to this.)
- The carrier shall not have overcommitted its infrastructure to at least three regionally-diverse national meet-and-peer points which they shall name and to which your entire 44.7mbps, in whole or part, shall be routable at all times. (Incidentally, I’m aware you can’t prove this except by later lack of performance, which is the next bullet point. Including it in the RFQ puts the carrier on notice that you’re going to proactively test and monitor performance and that you expect them to deliver every bit-per-second that you buy down their backbone and to exchanges all the time. It’s not a BS requirement, although some people will tell you it is.)
- The carrier shall agree to a three-nines SLA (99.9%) for both uptime and performance under the above. (This allows a bit less than one hour per month where the performance standards cannot be met.) For any breach the current monthly charge is abated. Upon two breaches within 12 months you as customer may cancel at your option without penalty or cost. The SLA shall not apply to your tail circuit which will be your responsibility (since you’re buying and provisioning it.)
- Contract term is no more than 24 months (nobody in their right mind agrees to more given changes in the industry; 12 months is better. If they’ll quote it get a 12 and 24 month commit.)
- Install (port and provisioning) charges are to be broken out and not "rolled up" into the monthly. That is, the installation charge and monthly recurring are to be separately stated on the quotation. (The install is a sunk cost and the port and provisioning does cost money. So will the hardware on your end, but you’re going to buy that on your own. The breakdown is necessary so you can compare actual transport costs only and ignore the rest in your analysis.)
Note that this quote does not does not include any "last-mile" costs of any sort, it does not include your infrastructure, billing support, customer premise equipment, your aggregation and distribution equipment (e.g. DSLAMs, cable Head Ends, etc), nothing.
You’re quoting a pure transport purchase.
For each of these that you buy, you can serve about 25 simultaneous Netflix watchers or ten HD-quality watchers. We’ll assume that the first person begins to watch a movie at 7:00, so the bucket on your ISP of people who get added to this includes anyone who then begins to watch a film between 7:00 and 9:00 PM. This is a reasonable approximation of peak load requirements.
If you prefer to use a higher-speed offering such as OC-3, OC-3c or similar, have at it. Only the divisor changes. Intentionally-excluded are so-called "carrier Ethernet" drops for a number of technical reasons that will make most of the people reading this spin their heads; they are not comparable on a pure multiplication basis for "hard data delivery" requirements such as is the case for a service such as Netflix. The service you quote must be private-line on a clear-channel basis such as a DS3 without an encapsulation such as ATM inside it. There are complex technical reasons to require this, most of them having to do with how certain "popular" protocols like ATM degrade when you approach the contracted bitrate. This inures to the benefit of the carrier and to yourconsiderable detriment, and for this reason your RFQ must explicitly prohibit such games by the carriers or you will buy something and find that you can’t actually pull the promised bandwidth from it.
If you cannot produce these quotes in your so-called "analysis" then you haven’t performed analysis.
You’re arguing from ignorance, someone’s talking points, or taking the word of others who have refused to produce evidence to back up their position and belief.
There’s been a lot of this on my forum, most of it whining, and plenty over on Seeking Alpha as well. Nobody has produced these figures to justify their claims. In order to make the Netflix model "work", assuming the customer’s Internet service costs $50/month and all they use it for is Netflix, and the carrier has NO last-mile cost nor ANY embedded expense in serving them the carrier must be able to buy each of the units of 44.7Mbps, assuming we’re talking DS3s here, for no more than $500, assuming the user is watching in HD, and no more than $1,000, assuming the user is watching in SD.
Note that this is the best-case analysis for Netflix as it ignores tail-circuit and distribution costs but those costs are real. Those tail costs are what ultimately make it worth it to pull your own glass and similar if you have rights-of-way (as cable companies do) but even if you own the ROWs and install all your own stuff instead of paying someone else to do it the cost exists and we’re intentionally ignoring it to get the absolute price floor below which one cannot sell and remain in business.
In addition this analysis ignores those Netflix users who have more than one device running at a time, even though Netflix permits this. That is, a "user" is presumed to be one account even though its entirely possible that there could be three people in a given residence watching three separate Netflix streams at once.
For these reasons this analysis is the most-favorable possible for Netflix’s arguments, as all of the possible confounding factors, known and unknown, are intentionally presumed to break for the Netflix bulls.
If you cannot hit these numbers then there is a cost-shift that is occurring – that is, the load that Netflix imposes on the ISP is being charged off on the ISP and not paid for by Netflix.
The argument that the customer bought the access and therefore it’s not their problem is a vacuous one. It is immaterial who bought what. What matters is whether the end-to-end bandwidth required to provide the service is paid for or not.
If the answer is "not" then someone’s bill is going to eventually rise. Either Netflix will find themselves forced to pay for transport by the ISPs or consumer prices for Internet access that include Netflix streaming are going to go up.
Either or both will happen until the cost equation for providing the bandwidth balances.
It is my position that building a business model on "taking the slack" out of someone else’s provisioning is a cute trick but is absolutely unsustainable and when, not if, that slack comes out your ability to play that game will end.
Those who wish to argue the other side need to come to the debate armed with their detailed quotes form those carriers and place them, along with the contractual terms, upon the table so they can be analyzed to determine if they really do represent a pure transport commitment sufficient to serve those customers.
If you’re an "analyst" or "pundit" and either can’t or haven’t come to the table with those figures, then it’s time to sit down and shut up.
