In large volume, 2/10ths to 5/10ths a cent. That suggests large ISPs pay something less than a penny per gigabyte. If you use 139 gigabytes/month, that costs your provider something like $1/month. (Doubling transit costs gives you a rough estimate of the cost to the carrier, which also has to carry the bits to your local exchange.)

Akamai recently bid "$0.002 per GB delivered," according to Dan Rayburn, a streaming media expert I trust. Dan notes that's the lowest price he's ever seen for a content delivery deal, and other recent bids have been as high as half a cent. Prices have been driven down by large customers, including Apple, doing their own content carriage. Amazon is in the game today as well.

Hurricane Electric is advertising Internet transit at dozens of locations for $0.20/Mbps, which works out to a similar price.

40% last two years, 30% expected next two years. NBN Chief Bill Morrow sees the Netflix surge passing and future growth slowing. The average NBN user draws 131 gigabytes/month, reports Geoff Long at CommsDay. (Not linkable. CommsDay is an excellent news source, available only via a professionally priced subscription.) 

Until two or three years ago, landline data traffic for the average user had been growing at 40% +-5% for a dozen years. Fortunately, Moore's Law has brought down the costs of the necessary components at a similar rate of about 40%. The marginal cost per subscriber per month has stayed flat at all the large carriers: About $1/subscriber/month across the developed world. Wireless growth has been similarly slowing.

Shared networks can work remarkably well. Across the U.S., AT&T, Comcast, and Verizon are in a fierce marketing battle to offer fiber to new developments. Table stakes now are a reliable gigabit network and premium TV offerings. 400,000 new apartments go up every year. They add less than 1/2 of 1% to the housing stock but are a prime growth market. 

I suspect there's a slight exaggeration here, but 97-99% would make most of us very happy. U.S. & U.K. government testing (SamKnows) have long demonstrated that most cable networks are darn close to 100% delivery. In 2014, FCC tests showed 95% of Comcast customers received between 109% and 119% of advertised upload speeds.

If cable/DSL is good, people won't switch. Kevin McLaughlin's article is unconfirmed but makes sense. Update Someone who knows thinks this is false. The analysis stands whether or not there are cuts today.End update.

 I have 200/20 cable; Jennie 50/50 fiber. Both are rock solid and uncongested. For most people that's fine and Google's gigabit isn't worth switching for. Reported result: Cost per new customer blows out the economics. Fiber needs to win ?30% to 50% of the market. That's hard if the existing carriers aren't so bad. LTE by the end of this year will be at a gigabit, (shared) with more antennas likely to allow servicing more fixed customers.

LTE by the end of this year will be at a gigabit, (shared.) Many antenna MIMO will allow servicing more fixed customers even before 5G is ready. Comcast and Cox are promising a gigabit (shared) to half the country within two years.

People hate to switch.

AT&T cutting 80K jobs. Carriers are gung-ho for SDN/NFV because it will be more efficient. Tens of thousands of people will lose their jobs at telcos, cablecos, and suppliers. In the long run, the economy benefits from the increased productivity, as do shareholders. But many of those fired will never get a decent position again, especially older workers. 

There are no easy answers. AT&T is paying for extensive employee retraining, hoping linemen can learn to be programmers for cloud computing. They donated $million to Georgia Tech's innovative fully accredited online Master's in Computer Science and are paying tuition for employees. Some will be able to make the change. 

Nokia Will Make Massive Layoffs and Ericsson plans to lay off thousands are recent headlines. None of these are related directly to SDN deployments, of which there are very few so far.

Singularity in half the U.S.? Cable is now 50/5 or higher for nearly everyone. The numbers from Leichtman are inescapable; Cable is clobbering DSL in the U.S. The question mark after "singularity" is only because this is a single quarter; another quarter or two and the death march will be almost irreversible. Telco broadband dying across much of the United States wasn't inevitable. DSL is still beating cable in Canada, England, and France. 

For several years, nearly everyone except a few top analysts and the companies involved thought cable was decisively knocking out telcos. I frequently pointed out that telcos were doing fine in about ~50M homes with faster DSL (AT&T U-Verse, millions at Century) or fiber (Verizon FiOS.) The losses were coming in the about ~25M homes AT&T and Verizon hadn't upgraded in a decade or more. They intend to shut down landlines for most of them, expecting to be more profitable because many would switch to their wireless.

SDN is object oriented programming brought to telco networks. Adtran showed an impressive but still early version at their press event. Named Mosaic, it is already managing G.fast at trials in 65 telcos. The 40 gig NG-PON2 units Verizon is testing also run under Mosaic. From a single console controlling more than a dozen pieces of equipment, Adtran showed they can run "Configure, Deploy and Activate" on any of the boxes, not all of which are made by Adtran. 

Their Huntsville SDN demo was along a long wall. To the left of the picture was an 8086 type of Linux PC, about $5,000 worth, running ONOS and Adtran's Mosaic software suite. Underneath it were off the shelf boxes, 48x10G switches, 32x100G switches, and 16x10G PON OLTs. To the right of the monitor was an Adtran box running SHDSL; a 16 port G.fast box, 2 x 10 gig OLTs, 2 G.fast CPE, a carrier Ethernet unit, and gateways connected to the 10 gig PON. The switches were leaf and spine, from an obscure vendor in a white box.

The controller could have been a thousand miles away in the cloud.

Verizon planning business customers in 2017. Vincent O’Byrne doesn't believe 10 gig down, 2.5 up is enough for Verizon's future. He has begun trials of 40 gig down, 10 gig up NG-PON2 with Calix and Adtran, with the intent to start purchasing within a year. Boston's new fiber network would be a natural place for them; Verizon would be able to offer 10 gig inexpensively to the giant companies in Boston's financial district. The same gear will probably be natural for Verizon's 2018 test of 5G highband, which will need massive backhaul. mmWaves do not go far and will need cells every few hundred meters.

The same gear will probably be natural for Verizon's 2018 test of 5G highband, which will need massive backhaul. mmWaves do not go far and will need cells every few hundred meters.

Also called TWDM PON, the OLT delivers 4 wavelengths at 10 gig each. Calix says they can bond them, to deliver 40 gig to a customer.

"Gigabit without a backhoe. ... We don't have to dig up the streets and can go incredibly fast. You're going to see us go coast-to-coast in the next 12 months," Comcast CTO Tony Werner tells Amy Maclean in a short but important interview. I'm told there are unannounced Comcast deployments of DOCSIS 3.1 in Chicago and California, as well as the large public trial in Atlanta. I'm guessing that Comcast has already upgraded the CMTS for millions of lines, ready to turn on large systems when they believe everything is ready. (That's consistent with what manufacturers are shipping but I don't have confirmation.)

Cable supplier Arris expects substantial sales of DOCSIS 3.1 modems starting early next year, which is consistent with Tony's prediction. Note that the high speeds are likely to be download only. Upload looks mostly to be 90% slower.

Drahi's Altice also is financing Layer3. They are offering a 4K set top with up to three terabytes, apparently designed to connect wirelessly to a simple network interface device. That avoids needing to wire in the home, which Fran Shammo says is often the largest cost in a Fios connection. They are about to launch in a Chicago neighborhood and are also hiring installers in Alexandria, Virginia. 

Cable tech legend Dave Fellows and partner Jeff Binder have raised $100M, hoping that Comcast, Verizon, and AT&T have loads of unhappy customers who will switch. Paulson is a gambler who made $4B betting that triple-A rated mortgage bonds would default in 2008.

Takeaways: The strike contributed but fundamental problems remain. They have been cutting capex since 2007 and they no longer have a major edge in quality. VZ is and will be a very profitable company, but maybe not worth $200B. Wireless data growth is way down (< 40%) as I've been forecasting for several years. 5G Highband will start rolling for fixed wireless in 2018, but probably slowly. It will not add to earnings until well into next decade. The CFO forecasts very slow revenue growth. McAdam thinks best quality, video and IoT growth, and large content revenue will solve these problems. I raise doubts. 

China is now 1/3rd of the broadband Internet. Over 240M are now connected by fiber home. The 3M added by CM in June was more than the entire rest of the world. It surpassed the previously monthly record, 2.65M at CM in March. China Telecom added 590,000, China Unicom 250,000, and China Mobile 3,189,000, for a total net gain over 4M.  

China Telecom's landline broadband total is 118M, Unicom 74M and Mobile 66M. Add about 20M for the smaller operators and the total is 278M. That's more than the EU 28 (166M) combined with the U.S. (103M). Nearly all are on fiber home or fiber/DSL. In two years, almost 100M have shifted from pure DSL to fiber.

LTE in China in July reached 600M LTE subscriptions.

Windstream, Fairpoint watch out. T plans 3M lines of fiber to the premises (including G.fast) each of the next four years, expecting to get to 20-25% of their territory. Some of them will be "out of territory," including some of the ~2M homes they have reached with fiber already. I can't recall either AT&T (almost half the U.S.) or Verizon (about a quarter) expanding this way in the last decade. 

AT&T, France Telecom, and Telefonica have discovered that fiber costs are down in favorable locations. Google Kansas City has proven fiber can be profitable, including pulling many customers from AT&T. Fiber in some places can be run for < $500/home; in others, the cost is $4,000-$5,000. It's a great business if you can cherry pick areas with low costs and weak competition.

That's exactly what AT&T is doing,

Sharon White can raise Internet speeds across Britain by 100 megabits with smart policy.  British Telecom, Vodafone, Sky, and TalkTalk would all benefit if White gets this one right. More important, the Brits get a much better Internet. The only losers would be John Malone's cableco and the now outdated policy for sharing of the binder. (BT has promised 10M lines of G.fast at speeds of 200-800 megabits. My examples here are from Britain, but other countries will face similar issues.)

Competition doesn't have to die. It does have to move from the cable to the exchange. Unbundling was a great idea in 2000 and is the main reason European broadband is 30-70% cheaper than American and Canadian. The 25 or 50 twisted pairs that run together from the exchange could be split between several carriers and performance was unaffected.

Future competition will require BT's local loop monopoly to hand off to others at the local exchange at the right price. The "right price" would include a profit proportionate to the investment. BT would be welcome to recover costs plus a profit; regulated prices would prevent exploitation of the monopoly.

That was the near-ubiquitous model for unbundling until recently. Currently in the EU, monopolies are being allowed to charge far more, after Kroes and Richards were blinded by a lobbying campaign brilliantly orchestrated by Luigi Gambardella.

Outside of China, the entire industry may have been negative for years. I knew the chip business was struggling but this chart is startling. For the last five years, growth has been only ~3%. For the last ten, only ~4%. After inflation, everything stagnant. 

For 2016, the new forecast is -2.4%, with the largest drop in the Americas. They forecast +2% in each of 2017 & 2018. 

China has committed $60B plus to the domestic chip industry and is sure to take market share. Including Taiwan, that's the industry future. There are virtually no interesting chip startups in the U.S. The San Jose area continues booming with Google, Facebook and software/web companies but no hardware. 

Silicon Valley is dying.

U.S. & Canada fixed 2016 +22%, 2020 +15%. The 2016 growth in U.S. & Canada mobile data looks to be 49%; the 2020 prediction is 38%. The VNI sees a major slowdown in both fixed and wireless growth around the world. Cisco estimates 2016 worldwide mobile growth to be 68%. They expect that to fall to 41% in 2020 when most people will have smartphones. Remember the world turns so no five-year forecast is certain.

Check the chart below and extremely rich data here and here. There's so much information I could spend a week in the analysis and still not catch all the inferences. Accuracy can't be perfect but they are clearly the best in the business.

The 100% mobile growth was an artifact, driven by people first getting smartphones. The average smartphone user is increasing data usage ~40%/year, per AT&T.

Careful: Shortage fears can lead to over-ordering and create a real shortage. Unless people panic, the problem is unlikely to be more than a few weeks delay in an order. Tan's comment in the financial call ("We're also working on resolving a few potential supply constraints in this particular segment." Details below) is likely circulating in the industry, and this article might. The instinctive action is to panic and double order, which is a pattern we see in chips on a regular basis.

Broadcom is the largest supplier of chips for both DSL & cable. The comment surprised me because the primary foundry, TSMC, is missing sales targets and generally has capacity. I haven't seen any huge increase in demand for broadband chips. Broadband subscriber growth in Europe and the United States is modest. Liberty Global is just getting started on increasing their cable coverage. Deutsche Telekom is two years behind schedule on it's big vector expansion. 

China added an unbelievable 50M fiber connections in 2015 and Hock says that trend is continuing.

March's 2.65M new subscribers at China Mobile is almost certainly a broadband record. I don't remember anything close. I checked 10 years of data from China Telecom, by far the world's largest broadband provider and didn't find a peer. Most of them are on fiber. China Mobile consolidated wireline subsidiary China Tietong and now is going full speed ahead.  China Telecom added 3m to 116M; China Unicom 1.1M to 73M. Add 18M cable modems and a few small players and the current Chinese subscriber base is over 270M. 

Most new Chinese connections are on fiber: China added 50M fiber subs in 2015 and entered 2016 with 130M fibered. China Mobile consolidated wireline subsidiary China Tietong and now is going full speed ahead. With 800M mobile customers, they have plenty of prospects. 

Q1 in the U.S. was 1.1M 

Update 7/22 Removed total of fiber customers. Looking at new data, I'm not sure how many are fiber home and how many fiber/DSL and didn't want to provide a figure. 

Cable added 1,000,000 DSL/telco fiber 10,000. The data below are from Bruce Leichtman, who has been consistently accurate. The trend to cable is clear in the 50M homes that are still running DSL versions that are ten years out of date. AT&T & Verizon are doing fine in U-Verse and FiOS territory but are getting clobbered elsewhere.

DSL beats cable in Canada, France, and England. The U.S. policy types who say a cable monopoly is almost inevitable are uninformed. AT&T & Verizon have decided to go wireless-only across about 25% of their homes. They virtually stopped investment in those areas and have been very open with Wall Street about their plans. The wireline business is profitable in those areas but wireless-only makes more money. Or so they expect. 

China Mobile added 2.65M new wired subscribers in March.

"Executives estimate that eventually AT&T could get by with one-third fewer workers," and, "senior executives say shrinking the work force by 30 percent is not out of the question." Quentin Hardy buried those quotes in a New York Times article. 30% of AT&T's 280,000 employees is over 80,000. My guess is that 50,000-60,000 is more likely, but I believe even 50,000 would be the largest cutback since Elisha Gray and Alexander Graham Bell invented the telephone. I hope this story will prove inaccurate but the sources are solid. In addition, moves like this may be necessary for AT&T to increase profits. 

Randall and his team have often seen trends before the rest of us. They are not cutting back to prop up earnings for a few years and get the stock price up. The analytic question is why and how. I have some of that below but I'd welcome opinions.

This is one of the biggest telecom stories of the decade and no one picked it up. When I mentioned it in an article, I got a disbelieving email from a senior executive at AT&T. The New York Times could have made a mistake, but Hardy is a veteran reporter who had more than one source. 

CEO Randall Stephenson in the same article preached the importance for AT&T workers to "Learn new skills or find your career choices are very limited."

It works, finally. As always, the telcos were cautious and the vendors promised too much. The dam has broken now, with BT going at 3M homes passed each year. http://bit.ly/BThalffast CEO Patterson will keep going until “almost all” Brits can get what he’s calling ultrafast broadband.

AT&T looks to be the next big order, millions of lines with fiber to the basement and G.fast to each apartment. http://bit.ly/GfastATT. Chunghwa in Taiwan is covering almost the entire country. In Australia, Parliament is debating how much of the National Broadband network should switch to G.fast. Rio de Janeiro, Panama, Belgium, and Switzerland are on the way.  

Much more at Gfastnews.com

Software unbundling could add 100 megabits to typical speeds in England. BT has just committed to 10M lines of G.fast in the next four years and "most of the country" soon after that. They are talking speeds of "up to 330 megabits" rather than the 500-800 now proven for G.fast from the local distribution point. Instead, they will put the DSLAMs in existing cabinets and save. 

Cioffi, who invented vectored DSL, believes using software unbundling would allow 100 megabits faster speed. Alternatively, you can extend the reach at a given speed. Traditional unbundling shares the physical lines in the bundle. Today's technology - vectoring and G.fast - provides much higher speeds but is very sensitive to other signals in the same bundle. The incumbents are saying hardware unbundling is now undesirable. They prefer to control everything and sell bitstream access.

ASSIA's multitenant software allows multiple companies to manage the parameters for each customer and troubleshoot many problems, The carrier would manage the physical network including the vectoring noise cancellation. Each company would have a management console and could support their own customers directly.  

BT's current plan is to run G.fast in frequencies from 22 MHz to 106 MHz. That leaves 0-17 MHz for other companies running older VDSL on unbundled loops. If the companies could agree that all would use the BT G.fast local loop, British homes would get 100 megabits more speed. BT would be able to use frequencies 2 MHz to 106 MHz rather than 22 to 106 MHz

Africa in 2015 has ~400M people with mobiles despite no signal where almost a third of the population live. Ericsson counts 170M smartphones in sub-Saharan Africa in 2015 and expects a remarkable 500M more in the next five years. 3G coverage in Q1 2016 is only 43% (GSMA) but increasing rapidly. The first map, from Afterfibre, very colorfully slows the emerging fiber networks. (Red are projected.) The third, from Paul Hamilton, has similar data but less color. It's easy to see the fibre desert in Central Africa. #2 is from the remarkable Steve Song, showing the undersea cables. Almost all have been built in the last five years. South Africa is served by five cables; Kenya and most of West Africa by three. Prices are a fraction of what satellite used to charge but in many places still very high. Larger versions below.