AT&T’s CTO, André Fuetsch, says their CAPEX “is not going up, it’s certainly coming down”, and that they’re already seeing significant cost savings from SDN/NFV deployment.

“If you look at the efficiency of the capacity we put into the network, just in 2015, 2016, we’ve actually put in 2.5x the capacity at 75 percent of the cost of what we paid back in ‘13 and ‘14. So you already seen the economics, the efficiencies coming in from these new technologies,” he said. “Going forward, we will continue to see the cost-per-megabyte transported to continue to drop. And we’re in the business of connectivity, and that’s what it’s really all about.”

Fuetsch further said that they might maintain the current CAPEX/revenue ration and use the spare money to make more investments in the network, giving little suggestion that it might be paid out in the dividend. The strategic importance of all this can be gauged from Fuetsch’s rise - having been the SVP responsible for the NFV strategy, he’s now the company-wide CTO and the president of Shannon Labs. John Donovan, meanwhile, has been retitled CSO and President of Technology and Operations, which may mean he’s going to be the next CEO.

Here’s a valuable corrective to all the virtual network rah-rah - Australian researchers benchmarked the major OpenFlow controller implementations on a variety of hardware and OS platforms and found most of them seriously weak. On a Tilera networking device, the best-performing controller couldn’t handle more than 17% of line rate. This went up to 68% for the combination of the Beacon controller (supported by Intel via the Floodlight project) and an Intel Xeon server - as the other controllers did poorly on the x86 machine, this suggests Intel knows something about packet processing in x86 they don’t, and that it optimised Beacon for x86.

No surprise, then, to find Intel working with AT&T to optimise packet processing in virtual networks on x86. Perhaps a bigger surprise was that the Cisco-championed OpenDaylight controller performed so poorly the researchers didn’t bother experimenting with it any further.

Intel also recently announced its first silicon photonics products, network cards for very fast (100 and 400Gbps) data-centre switching.

Deutsche Telekom has announced a new “pan-European strategy”, which turns out to mean creating three big data centres to concentrate its whole product line in virtualised form, and paralleling them with engineering and customer service offices in “production centres”. These will be in Hungary, Poland, and Greece, with the first one coming on line at the end of the year.

Very impressive, but you do wonder about this. Apparently they’re having a “close look at the corporate culture” to encourage ethical behaviour. Is there something they want to news-manage? Meanwhile, the company posted Q2 numbers - revenue was up 2.2%, EBITDA up 3.6%, but net profits down 12.8%. As usual these days, the growth was all coming from T-Mobile USA.

DTAG-SKT-VZ partnership: a sign of early 5G? US wireless >50% IPv6; small cells flop

More of those 5G partnerships are being announced. This one, between DTAG and SK Telecom, is interesting - CTO Bruno Jacobfeuerborn says that:

From our point of view, the next release of the official 3GPP specifications are set to form the basis of 5G

The next release, as it stands, is 3GPP R14, coming next year. Does that mean DTAG and SKT are joining Verizon on the early-5G bandwagon? Quite possibly, as SKT also signed a further partnership with Verizon itself, in parallel with their existing cooperation on 5G trials specs.

This statement from AT&T, meanwhile, has been read as a rebuke to VZ and friends for pushing ahead of the standards process. However, note that DTAG and SKT are explicitly listed in it as carriers co-operating with AT&T on 5G. Also, our previous link mentions that SKT and VZ will be cooperating on, among other things, the mobile variant of the Central Office Re-architected as a Data centre (M-CORD).

CORD is AT&T’s very own special contribution to network virtualisation, so you could make a case that there’s actually more cooperation than conflict in this message. Perhaps this line:

Linking trials to the standards process is the fastest path to large-scale global 5G deployment

actually means “get some agile development going, like we did with OpenMANO”? Interesting. Meanwhile, BT is getting in on the act with Nokia, now it’s a mobile operator again.

BT’s also announced that it will be providing its retail customers with IPv6 connectivity this autumn. Originally, it planned to reach 50% by April, but obviously that didn’t happen. Apparently you can expect a /56 prefix, dynamically assigned for some reason.

It’s been a big week for the global effort to fix the Internet addressing shortage - on the 17th, for the first time, Facebook observed more US mobile traffic from IPv6 networks than IPv4, while the World IPv6 Launch campaign confirmed that US mobile is now at 55% IPv6 penetration.

This is really interesting - MoffettNathanson reckons there are as few as 30,000 small cells in the USA. For context, T-Mobile USA is reckoned to have around 60,000 macro-cells, and AT&T’s 2012 Project Velocity foresaw 10,000 additional macro-cells and 40,000 small cells. We know AT&T deployed at most 20,000 of the planned 40,000, and may even have withdrawn some of those in favour of more capacity on macro sites after Leap’s spectrum became available. Sprint was fascinated with small cells for years, too. This suggests that small cells have seriously under-delivered in the United States - one of our conclusions in this research note.

Meanwhile, after Hans Vestberg’s sudden exit and a Q2 dominated by concerns about Ericsson’s accounting practices, the two huge family trusts that own the company and most of Swedish industry are after the chairman’s head too. The arch-rivals at Nokia have, meanwhile, bagged 30% of a huge China Mobile optical network contract. Ericsson, however, wins some sort of a hype prize - they tested a “5G drone” with CMCC, only days after Nokia tested a mere 4G drone with the carrier.

EU 25-year licences; OTT regs coming, or maybe not; Brexit vs Datacentre; US ready to hand over IANA

So, DTAG wants to hollow out its European OpCos and consolidate everything into those three production centres - cloud-based MVNO farms, basically, located very clearly in the cheaper bits of Europe. That sounds like it might have some regulatory consequences! And indeed, there’s a fair bit of regulatory action this week.

The European Commission is, according to a leaked document, thinking about a standardised European mobile operator licence, with a standard 25-year term and an EU veto on its terms. There would also be a voluntary option to issue multinational or even pan-European licences.

We can also expect a statement in September about the handling of customer data and possible regulation of OTTs. As usual, Oettinger has been briefing the German press that this is an exercise in giving telcos concessions in exchange for investment (e.g. here) although note the remarks at the first link from the Digital Single Market spokeswoman pushing back on that. The tug of war between Commissioner Oettinger and Vice-President Ansip goes on.

Meanwhile, British technology companies and telcos are lobbying the government to do something about the European arrangements that let you process personal data between countries. If the UK leaves the EU without replacing them, they may have to move the data and its processing. As 43% of EU data centre capacity in the top 4 markets is in the UK, that would be a bit dear.

The US government, meanwhile, is saying that it wants to wind up its role in the Internet’s governance as soon as the 1st of October, when the contract between NTIA and ICANN to provide the Internet Assigned Numbers Authority functions runs out. The US Department of Commerce proposes just to leave the job of managing IANA and the DNS root zone to ICANN. Stand by for a row.

Here’s your update on the FCC’s Business Data Services consultation. Public Knowledge has filed its concerns; Cox and the ILECs argue the cablecos’ Ethernet over DOCSIS services mean there’s plenty of competition; Sprint denies it.

Harold Feld blogs the legalities of a case that stops the FCC pre-empting a state ban on municipal broadband.

Comcast takes 1Gbps cable to Chicago, targets 6m DSL users; pay-TV disaster; Charter beats VZB; PE fund buys RCN

Here it comes. Comcast has launched gigabit cable in Chicago. Their VP of consumer services, Marcien Jenckes, makes it abundantly clear that they’re coming for the 6 million households still on telco DSL in their footprint. In Q2, cable net-added 550k broadband subscribers, 360k of whom came from the telcos, and 220k of those went to Comcast.

TV isn’t helping the telcos either - the pay-TV sector generally had a horrible Q2, but the telcos took the cake with a 500k net-loss. 391k of those were from AT&T U-Verse; even though DirecTV net-added 342k, it’s not enough to soak up the U-Verse exodus. It’s blow fibre or die.

It’s no different in the enterprise market. Charter has pushed on into third place for carrier Ethernet service by port count, edging out Verizon of all companies. It’s not just the mergers, either - 60 per cent of new connections went either to an independent or a cableco in Q2.

More upshot: private equity fund TPG Capital is buying both RCN and Grande Comms.

Jenckes also pointed out that Google Fiber had been slower to deploy than he had feared. That may be about to change - more and more 5G and other fixed-wireless trials at Google are seeping into the public domain.

A new independent FTTH deployer has made its appearance in the UK. Community Fibre has filed for code powers, saying it aims to reach 500,000 homes in London with 1-2Gbps fibre. Its biggest problem may be Openreach deciding it now needs to build FTTH in its areas, like they just did in Kensington.

And here’s an effort to model the impact of Long Reach VDSL on UK broadband coverage.

Intel makes ARM chips; Xiaomi shipments down 7m; Samsung = 82% of Android profits

Well, here’s a historic moment. Intel has agreed to manufacture ARM chips on its new 10nm fab lines. Up until now, Intel reserved its superb manufacturing solely to its own x86 designs; for the first time, it’s a contract fab for somebody else. In part this is just a nice-to-have way to get more utilisation on the hugely expensive machinery, but it’s also a confession of ARM’s triumph in mobile and Intel’s failure to make any headway with its own technology.

Of course, the biggest deal imaginable in the ARM world would be the system-on-chip for the next iPhone. Meanwhile, the Intel Optane memory and storage chips have been delayed after manufacturing problems with the DIMM (i.e. RAM) version of the technology, which builds up the circuits in three dimensions.

Xiaomi is one endangered unicorn (sorry, decacorn) right now. IDC reckons its shipments are down 7 million from this time last year, putting it down to fourth place. Management are claiming IDC got the sums wrong, often a bad sign. The company has abandoned its no-ads, web-only distribution model, which has given it a new problem - a marketing budget to feed.

It looks like Lenovo has a margin of -5% on each Android tablet it ships. Fortunately, volumes are declining!

Who’s winning, then? Samsung, according to Strategy Analytics, with 82% of the Android profit-pool, and a powerful new gadget in the Note 7. ZDNet thinks Samsung is likely to sweeten its rather toppy RRP ($800) with gimmes like free Gear VR headsets, hoping incidentally to lock users into a Samsung ecosystem. The case is made further here.

Apple earns about $1/customer/day all the time. It’s also sharply increased its spending on research and development.

Here’s Intel’s Project Alloy VR/AR device.

Son still wants Sprin-T; unlimited is back, max buyout hits $850; China Mobile, Unicom H1; MTS Q2

Masayoshi Son apparently still wants to buy T-Mobile USA. That’s understandable - who wouldn’t want to have a go on that company? - but he also wants to keep Sprint. In the ring, meanwhile, T-Mo announced the return of the unlimited data plan. Before everyone writes at once, there is a throttling level after 26GB/mo, but that’s a pretty huge bundle nonetheless. At the 26GB chilly limit, that implies a marginal data price of $2.88/GB, a substantial price cut compared to the usual $5/GB in its prepaid rate card.

Sprint responded with something similar, plus a massive slug of subscriber acquisition money - the maximum buyout is now up to $850, consisting of up to $650 towards buying out your early-termination fee and paying off device instalments, plus $200 in service credit. When you think that Sprint’s quasi-unlimited plan costs $60/mo for the first line, this starts to look more like vendor financing than marketing spend.

It looks like all the US national carriers will throttle rather than charging overage.

Verizon Wireless has a new line of business - give them $1-2 per install and they’ll preload your app on their Androids.

China Mobile’s H1 was strong - revenue up 7%, profit up 6%, LTE penetration surging. Unicom, though, saw their profits plummet 80%. 

Singapore’s sovereign wealth fund, Temasek, is selling a stake in Bharti Telecom (the eventual holding company for Bharti Airtel) and another in AIS to Singapore’s state telecoms operator, Singtel, which it partly owns.

EE wants a licence variation to let it use its 10MHz of unpaired/TDD 2100MHz spectrum for LTE, either for the Emergency Services Network or for wireless backhaul.

It looks like MTS is having some pricing trouble.

98 Facebook data points on you; Google’s new IoT OS; exit Avaya? exit Lyft?

Here’s a list of 98 data points advertisers can use to target you on Facebook. Also, if ad buyers can use them to target you, anyone with access to the targeting tool (which is nearly anyone) can find them out about you (see graph.tips for examples).

Google is building a new operating system for Internet of Things devices. It’s optimised for low power consumption and uses a capabilities-based computing security model. That sounds eerily like Symbian.

Wind River has a new cloud-based management system for IoT devices.

Microsoft acquires WebRTC-based game-streaming app Beam. They’ve also delayed the release of Azure Stack, the private-cloud version of Azure, and customers aren’t happy.

Is Avaya about to be broken up or even go bankrupt? Here’s their problem - you can make a very decent CRM/call centre solution out of web tools and Twilio. Having acquired Nexmo, Vonage is trying to pivot to being a Twilio competitor.

How Obama for America tested its IT infrastructure.

Scaling PayPal.

And Lyft seeks a buyer to make that unicorn valuation into cash…but finds it not.