- 250 million 4G connections in 25 months at Reliance Jio India.
- 344 million fibre to the home connections bringing China gigabits
- 5-10 times more capacity with Massive MIMO and Carrier Aggregation, 4G or 5G. Real speeds 100-400 megabits.
- Verizon's millimeter wave, the most advanced network in the world.
- Wireless costs falling 40%/year.
- Google and Facebook, web-scale giants building their own international backbone.
- 5G is just a name; 80% to 90% of connections little faster than 4G. More below.
- 10G fibre almost as cheap as GPON.
- Almost no 5G phones until Samsung and TSMC build more 7 nm capacity.
- Still to come
J., a brilliant and successful Internet entrepreneur, was shocked when I told him what was happening around the world. He wanted to know more about changes that matter to his companies. I created a presentation.
Mukesh Ambani, India's richest man, has spent US$30 billion on a state of the art, ultra-efficient 4G network now reaching 96% of India's 1.3 billion people. More than 10 million are signing on every month. Average speeds are 20 megabits. Although the average customer pays only a few dollars per month, Jio is profitable. Data costs have come down 90% and India suddenly is an Internet superpower. Coming next: 50 million fibre to the home connections, three times as many as the U.S.
I was very skeptical when I first saw these figure in company financials and government reports. But analysts tracking the number of chips sold in China confirmed the figure. Five years ago, the government decided that a great Internet would help the country. It told the companies to build. They did.
5-10 times more capacity with Massive MIMO and Carrier Aggregation, 4G or 5G. Real speeds 100-400 megabits.
Massive MIMO, invented by Tom Marzetta at Bell Labs, uses 64 or more antennas to increase performance. Softbank Japan and China Mobile started deploying in 2016; results are typically 3-7 times as much capacity. The antennas steer each of 8-12 separate signals. (Beamforming.) Massive is spreading rapidly in 2018 and 2019, especially in frequencies above 2100 MHz. 3.5-4.2 GHz bands, previously impractical for mobile, work well with MM. Telcos just paid over six billion in the Italian auction, mostly for 3.5 GHz mid-bands.
Carrier aggregation allows using multiple spectrum bands. For example, AT&T uses 700 MHz, 1800 MHz, and 2300 MHz. Originally, LTE was limited to 20 MHz "carriers" in a single band, yielding speeds in the tens of megabits. With aggregation, 5 or 6 carriers can be joined, leading to speeds nominally over a gigabit. Speeds of 100-500 megabits are now becoming common around the world.
5G's original definition included a peak speed of 20 gigabits. That could only be achieved with the massive amounts of unused spectrum in higher frequencies. Verizon is using 400-800 MHz at 28 GHz, compared with the 20-100 Mhz practical in lower frequencies. The result is customer speeds that often are over a gigabit, although the first homes are only getting 940 megabits. Unfortunately, in early 2018 the industry changed the definition of "5G" to include almost all systems now shipping if a software tweak (NR) was added. It's 70%-90% slower than the mmWave Verizon is using.
As I write in October, 2018, Verizon is the only telco in the world committed to a large deployment of millimeter wave. All the rest of "5G" is much slower.
The efficiency of new equipment, with MIMO and carrier aggregation, has been increasing at a furious rate. Verizon CEO Lowell McAdam estimates the cost per bit is only 1/10th the cost of early LTE. Lee Hicks of Verizon in August, 2018, said his costs are coming down at 40% per year, which he expects to continue. Verizon has maintained profitability despite high traffic growth and unlimited offerings/ That would not have been possible without the cost drop. As traffic growth slows - Cisco estimates 30% in 2021 - capacity will outrun demand.
Mobile competition determines prices, but this cost drop is so large price/bit has and will come down. Internet companies can count on constantly lower cost per bit, allowing higher definition video and other services needing bandwidth.
Both companies are now so large they are building undersea fiber networks with practically unlimited capacity. They lead some of the groups and long term lease on others. The giants are almost impossible to compete with, with market-dominant shares of 70% to 90% in most of the world outside China. The near-zero marginal cost on their fiber gives them yet another advantage.
5G is just a name; 80% to 90% of connections are little faster than 4G.
5G millimeter wave is a big jump, about three times as fast as the best 4G. The claims for "5G" in low and mid-band are scams; the capacity improvement is 15%-50%, far less than the gain in 4G from Massive MIMO and carrier aggregation.
5G was so effectively promoted that by early 2018, everyone wanted to jump on the bandwagon. So the industry decided to rename all new 4G systems 5G if they added the NR software. 5G in low and midband uses the same equipment as 4G, but the new name has bamboozled most regulators and reporters. Engineers like Telefonica CTO Enrico Blanco did not want to accept the change, but the marketing people and lobbyists have won. Resistance is futile.
Engineers don't care whether the systems are called 4G or 5G. The network buildout remains the same: throughout the developed world, most networks are being rapidly upgraded to deliver typical speeds of 100-400 megabits and nominal speeds of a gigabit. It was called "Gig LTE." The same systems are now called "5G NR."
The new systems have the capacity and low cost necessary to meet the increasing traffic demand. They will be deployed almost ubiquitously but do not require a meaningful increase in investment. But bamboozled regulators are being pressed to give telcos tens of billions in concessions, The lobbyists say "5G won't be built" without the giveaways. Pure b______.
Xavier Niel's Salt in Switzerland and HKB in Hong Kong are selling 10 gigabits for the same price as others sell 1 gigabit. One large incumbent telco plans the same in 2019. Huawei tells me the difference in price is only $5-10 per connection in large quantities. I almost left this off the list because very few people have much practical use for even a gig, much less ten. But customers are buying and smart telcos will deploy.
There will be 5G phones by Xmas according to top researcher Ted Rappaport. That may prove true, but they will be very few. Verizon CEO wants them ASAP, but says his suppliers won't be ready until the middle of next year. His CFO didn't think even Verizon would have many before yearend 2019.
The problem is that 5G chips require the most advanced process on earth, called 7 nanometer. Linley estimates 5G chips are ten times as complicated as 4G. Without 7 nm, they would require too much power and run unacceptably hot.
This week, Samsung announced they were finally moving into regular production of 7 nm. Until now, the only vendor was TSMC. Apple has purchased 75% of TSMC's capacity for iPhone chips. Huawei has purchased another 10%, most of which will go to the remarkable Kirin 980 for gigabit LTE. TSMC is doing everything it can to build more capacity, including deploying EUV steppers at US$120 million each. But it will be well into 2019 before they catch up.
Samsung will be challenged to produce enough at 7 nm to meet its own needs. It won't have much capacity to sell to Qualcomm for a while.
Intel is the only other chipmaker capable of producing chips at this level, although they call them 10 nm. They are having severe problems bringing this generation into production and have pushed back delivery until late 2019. Global Foundries has given up on developing 7 nm, which would have required a multi-billion dollar investment.