AI: Nvidia, AMD, Intel Similar

Databricks tests have similar results

Databricks, a large service bureau, has the scale, expertise, and incentive to do things carefully. There are small differences that would change with a different set of tests. These numbers are strongly suggestive of very rough parity between the chips on the market in December 2023. Intel & AMD are not far behind Nvidia,

Continue reading

AT&T’s Scary 300% Hurdle Model

AT&T is cutting capex, usually a mistake in telecom. Chief Data Officer Andy Markus told WSJ “AT&T aims for at least a ‘3x return on any AI investment,’ ideally within the first year of deployment. …“Every dollar you invest with us to create an AI solution … should return three or more back.”

A while back a friend at AT&T complained that he was required to get his money back in a year. This is much worse. Even spread over 2 years, this is a higher hurdle rate than I recall seeing.

If Markus didn’t misspeak, AT&T is screwing shareholders who don’t sell very quickly.

Continue reading

From 2017: Maintaining the Internet connection with autonomous networks

As a propaganda campaign about Internet splintering grows, I thought to reprint this

The question: how to arrange “robust interconnection” of the Internet without being dependent on ICANN. Vladimir Putin is creating an alternate root, with support from India and China. Columbia Professor Eli Noam convinced me a “network of networks” was possible and could be a good thing in theory. I doubt ICANN will actually shut out the Russians, but it’s reasonable for Russia to protect itself.* My question here is technical. Thousands of people, especially at the IETF, have worked to build the Internet we have today. The principles are simple; the implementation is demanding. So I’m asking engineers, “What technical systems must be built to ensure robust interconnection, assuming everyone wants to work in good faith?”

     ICANN CEO Fadi Chehadi confirmed to me there was no technical reason the Chinese, Hebrews, Verizon or any other competent party couldn’t set up independently. Vladimir Putin intends to test that, creating a new root that is not controlled by a California organization. The primary benefits of “The Internet” could be maintained so long as there was “robust interconnection.” Fadi added the rub was how to ensure that robust interconnection. 

    I think Fadi was worried about censorship, a real issue.

Continue reading

Massive MIMO explained in depth

Eom Marzetta invented Massive MIMO in 2010 and the first production units shipped in 2017 in Japan & China. As the build of 5G mid-band picked up in 2019 and 2020, M MIMO became the standard choice. As I write in late 2022, China has deployed over two million cells.

MIMO uses multiple antennas to send several signals in the same spectrum. With 4 antennas, typical speeds double. Results vary based mostly on the terrain.

M-MIMO was the next step. The typical M-MIMO uses 32-64 relatively small antennas. Lower frequencies (below ~2300 MHz) require larger antennas so it is generally impractical to have more that 4 or 8 antennas so rarely implement M-MIMO.

Continue reading

5G Phones Have Taken Over

5G phones are now so cheap that all but the poor are buying them. They still do nothing important but for $0 (Apple) to ~$40 more (Most others, falling) most people don’t want to buy a phone soon to be obsolete.

80-90% of the phones sold in China & Korea are 5G. >70% in the US. Similar figures for Japan, Australia, Canada, and Western Europe, although I don’t have data.

Continue reading

Soon, Almost Every Mobile Phone Will “Have” to Pay Huawei Royalties

China’s largest tech giant almost doubled its R&D budget over the past half-decade to $22.1 billion in 2021 — more than any company in the world outside America. Bloomberg

2021 R& D (billions) from Bloomberg

Huawei is the driving force behind Sisvel’s Wi-Fi 6 patent pool and will collect a significant % of the 50-60 cents Sisvel is demanding from every phone, Wi-Fi router, etc. That could easily amount to over $100 mi;llion per year. It will be hard to circumvent Sisvel, which also includes Phillips and Mediatek.

Continue reading

Editorial: New York Should Build Robert Sokota & Nick Colvin’s World’s Best Wireless Every Block Network

Put a small cell on every block in New York and you’ve built the best wireless network in the world. 25 years ago, Paul Baran was ahead of his time putting Ricochet modems on every block of my New York neighborhood, some on light poles.

The time is now ripe. Robert Sokota. Nick Colvin, CityBridge, and ZenFi have now begun putting 32-foot kiosks around the city. They have a city franchise, a demo unit in Queens, and $175 million to go to work.

Continue reading

250M 5G Subscribers End 2020 150M Q3

230-250 million 5G yearend as phone prices fall to US$199-260.

5G subscriber numbers are exploding. 100-120 million 5G phones will sell in Q4, many of them Apples. Samsung is also selling 5G aggressively in Europe and the U.S. China will add at least 50 million. 70% of the phones selling in China are 5G and likely over 90% by mid-2021.

I’m reworking my 2021 & 2022 estimates, which were already among the highest in the world. Nearly half the phones sold in the US, UK, and Australia are iPhones, most of which will be 5G from now on.

As Chinese phone prices (US$199-260) reach the West, most people will choose to buy 5G. Apple is selling 5G at the same price as 4G. Other companies will bring down the premium and it will make sense for all but the poor to choose 5G.

See 5G Phones $199-260 including screenshots of phones on sale at

“5G still doesn’t have any use cases,” writes top Wall Street analyst Craig Moffett. 5G is of almost no practical use today but most people keep phones for about 3 years.

2020 Q2Q3Q4 est
Rest of world448
Source: Analysis Branch telecom

Analysis Branch and our reporting date back to 1999, when Dave Burstein started covering broadband from its very beginnings. Since then, we’ve had the chance to learn from hundreds of the best industry and academic leaders. Jennie Bourne and Dave have written two books, been quoted by the WSJ, NY Times, & Washington Post, spoken at Columbia University, chaired 8 conferences, and traveled the world looking for news.

Contact Dave Burstein 347-603-6442 (New York) Deadlines understood. Ask for background, short quotes, sources or whatever you need.

5G Need to Know: Sales exploding, performance poor, 0 new apps

147 million subs September 2020, 117 million in China. Yearend likely 230-250 million, including 30-50 million iPhone 12. Decent 5G phones cost US$200-250 in China, with 6.5″ screens and multiple cameras. As China prices move West, 2021 will blow out.

Forget gigabit speeds and 1 ms latency. Mid-band speeds mostly 100-400 Mbps. Low-band typically 50-100 Mbps, the same or less than 4G 50-150 Mbps. Low-band is really 4G + 5G NR software, which adds little. The Emperor Has No Clothes.

Verizon specifies latency of 25-40 ms, including mobile Edge. PCMAG had some tests at 10-15 ms on Verizon, which is what to expect with an Edge Network. Edge will deploy slowly outside of Asia. 4G latency had been 35-60 ms, but it is falling as well.

All major apps run about as well in 4G as in 5G. Telemedicine and IoT are fine in 4G. Tesla and Ford point out autonomous vehicles don’t need 5G or any connection. What would they do out of signal.

One major app coming: surveilling and killing people. US DoD awarded $600M in 5G contracts. Telcos tell me police and surveillance outfits are very interested.

562,656 Koreans go back to 4G

Realme price US$150 for 11/11 sale

5G has so few practical uses that half a million Koreans have gone back to 4G. The saving is modest; otherwise, millions would have gone back. 4G speeds averaged over 50 Mbps in 2019. 4G is probably much faster today.

5G averages 300 Mbps across about a third of the country. So what? What can you think of that runs much better at 300 Mbps than at 50 Mbps? The latency is about the same. YouTube, Facebook, 4K video, and almost everything else is exactly the same on 50 Mbps 4G or 300 Mbps 5G.

Korea expected 15 million 5G users by the end of 2020 but will only reach 10 or 11 million. 800,000 chose 5G in August 2019 as the carriers deeply discounted the phones. The carriers somehow agreed to stop the discounting and sales plummeted.

China’s 5G looks likely to be 170-180 million by yearend. That’s more than the 150 million Chinese target and the typical Western estimates of ~100 million. People refused to believe my Dec 2019 estimate that 2020 would see 210 million worldwide and 150 million in China. It will probably finish higher.

That’s because the price difference between 4G and 5G is collapsing. If the extra cost is modest, it makes sense to go for 5G even if it does nothing useful now. Perhaps that will change in 2 or 3 years.

The difference in the bill of materials cost between 4G & 5G is US$15-35. (mmWave is more.) With eleven phonemakers fighting for market, I expected prices to plummet. They have in China, where more than 60% of the phones sold are 5G and the rate goes up every month.

For the 11/11 big sale, Realme is advertising 5G at US$150. Several other have 6.5″ 5G phones with multiple cameras and other features for US$200-250. The service costs the same. All but the poorest will be buying 5G.

Apple has now brought iPhone 12 5G premium to $0. All models have mmWave wave 5G, the most expensive kind. It costs less than $2 to airfreight a phone from China to Europe and a grey market is already emerging. When the Chinese prices come West, 5G will quickly dominate. My 2021 and later 5G estimates need to be raised.

Continue reading

844 low latency Starlink satellites: Musk changes rural possibilites

Starlink latency averages 30 ms, similar to 5G’s reality. Speeds measure up to 100 Mbps, although some only get half that. Upstream is usually 10-40 Mbps. Caps haven’t been specified but will almost certainly be higher than current satellite offerings.

By most standards, the performance is decent. A Redditer posts

Starlink is 600x better than my current ISP BEFORE you consider data cap. My jaw dropped when I saw the official numbers. I live in a rural village in Alaska and pay around $200/mo for service that is running fast if it hits 500kbps with a 40GB data cap.
Half the price for up to 300x faster service? Elon please start launching some polar orbits.

Heavy video watchers may need to add satellite broadcast as well, depending on the bandwidth cap. It is definitely a reasonable choice where the cost of fiber or decent wireless is prohibitive. That’s about one million locations in the US.

Beta testers are being charged $500 for the terminal and $99/month. That’s high: Most wireless or wired Internet in the US costs $50-80, including equipment.

Continue reading

Need to know: Real 5G speeds and spectrum (3 minute read)

Takeaway: 95+% of 5G is lowband (< 1800 MHz or mid-band (2500-4200 MHz.)
Asia, TMobile US, and much of Europe is mid-band.80-200 MHz of spectrum usually delivers 100-400 Mbps.
Low-band (AT&T, much of DT & other European, runs at 4G speeds and sometimes lower. See AT&T data below.
Millimeter wave often does 500 Mbps and a gigabit, but will be rare for years.

Gigabit and 10 gigabit 5G speeds are mostly a fantasy outside the lab. That will be true almost everywhere until 2022-2023, when large-scale millimeter wave deployments begin. Outside of Asia and possibly Verizon, gigabit speeds will be uncommon until late in the decade.


Spectrum and the number of cells/antennas mostly determine speed. Every engineer knows 5G is little faster than 4G. The 80-200 MHz of new mid-band spectrum is where the speeds come from, whether 4G or 5G.

Low: 50-125 Mbps Mid: 100-400 mmWave 500-1500

Need to Know: 5G Spectrum – Think mid-band

Takeway: Decent 5G until around 2025 will be mostly mid-band, 2500-4200 MHz. Frequencies lower than 2100 MHz essentially deliver 4G capacity. Millimeter wave, 20 GHz and above, can deliver gigabits but will be rare until the middle of the decade or later. It’s expensive, and the carriers don’t see much market.

Heretical but true: 5G delivers little more capacity than 4G, in the same spectrum and using the same antennas. T-Mobile US CTO Neville Ray estimates that in ideal circumstances, 5G delivers between 19% and 52% more capacity. “Ideal circumstances” are years away; below 2100 MHz, 5G is slower than decent 4G. 5G in lower frequencies adds nothing to capacity and has few if any practical advantages. It’s great or.

Dwight Silverman of the Houston Chronicle tested AT&T 5G around the city. Most speeds were in the 75-125 Mbps range typical of AT&T’s latest 4G. In the lower right, the single mmWave radio tested over 800 Mbps.

Mid-band, from 2500 MHz to 4200 MHz, is the primary 5G band in use. 5G can efficiently use 100 MHz bands, which usually deliver speeds of 100-450 Mbps and occasionally more. Like 4G, theoretical and lab speeds can go over a gigabit, but that’s misleading.

100 MHz contiguous is the standard 5G band, although not all carriers can achieve that. The lower part of the band has longer reach. The 2500 MHz spectrum of T-Mobile US and China Mobile is golden. (Both actually have 160 MHz, beyond the 5G specifications. Huawei tells me its radios can use all 160 MHz.)

Millimeter wave, from 20 GHz to over 60 GHz, can work with 400 MHz bands and deliver gigabits to many. Initially, it was the only 5G. It is in very limited use until 2022 and later. No carrier needs the capacity before 2022 and many are fine just using mid-band until late in the decade. Because shorter wavelengths have less reach, they cost more to deploy.

The industry, at 3GPP, decided to call anything that ran “5G NR” software “5G.” I don’t think something slower than decent 4G should be called 5G, but that’s become the norm.

Open RAN Experts: Not quite ready for prime time

Open RAN works, but the strongest supporters are clear it has a way to go. CTO Tareq Amin of Rakuten, with the most advanced deployment in the world, had to spend “hundreds of millions” on custom chips to get the performance he needs. My unofficial numbers are custom chip cost wiped out all the initial savings. Rakuten hasn’t turned on 5G yet.

Please don’t misread me as an opponent of Open RAN. It is probably the right choice for a new network today and will play a growing role in the future. But there are problems that must be solved.

Santiago Tenorio, Vodafone Group’s Head of Network Strategy & Architecture and Chairman of Telecom Infra Project (TIP), is one of the most enthusiastic supporters and has four field trials underway. Tenorio says

The big suppliers currently have the TCO upper hand. The traditional vendors have decades of experience, have thousands of employees, and are able to decide what would work better for operators. If you’re going to deploy [OpenRAN] in say, 25 sites, you may get better commercial conditions from incumbents.

There are numerous industry challenges if operators are to reap the promised open RAN benefits of a “richer ecosystem” and lower total cost of ownership (TCO). … We haven’t even scratched the surface of system integration challenges. There are one million different ways in which you can actually build a product that satisfies O-RAN specifications.

Continue reading

Analysis Branch: 100 million 5G subscribers in August, 84M+ in Q2, prove 5G is real.

210 million 5G yearend as phone prices fall to US$199-260.

New York August 31, 2020
Contact: Dave Burstein, 347-603-6442

  • 5G use is exploding. At the end of June, China had ~67 million 5G phones, Korea 7.35 million, the U.S. 4-5 million, and the rest of the world perhaps 4 million. 5G fixed wireless added perhaps 4 million more.
  • China’s 88M the end of July, combined with my reporting below, implies 5G reached 100M users between July 20 and August 5.
  • Saudi Arabia average 5G download 414 Mbps, 34% coverage leads the world. US, Europe mostly dismal. Open Signal
  • As Chinese phone prices (US$199-260) reach the West and Apple releases the 5G iPhone, monthly growth will reach more than 20 million per month. See 5G Phones $199-260 including screenshots of phones on sale at
  • December 2020 will almost certainly reach 200 million. 210-220 million is more likely.
  • “5G still doesn’t have any use cases,” writes top Wall Street analyst Craig Moffett.

This release is different We show our work so you can judge for yourself the accuracy. This is the full report, with details. See a discussion of sources and accuracy is at the end.

Q2 2020 5G: 84 million subscribers

China ~65 million, based on the number of phones sold reported by CAICT. Perhaps 4 million routers and fixed systems.
Korea 7.35 million, reported by the companies in financial reports
U.S.A. 4-5 million. That figure is based on phone sales reported by M-Science & Strategy Analytics.

Contact Dave Burstein 347-603-6442 (New York) Deadlines understood. Ask for background, short quotes, sources, or whatever you need. Broadband and telecom since 1999

100 million were connected by late July

China added about 14 million more in July. Korea probably 500,000. Japan is picking up, with NTT DOCOMO at 90,000 in July. The big gains in the U.S. will start in September (T-Mobile, now advertising heavily.) When the iPhone 5G ships in October or November, Verizon and the Europeans expect strong demand.

Yearend 2020: 210 million 5G users

The 210 million estimate and 65 pages more of analysis is at I’ll have an updated version shortly and a followup release.

By the end of 2020, we expect 210 million 5G subscribers. With 30 million phones sold in China in June and July. China is on track to easily meet the 150 million plan for the year. The U.S. will also accelerate. T-Mobile is upgrading about 3,000 towers per month to 100-400 Mbps and has begun aggressive sales.

Decent 5G phones are shipping in Europe for 400€. Huawei, Xiaomi, Realme, Oppo, and Vivo are offering decent, 6.5″ 5G phones in China for US$199 to $260. See 5G Phones $199-260 and Coolpad $199 5G phone with Unisoc Ziguang Zhanrui Chinese chip

When those prices reach the West, many Android buyers will choose 5G because the price difference is modest. When the 5G iPhone ships in volume, the Europeans and Verizon expect very high sales. (Possibly in October.)

5G in 2020 is mostly 100-400 Mbps, not the promised gigabits

The low prices will drive 5G sales in 2020, not any new 5G applications. There aren’t any that inspire people.

The performance hype is ridiculous. Open Signal reports latency is similar to 4G. Verizon claims 30 ms. 1 ms latency is a fantasy outside the lab. Low-band speeds are often slower than 4G, especially at lower frequencies. See Finally, Data: US 5G slower than Canada’s 4G. Believe it

Analysis Branch figures are higher than almost all subscriber estimates in the West

In December 2019, I put out a 210 million estimate for the end of 2020. Xiaomi dropped the 5G price to US$285, a demand driver. China officially set a plan for 150 million, which was forcefully echoed by the three telcos. Although most estimates for 2020 were ~ 1oo million, my research suggested the Chinese would deliver.

Chinese telcos, among the largest in the world, have consistently made their numbers for the last decade. 300 million were connected to fiber to the home in about four years. Minister Miao Wei last spring said, “Accelerate!” The carriers have delivered.

Company leaders no longer go to jail for missing quotas, but MIIT can and often has fired any executives who come up short. The $100 billion (sales) $14 billion (profits) China Mobile could increase marketing and phone subsidies enough to reach 150 million. It’s on track already.

5G is selling far above almost all predictions except mine because the phone price in China is little more than 4G. Decent phones go for US$199-260 in China, with prices falling there and everywhere else. Tens of millions of people have decided to pay the small premium for a phone that won’t be obsolete as soon. I would.

About the data

I have only indirect data on most of the world. If you want to be accurate, please think of the range of 83 million to 92 million rather than the 84 million headline figure.

I’m including a figure of 4 million 5G routers and pucks. Unfortunately, I can find no primary data. The 4 million is a guess. I have not tried to divide them by country. Data extremely welcome.

An analyst firm put out a 63 million figure for Q1, almost certainly a mistake but frequently repeated. The highest plausible estimate of 5G phone sales in 2019 and the first quarter of 2020 is 45 million and it is probably a lower than that. (Strategy Analytics reports 24 million for Q1 2020)

I’ve urged them to put out a correction and am not naming them here.

“I make many mistakes,” the Butler said. I’m sure I have some, although I’ve done a great deal of research, I’ll issue a correction ASAP.

Country details:

China: ~65 million.
China’s telcos are reporting ~115 million “5G contracts” but an authoritative government source (CAICT) reports only 64 million 5G phones shipped. Since 4G and 5G contracts are the same price, I assume the telcos are persuading many 4G customers to sign up for a “5G contract.” China doesn’t need to overstate the numbers; even the lower figure is three times as many as the rest of the world.
Over 410,000 base stations have been upgraded and 15,000 more are being done each week. China Mobile expects a total of 600,000 5G cells yearend, covering about 700,000,000 people. All is mid-band.
17 million 5G phones shipped in June, many selling for US$230-260. 30-gigabyte service costs $13-18/month. China is on track to easily meet the 150 million year-end target. Counterpoint reports 60% of 5G phones in June were Huawei, which has shipped over 20 million 5G phones in China and probably over 30 million worldwide.
Unofficial sources claim July is far ahead of June.

Korea: ~7.35 million
All are mid-band, mostly 100-400 Mbps down. Open Signal data implies the indoor coverage is terrible. See 5G #fail. 85% no 5G in “90% covered” Korea

U.S.A. 4-5million.
US 5G coverage is awful, so I was surprised when Strategy Analytics reported Samsung sold over 3 million expensive 5G phones in Q1. Most probably were sold by Verizon, despite Verizon customers only connecting to 5G 0.4% of the time. I infer that high-end Samsung buyers are spending more for a phone that will not be obsolete in a year or 2. CEO Hans Vestberg has said people are buying 5G phones even where Verizon does not have 5G coverage.
So far, almost all AT&T & T-Mobile has been the ridiculous “low-band 5G,” actually slower than much 4G. See Finally, Data: US 5G slower than Canada’s 4G. Believe it The companies are making it nearly impossible to separate the 5G at 4G speeds” from other 5G. I will exclude them if I can. Any reporter or analyst who doesn’t try to make the distinction should point out that much “5G” is slower than much “4G.”

Europe ?2 million
No European carrier has enough 5G customers to release a figure. I infer from that and the limited availability that there are few actual subscribers. More data welcome.

Gulf ? 1 million
The UAE, Qatar, and Saudi Arabia have some of the most extensive deployments of 5G. There is little or no public data on the number of subscribers. Ooredoo Qatar reports 200,000 subscriptions Q2.

Japan ? 0.3 million
3 carriers are deploying. NTT DOCOMO reports 150,000 subscription Q2 and 90,000 more in July. It is shooting for 2.5 million early next year.
Rakuten, the first telco in the world to build a completely virtual system, is not yet ready to turn on 5G. When it does, expect major changes. It has already covered a quarter of the population and expects to reach 70% early next year. See Rakuten virtualized 4G now covers quarter of Japan. It is half as expensive as NTT and will be a fierce competitor. 2021 totals for Japan look to be 10-12 million.

South and Southeast Asia ?0.3 million
Viettel and almost all the Southeast Asian countries are just starting to deploy. Jio in India is ready to move rapidly when the government approves. Look for very rapid growth in India in 2022 and possibly earlier. The projections of 18 million in 2024 are far too low.

Australia ?0.2 million
Lots of pr, little data

Latin America ?0.1 million
Almost all talk so far.

Africa ? 0.1 million
MTN in South Africa has recently deployed mid-band, but few subscribers so far.

Canada ? 0.1 million
Just getting started

Russia, most of Latin America, and almost all of Africa have little more than pr.

Total: About 80 million phones Q2 and perhaps 4 million fixed home systems.

Sources and accuracy

China’s government CAICT is the source for the phone sales figure. It reports phones shipped so I have to adjust for units in transit and dealer inventory. The Chinese carriers are reporting about 115 million “5G contracts” but only about 70 million 5G phones have shipped in China.

Korea’s 3 telcos provided figures in their quarterly financials.

The U.S. estimate is based on Strategy Analytics estimate of 5G phone sales plus a small number from 2019. M Science reports about 1 million fewer sales. The companies say nothing.

Few other companies have reported subscriber numbers, from which I infer they have very few. The 4 million figure for Rest of World is highly uncertain.

I have found no figures for the number of homes connected by 5G routers. My 4 million estimate is also highly uncertain.

The most widely reported figures — not ours — for Q1 almost certainly are 40-50% too high.

For estimates of year-end 2020 and through 2025,

Analysis Branch and our reporting date back to 1999, when Dave Burstein started covering broadband from its very beginnings. Since then, we’ve had the chance to learn from hundreds of the best industry and academic leaders. Jennie Bourne and Dave have written two books, been quoted by the WSJ, NY Times, & Washington Post, spoken at Columbia University, chaired 8 conferences, and traveled the world looking for news.

Contact Dave Burstein 347-603-6442 (New York) Deadlines understood. Ask for background, short quotes, sources or whatever you need.

$Hundreds of millions in extra chip costs prove Open-RAN not quite ready

Tareq Amin’s Rakuten is the most advanced Open-RAN and virtual carrier on earth. But he wasn’t able to get there using standard parts and equipment.

Rakuten pays hundreds of millions of dollars in non-recurring engineering fees to chipmakers like Qualcomm to obtain the components it needs. …We cannot find the right material at the right cost, the right architecture, to address the future requirements for 5G radios

Tareq Amin Rakuten to John Hendel of Politico

That’s an enormous sum, enough to design state-of-the-art chips. I would guess the extra cost ate up all the expected cost savings from the initial deployment. (Below) Important: Please don’t infer from this that Open-RAN is a failure, especially for new networks. Most of these problems will be solved. Meanwhile, don’t believe the hype.

Alex Choi of Deutsche Telekom is enthusiastic about Open RAN and the other buzzwords as near-future technologies, still with challenges. That’s the near-universal consensus of the top network engineers. That doesn’t mean the new networks shouldn’t pioneer, but I’ve been seeing some unfortunate datapoints:

  • Vodafone CTO Scott Petty has been one of the most enthusiastic supporters and is doing some deployments. But Laurie Clarke of New Scientist quotes Petty, “We believe by 2023, we may be able to deploy some scale in the rural parts of our network, but it will take until 2025 to be able to deploy at real scale in our denser urban and suburban areas.”
  • AT&T was the first strong backer of SDN/NFV/Ecomp, pouring a fortune into open software. It is quietly cutting back. A friend has been receiving resumes from senior people at AT&T, expecting layoffs.

Contrary to general belief, the initial saving from open and virtual RAN is modest. The hardware is cheaper, but hardware is only a small part of the network cost. For large carriers with bargaining power and good negotiators, the price of the hardware is already low.

China Mobile & China Telecom are paying US$23,000 per cell. The antenna, power supply, and a high-performing processing unit will still be required. How much could they really save if they bought radios from Mavenir or Jio/Radisys?

Smaller telcos, like AT&T or Telefonica, order closer to 10,000 cells. Nokia & Ericsson are masters at extracting maximum revenue from customers but if pressed will give great prices on the initial purchases.

Over time, the more flexible systems should be much cheaper to upgrade. Moore’s Law may be slowing, but the systems in five years will be far more capable than today. If open systems deliver on their promises, the upgrade should be cheap and easy. (That’s still unproven.)

Carl Russo at Calix has produced some of the best SDN systems, Verizon tells me. He tells me carriers who just look for initial savings will not reap the full benefit of SDN. Only those who use the flexibility and integrate better management, especially of customer offerings, receive a full return.

Nearly all telcos upgrading and densifying existing systems will likely follow a similar timeline as Vodafone, limiting new systems for several years. New builds are almost all choosing Open.

All of this is speculation until we have good data from the field, of course.

Primary: Cost per bit of wireless is falling at a ferocious rate

Even experts are skeptical that Verizon’s cost per bit is falling at 40-50% per year, but I’ve confirmed it to my satisfaction. Three well-established datapoints:

  • Traffic was growing at 60-80% per year and only now is growth falling below 40%
  • Capex has been flat to down
  • Speeds have been going up consistently and congestion is uncommon

If wireless cost had not been plummeting at a rate similar to traffic growth, without a rise in capex the network would seriously degrade. Ergo: Deployed technology is improving very rapidly.

Verizon CEO Vestberg calculates his cost per bit has fallen 90%. His peer at Deutsche Telecom estimates 70%. Neither figure is precise since neither offers details, but the trend is clear.

The trend has been clear at least since 2013, when carrier aggregation and MIMO became widely deployed. It will continue at least until ~2023, based on the equipment (Massive MIMO,) mid-band spectrum. and the likely maturity of SDN, NFV, O-RAN and the like.

At the same time, routers, switches, optical backhaul and other network components have been going down at a similar, extraordinary rate. Moore’s Law is slowing as chip features aproach atomic dimensions, but More than Moore techniques keep improving productivity.

All of which is wonderful, especially if carriers can find a way to sell what they can deliver.

Do also read Moffett’s Remarkable Insight: Low marginal cost means nothing if you can’t sell it for insight on the implications of technology advance.

Low cost can be an extraordinary weapon

Low costs, passed on to customers, can re-create the market. The biggest telecom story of the last decade is the 400 million Indians newly with 4G. Mukash Ambani spent US$30 billion for a state-of-the-art, extraordinarily efficient network covering 95% of 1.353,000,000 Indians.

With prices from $2-$10, he rapidly won customers from the 10 other Indian telcos. Most were selling 2G and 3G, saying 4G was too expensive for India. This was hogwash; 4G, the more recent technology, is an order of magnitude less expensive. Ambani makes money at prices so low all but two other companies are gone. Survivor Bharti wants a government handout and Vodafone is on the edge of bankruptcy.

Once the network was built, Ambani’s cost per bit or additional user was very low. He took that to the extreme, literally giving away three months or more of service to new customers, without catches. Ten million signed up each month and soon became paying customers. Jio is now at scale, profitable, and soon the most comprehensive “platform” on earth.

Ambani is extremely proud he is creating the first great Indian Internet company.

Low costs are not a guarantee of success

The economics textbooks suggest being the low-cost provider is a crucial strategy. This isn’t necessarily true in telecom, where fixed costs are high and marginal cost only a modest factor. See Moffett’s Remarkable Insight: Low marginal cost means nothing if you can’t sell it

Rakuten in Japan, Dish in the US, 1&1 Drillisch in Germany, and possibly Oi in Brazil are hoping to replicate the success of Jio. The other companies in India were just getting started on 4G. Most were selling 2G & 3G, and most consumers thought 4G phones too expensive. But 4G smartphones had fallen to US$50. Ambani asked for bids on tens of millions of 4G feature phones. That brought the phone price down to ~$25. It makes sense to find ways to get all customers on 4G, because they will cost much less to service.

In Germany, Japan, the U.S. and Brazil, the empires can strike back powerfully

The new 4th carriers will have a more modern network than even giants like Deutsche Telekom and NTT DOCOMO. But the incumbents are in far better shape than the 2G and 3G networks Jio wiped out. Most have 90-98% 4G coverage and are starting to go beyond pr in 5G.

Even in Brazil, customers of TIM get a 4G signal 88% of the time. Claro and Vivo are not that far behind. Vivo has passed 13 million homes with fiber, which also provides robust wireless backhaul. If Marc Ganzi takes over Oi, Brazil’s $4, he will face strong competitors.

If the newcomer is successful, the incumbents can and will strike back. Verizon and AT&T typically make $20 billion a year. NTT DOCOMO, KDDI,DT, Telefonica; and American Movil/Telemex make billions. Their capabilities are immense.

If Verizon raised capex by 3% of its $125 billion in sales for 3 years, that would add $10 billion, enough to rebuild most of the network to latest and greatest. Actually, they’ve been doing that already. Verizon’s “One Fiber” program cut network costs by 50% in the first year alone, Lee Hicks told me.

Even with advanced technology, being the fourth carrier is tough. The main opportunity is when the first three carriers have cartel-like pricing, which is common.

Virtualization savings are not enough

Rakuten’s Tareq Amin believes his new Japanese network is costing 30% less. It’s cloud-native, software defined, fully virtual, and uses a software RAN. It’s got a way to go, but let’s start with his 30% figure How much difference does that make?

Let’s assume the marginal cost of serving a new customer is about 15% of the revenue, once customer acquisition costs are covered. A saving of 30% would lower the cost from 15% of revenue to 10% of revenue. That’s about 5%. It’s meaningful but usually dwarfed by marketing and customer acquisition costs.

The engineers actually building networks , like Alex Choi, are enthusiastic about the potetial but are reporting numerous challanges still to solve.

The big productivity improvements came from MIMO, Carrier Aggregation, Massive MIMO, rebuilding the network inside the carrier, and more spectrum

The huge productivity improvements began around 2013 in 4G and will continue driving productivity at a frantic pace at least until 2022-2023.

There’s a myth that 5G NR yields major capacity improvements. The highest plausible estimates come from T-Mobile CTO Neville Ray, He swore under oath at the FCC that 5G NR could result in 19% to 52% improvement over what 4G can deliver today using the same spectrum. 4G LTE is so close to Shannon Law limits it’s impossible for any software to massively increase capacity.

Anyone claiming otherwise is ignoring the progress made in 4G. Between 2013 and 2020 of the typical 4G network went up 10-25 times while capex was flat to down. If demand was greater, most networks could have done twice as much while still keeping capex reasonable.


Paulraj’s invention produced about a 2-4x improvement starting around 2012 by using 4-8 antennas. The big deployments were from ~2017. Neither DT nor AT&T has deployed MIMO everywhere and room for improvement remains. The results vary by terrain.
Typically, a 2-4X improvement.

Carrier Aggregation

LTE was designed to use 20 MHz carriers. By around 2013, LTE Advanced allowed joining carriers, initially 2 or 3 but now up to 6. A company like AT&T might combine spectrum at 700, 1799, 1900, and 2300 MHz. Five bands at 3.5 GHz would be possible in 4G, although most of 3.5 GHz from telcos will be 5G.
AT&T is actively combining Wi-Fi/unlicensed spectrum at 4.9 GHz, sometimes doubling performance. This works in 4G but not yet in 5G, one reason 4G is often faster than the 5G deploying today.
Typically, a 3-5X improvement is possible.

Massive MIMO

Tom Marzetta invented M-MIMO in 2010 at Bell Labs, but Huawei and ZTE were the first to deploy in 2016. The results from Softbank Japan and China Mobile in 2016 were improvements of 3X-10X in TD spectrum, widely varying by terrain. Nokia and Ericsson couldn’t ship M-MIMO in quantity until 2019, so some people think it is 5G. Actually, 5G offers only very small improvements over 4G.
Massive MIMO is key to opening up 900 MHz of spectrum, almost doubling the available spectrum. Without Massive MIMO, the throughput for mobile data in mid-band was not enough to cover the cost. Now, that spectrum is selling for billions.
M-MIMO works poorly in FD bands, below 2300 MHz. Some hope for a technological breakthrough.
Typically, M-MIMO results in a 3-5X improvement in TD spectrum.

Other technology improvements (Dozens)

Coding more bits per Hz (256 QAM) raised capacity by almost a third starting in 2016. It’s still not rolled out everywhere. SON – self-organizing networks – reduce interference between cells. Ibrahim Gedeon of Telus told me in 2018 that made an enormous difference.

More spectrum

Opening the band from 3300 MHz to 4200 MHz is a big advance, doubling the available spectrum. Even more important was the effect of carrier aggregation on the use of bands telcos already owned. Carriers like AT&T have been warehousing spectrum for a decade, mostly to prevent competitors from buying it. (Ask any expert on spectrum auctions.)
As recently as 2018, AT&T’s CFO told Wall Street half of its spectrum was fallow. (No one believed that but I confirmed similar with several carriers.) In fact, AT&T was actively soliciting buyers for its spectrum at 2300 MHz because without carrier aggregation it was impractical to use. As 3-5 carrier aggregation became practical, and phones began to support it, the effective spectrum at many carriers increased 50% or more.

In 2014, world-class engineers debated whether wireless would soon see a 10X or 100X capacity improvement

In a remarkable symposium accompanying Paulraj’s Marconi Award, Paul, John Cioffi, and others extrapolated wireless progress. Vint Cerf, Tom Marzetta, and a dozen other luminaries took part. For the last six years, I’ve watched the vendors deliver much of what the pioneering engineers predicted.

At the dinner for the award, Paulrj quietly told me, “I’m confident about 50X, but not so sure about 100X.” I bet if he looked forward again, he’d expect the 100X this decade.

Moffett’s Remarkable Insight: Low marginal cost means nothing if you can’t sell it

The low cost per bit of 5G means almost nothing if you can’t sell the bandwidth.  I’ve been warning for at least five years that telcos’ biggest problem is overcapacity. In 95% of locations, most of the time, major telcos have excess capacity. The tech is moving incredibly fast, driving cost per bit down at 40%-50% per year.

Hans Vestberg at Verizon discovered this the hard way. He told an interviewer his 5G network would cost 90% less per bit. He would use that cost advantage to create profitable new products. When asked which products Verizon would offer, he said he didn’t want to tip off the competition.

A year and a half later, Verizon has no significant new products or revenue sources. None.

(The network also isn’t performing. Open Signal connected to 5G 4/10ths of 1% of the time.)

Craig Moffett, the best on Wall Street, warns of “The Fallacy of Marginal Cost Advantage.”

All of us are taught from our earliest studies of microeconomics that marginal cost advantage is… well, everything. The cost-advantaged widget maker will always win…

It’s not true with telecom.

Years ago, I was asked by one of the legacy long-distance networks whether the spate of next-generation fiber long-distance networks … would indeed have an “order of magnitude” advantage in marginal costs … We concluded the “next-gen” networks would indeed have marginal costs that were an order of magnitude lower than those of the legacy providers.

But that wasn’t the bad news. The real problem wasn’t the marginal cost disadvantage. It was excess capacity. The capacity of long haul fiber networks had exploded in 1997 and 1998, driven by technology advances.

We are regularly asked whether we buy the thesis that Dish Network can build a next-generation virtualized ORAN network that will have a significant marginal cost advantage,,,

Sure they can. The cost of a network is almost entirely fixed; there’s the cost of network construction, and there’s the ongoing cost of powering, infrastructure rental, repair, and maintenance. Sending more or fewer pulses of light doesn’t change any of those.

Any self-respecting network engineer will tell you that one of the first things they ever learned about network economics is that networks are built to handle the moment of peak demand. It therefore follows that the cost of a unit of capacity at any moment other than the moment of peak demand is zero. It is excess, it is not recoverable, and it is therefore valueless (a unit of capacity one second ago that expired unused isn’t helpful to anyone, and is lost forever).

In the end, there are only three things that matter: what does the network cost to build? What does it cost to operate? And how much revenue can it generate? What Dish is planning in the U.S., is a reminder that building and operating networks is hard. And earning a return on them is even harder.

The key question for Dish, Rakuten, 1+1 Drillisch, and other new networks is not whether their costs are low. It’s whether that can be turned into sales. Jio in India proved it was possible, reaching ~400 million subscribers in 4 years. It won’t be easy.

Early results on Open Ran, SDN, NFV, and the like don’t match the hype.

Alex Choi of Deutsche Telekom is enthusiastic about Open RAN and the other buzzwords as near-future technologies, still with challenges. That’s the near-universal consensus of the top network engineers. That doesn’t mean the new networks shouldn’t pioneer, but I’ve been seeing some unfortunate datapoints”

AT&T, Ciena,…”5G’s attractive, transformative services will likely introduce threat vectors not yet seen or experienced”

5G Americas has a new paper reflecting the almost universal opinion among security experts that 5G has important vulnerabilities that can’t be ignored. 5G systems are perhaps an order of magnitude more complicated than 4G and correspondingly hard to protect. They are also massively distributed, from your home IoT up. That huge attack surface is all but impossible to guard.

Verizon CEO Hans Vestberg, who should know better, said 5G is completely secure. When I asked how he knew it, I was pointed to statements from 3GPP. 3GPP has some of the best engineers in the world, but their promotion always needs factchecking. Great engineers are often bad liars.

The 5G paper contains good short descriptions of 5G software and hardware considerations. It’s somewhat less technical than most, but no one has yet written “5G for the intelligent layman.:

Proposed 5G architectures are designed to close security gaps from previous iterations of cellular networks, but the pervasive nature of 5G introduces new security challenges outside the traditional space. 5G’s attractive, transformative services will likely introduce threat vectors not yet seen or experienced. This paper will look at how 5G differs from other wireless architectures, and what threats, vulnerabilities and attacks are therefore possible. Security considerations will examine various aspects of software, virtualization, automation, and orchestration, as well as Radio Access Network (RAN) considerations. Zero-Trust security, as well as several other techniques, will be discussed to mitigate the threats, and various recommendations will be proposed for security.
5G will usher in an age of accelerated innovation, but with that promise comes the inevitable potential for
attacks. The telecommunications industry needs to be prepared to defend against these attacks and have
mitigation plans in place for current and future attack vectors.

The work was led by David Krauss. 5G Americas is sponsored by AT&T, Cable & Wireless, Ciena, Cisco, CommScope, Crown Castle, Ericsson, Intel, Mavenir, Nokia, Qualcomm, Samsung, Shaw Communications Inc., T-Mobile USA, Inc., Telefónica and WOM.

China is > 70% of 5G as subs pass 80M

At the end of June 2020, China had ~65 million 5G phones, Korea 7.5 million, the U.S. 5-6 million, and the rest of the world perhaps 4 million. In June 2019, Minister Miao Wei told the giant Chinese telcos “accelerate” and the results are extraordinary. China has upgraded 415,000 base stations and are adding 15,000 more each week. It will easily surpass the 2020 goal of 150 million.

Korea is second with 7.5 million subscribers but will soon be passed by the U.S. Korea added 800,000 5G users in August 2019 but the telcos cut back promotion and now add ~500,000 per month. The U.S. began the year shipping over 1 million phones per month, but closing stores slowed down the pace. With the 5G iPhone and accelerated marketing from T-Mobile, the U.S. will be at least 12 million by year and possibly as many as 17 million.

Sales in the rest of the world are so slow that the carriers refuse to release figures. 4 million is a plausible total but data is limited.

Decent phones sell for US$230-270 in China, as 11 makers are fighting for a market where all but 4 or 5 will struggle. The Chinese and Korean deployment is all mid-band, with speeds typically 100-400 Mbps down.

The much-hyped low latency is still in the labs and “1 millisecond” will be a fantasy for years. Verizon claims 30 ms latency, although some tests are lower. Open Signal reports “no significant improvement in 5G latency over 4G.”

Which means none of the promised use cases are gaining traction.

June 2020 H1 5G subscribers 86M (84M-92M)

With decent phones selling for US$230 and the price going down, China will almost certainly meet the yearend 150 million plan. Verizon and others expect a big boost from the iPhone 5G, so I’m confident in my 210 million year-end figure. That estimate and 65 pages more of analysis is at

5G performance is highly disappointing. While Korea claims over 90% coverage, Open Signal only connected to 5G 1 in 6 tries. 5G #fail. 85% no 5G in “90% covered” Korea I believe, but haven’t confirmed, that the problem is that mid-band struggles getting indoors. Low-band speeds are often slower than 4G, especially at lower frequencies. See Finally, Data: US 5G slower than Canada’s 4G. Believe it

Latency is about the same as decent 4G. Verizon claims 30 ms. 1 ms latency is a fantasy outside the lab. Essentially none of the highly touted “use cases” and “new apps” have appreciable takeups.

5G is selling far above almost all predictions except mine because the phone price in China is little more than 4G. Decent phones go for US$230-260 in China, with prices falling there and everywhere else. Tens of millions of people have decided to pay the small premium for a phone that won’t be obsolete as soon. I would.

Much more June 2020 H1 5G subscribers 86M (84M-92M)

For estimates of year-end 2020 and through 2025,