Update December 2021: Vestberg was dead wrong about a $200-400/cost per home passed for mmWave. The word from Verizon in 2017 was to expect 600 meter reach for mmWave. Testing by Earl Lum found Verizon built to a 200 meter radius. The economics of mmWave deteriorated because the capability was massively overestimated.
On the other hand, mid-band Massive MIMO is proving incredibly productive and surprisingly cheap. It produces more bandwidth than carriers can sell at most locations. Net result: worldwide capex roughly flat 2018-2024, serious over-capacity.
Hans Vestberg of Verizon is passing 30M homes without raising capex.
NTT DOCOMO's very respected CTO Seizo Inoe in 2016 called high costs of 5G "a myth" in presentations at the Brooklyn 5G Summit and an IEEE conference in Kuala Lampur. He pointed out that LTE was cheaper than 3G and that much of the 5G would use existing towers and backhaul.
Vestberg, former Ericsson CEO now running Verizon's network, sees mmWave costing $200-400/home passed as they deploy to a quarter of the U.S. (Obviously, some other areas will be more expensive.) He just told a CITI investor conference 5G is
"Massively, massively cheaper than having a fiber all the way to the home to have sort of a beam in the air going to the home."
Verizon has previously said their costs to pass a home with fiber were $700 in 2007 and went down from there, presumably to $400-600. AT&T has confirmed similar as they deploy 15M lines of fiber home. I'm inferring that "massively, massively cheaper" would be $200-$400.
AT&T has implied similar costs by saying their capex would not have gone up except for the tax cuts.
All cost numbers like these are rough estimates and vary enormously based on density and physical circumstances. In particular, the last 2-8% of rural fiber costs into the thousands and the last ~1% is prohibitively expensive.
The main reason the cost projections are coming down is that mmWave reach is proving much better than expected. Vestberg is often getting a gigabit 300-700 meters. A major European government study was based on 200 meter reach
The (still early) results from the field are much better. I believe the improvement is mostly due to advanced beamforming performance. 256 antennas at mmWave frequencies are the size of a large chip. Improved algorithms and more processing power take advantage of the ability to steer the beam were needed, improving capex. Near Line of Sight, with tree and building obstacles, is doing remarkably better than almost anyone except Ted Rappaport expected.
The high cost estimates for "5G" from outfits like Deutsche Telekom are pure b_____, designed to get government subsidies and concessions. In particular, much of the "5G" likely in Europe before 2022-2025 is 4G TD-LTE with a minor software tweak. Three years ago, when early work in 3.5 developed, it was all called 4G TD-LTE
None of the above is certain until we have more data from the field.
One of the boldest statements in Onoe’s speech was that deploying 5G will not require a ton of investment. ... Better service, he insists, does not always correlate with greater capital expenditures. NTT DOCOMO’s 600 billion yen in capital expenditures last year marked a 15-year low, even as the data traffic across its networks grew 6300 percent since 2000.
In fact, Onoe actually expects capital expenditures for NTT DOCOMO to drop throughout 5G deployment, which he says would keep with trends for earlier wireless generations. To illustrate his point, Onoe opened a chart of the company’s capital expenditures over the past 20 years and asked the audience to guess when the company rolled out 3G and 4G LTE service. It’s impossible to tell based on expenditures alone. “For LTE, there was no increase in CapEx before the LTE launch,” he says. “That's a fact”