If SpaceX Independently Builds a U.S. Terrestrial Mobile Network, the Base-Case Cost Is About $71.5B, with Low-Band Spectrum the Largest Variable

Overview

Using bottom-up models of wireless propagation, population distribution, capacity, spectrum, and unit costs, the report estimates the sites, investment, and time SpaceX would need to independently build a U.S. mobile network without relying on terrestrial carriers. The central conclusion is that an independent build is not impossible, but low-band spectrum, construction capacity, and uplink coverage would make it a lengthy and expensive undertaking; the base-case nationwide network plus Grain low-band spectrum scenario would require about $71.5B and 8.4 years.

Core views

Bernstein remains positive on SpaceX's overall opportunity, including launch, orbital data centers, and Starlink broadband for consumer, enterprise, and government customers, and assigns SpaceX an Outperform rating and a $248 target price; the table lists an August 28, 2026 closing price of $141.50. However, the report considers direct-to-device mobile service the most difficult of SpaceX's businesses. Mobile V2 satellites are expected to begin launching on Starship in mid-2027, with larger antennas, greater capacity, denser beamforming, and lower orbits potentially improving the downlink experience. Yet handsets have only limited transmission power, antenna gain, and battery capacity, while indoor penetration losses further weaken the uplink. Receiving notifications, loading webpages, and browsing social media outdoors may gradually become usable, but whether continuous two-way voice, video, and future AI connectivity use cases can operate reliably remains uncertain. Improvements in satellite capabilities therefore solve only part of the problem, and a credible mobile service would still require a terrestrial network; Bernstein ultimately continues to view a partnership model as more likely than a fully independent SpaceX build. The report constructs three network tiers—urban, nationwide, and premium—and, for each, assumes either obtaining or not obtaining Grain low-band spectrum, producing six scenarios. The urban network covers 70% of the population and is the lowest-investment but least competitive option. The nationwide network covers 92% of the population, resembles a Sprint-style network, and serves as the report's base case, supporting a network positioned for approximately 50M-75M subscribers. The premium network covers 95% of the population and aims to approach incumbent carriers, achieving network parity on several metrics. The six scenarios require 32,362-117,384 macro sites, network capital expenditure excluding spectrum of $14.3B-$78.0B, and total project costs including spectrum of $46.9B-$126.5B, approximately stated as $50B-$130B. The report views “nationwide network + Grain” as the most likely independent-build option: it covers 92% of the population and requires 56,705 macro sites, 25,262 conventional small cells, and 4M Starlink-backhauled residential femtocells, at a total cost of $71.5B, comprising $29.7B in network capital expenditure and $41.8B in spectrum. At the maximum deployment rate, it would take 8.4 years, with the model indicating completion in 2037. The nationwide scenario without Grain would require 85,858 macro sites, $79.5B, and 12.7 years. Coverage, rather than capacity, initially determines network scale. Incumbent carriers built their low-band coverage layers long ago, so routine planning focuses more on capacity; a new entrant, however, must first ensure that signals can reach populations, rural areas, and highways connecting cities. Rather than assuming a site count and multiplying it by unit cost, the report begins with where people actually live and calculates sites based on four population settlement types, propagation radii at candidate frequencies, and hexagonal coverage geometry. The incremental cost of covering approximately 70% of the population is similar, after which the marginal cost of covering additional people rises rapidly as the network enters lower-density areas: urban-core population density is approximately 940 people per square mile versus about 24 for rural coverage, a roughly 40-fold difference; more than two-thirds of the U.S. population is concentrated on less than 3% of its land area. The model is anchored to the 2,613 urban areas in the 2020 Census, where 265,149,027 people, or 80.0% of the U.S. population, live within 103,872 square miles, or 2.94% of U.S. land. For the portion above 80% population coverage, it uses probit interpolation with log density approximated by a normal distribution. Varying rural land area by 30% in either direction changes the nationwide scenario's site count by only 6.9%. Highway coverage raises costs further. Based on 2023 FHWA data, the model includes 29,109 miles of rural Interstate highways and 97,964 miles of rural expressways and principal arterials, covering 20%, 45%, or 70% of them depending on the scenario, while accounting for natural overlap with population coverage, terrain blockage, and signal spillover. In the nationwide plus Grain scenario, the highway layer requires an additional 11,413 sites, approximately 24% on top of the population-driven site grid. Highway coverage is a linear problem, so site requirements vary with radius rather than radius squared. Consequently, residential femtocells or satellite traffic offload cannot replace coverage sites on remote highways. On May 12, 2026, the FCC waived SpaceX's AWS-4, H-block, and AWS-3 terrestrial buildout obligations and replaced them with satellite D2D performance metrics, making terrestrial coverage levels a commercial choice rather than a regulatory minimum. Capacity is not the primary constraint in the network's early stages. The report separately calculates coverage and capacity grids for each type of area and uses the larger site count; in every scenario, propagation distance becomes binding before traffic. On the capacity side, it converts monthly user traffic into busy-hour demand using 7% of daily traffic, then estimates per-site capacity using spectrum bandwidth, spectral efficiency, three sectors, and network load factors of 55%-72%, taking the lower of downlink and uplink capacity. The nationwide scenarios without and with Grain both target 25.2M users. Before Upper C-band activation, their grids could support 77.3M and 54.5M users, respectively, equal to 24.6% and 17.3% of the covered population, with capacity headroom of 3.1x and 2.2x. While Grain reduces the site count, it also reduces capacity headroom by approximately 25%-30% in paired scenarios. If per-user traffic grows by about 13% annually, 2.1x headroom lasts only around six years, close to the six- to ten-year deployment period. Upper C-band therefore ultimately is not optional but a necessary capacity layer for supporting a mature subscriber base. The uplink will also become more important as the traffic mix changes. Large antenna arrays improve the downlink and enhance the base station's ability to receive handset signals, so the uplink becomes the capacity constraint only when the downlink-to-uplink traffic ratio falls to approximately 5:1. U.S. networks are currently at about 8:1 to 10:1, but video calls and cloud backups are driving the ratio lower. The report cautions that capacity output is the least robust part of the analysis: if spectral efficiency falls by one-third, or 2031 rather than current per-user traffic is used, the pre-Upper C-band penetration ceiling would decline from 16.5% to about 11% of the covered population. SpaceX proposes deploying self-installable, licensed-band residential femtocells next to approximately 12M installed Starlink terminals in the United States, with direct satellite backhaul. Because about 90% of mobile usage occurs indoors—and indoor traffic is precisely what satellite beams struggle to serve but residential femtocells can absorb—the model deploys 1.5M-8M residential femtocells depending on the scenario, offloading 4%-18% of total macro-network traffic, with D2D satellites offloading another 0.5%-3.0%. However, this offload occurs mainly in urban and suburban areas, relieving capacity and the less expensive portion of deployment, but it cannot add coverage on unpopulated highways or in rural areas. Under current technical assumptions, it therefore does not materially reduce the macro-site count or total capital expenditure. Spectrum is the largest determinant of cost and site count. SpaceX acquired 40MHz of AWS-4, 10MHz of H-block, and up to 15MHz of unpaired AWS-3 from EchoStar, totaling about 65MHz, for approximately $20B, and provided around $2B to fund interest on EchoStar's debt. Every scenario includes approximately $21.6B of committed spending. Only 20MHz of this portfolio is available for uplink versus 45MHz for downlink, creating a 2.25:1 asymmetry, while coverage is determined by the 1.7GHz mid-band uplink. SpaceX currently has no low-band spectrum, whereas incumbent carriers built their foundational coverage layers at 700MHz or 600MHz. Handset transmit power is about 0.2 watts, far below a base station's approximately 40 watts, so coverage distance is determined by the handset-to-base-station uplink. Grain's Band 26 holdings are the asset that most materially changes deployment costs in the report. The band is located at 817-824MHz and 862-869MHz, with approximately 10MHz available as a contiguous coverage carrier. It can lower the coverage layer's center frequency from 1.70GHz to 0.82GHz and increase the rural cell radius from 2.49 kilometers to 3.13 kilometers. In one coverage-engine comparison, the nationwide network site count declines from 88,563 to 58,423; in the six-scenario summary table, nationwide macro sites decline from 85,858 to 56,705, both indicating savings of roughly 30%. Grain spectrum has limited capacity, so other bands would still need to carry traffic, but the avoided site costs exceed the incremental license expense, lowering total costs in every paired scenario. Grain acquired the spectrum at an all-in cost of approximately $3.6B; Bernstein values it at $4.5B-$8.5B, or $0.97-$1.83/MHz-POP. SpaceX has not yet contracted for the asset, and AST SpaceMobile is also bidding. The FCC requires Grain to select a satellite partner by November 5, 2026 and submit an application by December 4; the transaction structure may also resemble a long-term lease rather than an outright transfer. Upper C-band solves only capacity, not coverage. Auction 115 will offer 160MHz of spectrum at 3.98-4.14GHz, with bidding beginning April 27, 2027. The report assumes SpaceX obtains 20-60MHz at a cost of $6.5B-$14.2B. The band cannot be activated until December 31, 2030 at the earliest in the top 75 markets and July 1, 2031 elsewhere, so 2027 auction spending could sit idle on the balance sheet for about four years. The report treats it solely as a capacity overlay on existing coverage sites. Overall, spectrum spending totals $29B-$58B, or 40%-70% of network deployment costs: SpaceX's committed purchases provide capacity but lack coverage, the Grain asset that would provide coverage remains uncertain, and the Upper C-band spectrum that would provide large-scale capacity requires an approximately four-year wait. The unit-cost model itemizes radios, antennas, baseband computing, cabling, mounting hardware, installation labor, power, backhaul, engineering, zoning approvals, and project management. The all-in cost of one colocated macro site is $258,000-$364,000, versus DISH's actual approximately $317,600. The model assumes 3%-5% of sites require new towers and project management costs equal 10%-17%. Open RAN receives no unit-cost discount because the cited evidence does not show an inherent material cost advantage. Site rent, land leases, backhaul, power, and maintenance increase by 2.5%-3.0% annually, with 15 years discounted at 9%. In the visible scenario table, annual operating expenditure after full buildout for urban and nationwide networks ranges from $2.3B to $5.7B; the 15-year present value of capital expenditure, spectrum, and operating expenditure totals $108.7B in the nationwide plus Grain scenario. Elsewhere, the text states a range for fully built network Opex of approximately $2B to approximately $78B. Actual progress is constrained by construction capacity rather than funding. Only about 1,200-2,000 genuinely new towers are built annually in the United States, and roughly one-quarter are not for wireless communications. The model limits new towers to 1,200-1,500 per year and additions of tenants to existing towers to 5,000-9,000 per year, versus DISH's historical peak of approximately 5,000. It therefore assumes the vast majority of deployments use existing structures. The nationwide plus Grain plan requires 8.4 years, while the nationwide plan without Grain requires 12.7 years; even at a more aggressive pace, the premium scenario without Grain requires 8.1 years. The model is validated against DISH's actual deployment: after inputting DISH's spectrum and coverage targets, it predicts 22,952 towers versus approximately 25,000 actual towers, an 8.2% underestimate. The base-case plan serves 5,386 people per tower versus approximately 4,087 for T-Mobile, making it a sparser but still incumbent-scale, coverage-first network. DISH serves approximately 10,700 people per tower and can retain only around 17% of its own users' traffic on-network. Sensitivity tests show that requiring an additional 3dB for deep-building signal penetration or raising cell-edge standards changes tower counts by 38.5% and the capital bill by 35%. In contrast, a 25% increase in all unit costs raises total project cost by only 6.4%, because spectrum represents about two-thirds of the base-case bill. This demonstrates that engineering coverage standards and low-band spectrum choices matter far more than the price per tower. The investment implications are conditional: if SpaceX actually pursues an independent terrestrial build, extensive colocation on existing towers and long-term lease spending would directly create monetizable demand for communications-tower companies and could add a competitor to the U.S. mobile market. However, the spending would take years to be fully realized, and Bernstein still sees a partnership as the more likely outcome. The report does not assess the full network's investment return, customer acquisition, handset subsidies, retail, working capital, or satellite-constellation costs, so the $71.5B base-case bill is not a complete mobile-business economic model.

Analysis framework

Bernstein first assesses whether satellite D2D alone can satisfy mobile communications coverage and two-way connectivity requirements, then defines urban, nationwide, and premium network-quality targets, each with and without Grain low-band spectrum. The report then uses population density, land area, highway mileage, and 3GPP propagation parameters to calculate coverage sites, and busy-hour traffic, uplink/downlink capabilities, and offload ratios to calculate capacity sites, taking the larger figure. It subsequently estimates site, spectrum, and operating costs item by item, applies deployment-speed constraints, and finally validates the model using DISH deployment data, incumbent-carrier density, and one-variable stress tests.

Methodology notes

• (Method Outside the Vocabulary)

  Bottom-Up Site Geometry Model Based on Settlement Types and Wireless Propagation

  The report calculates sites based on actual population distribution, four urban-rural settlement types, frequency propagation radii, and a hexagonal grid rather than assuming a site count upfront. This method captures the rapidly rising marginal cost as coverage increases from 70% to 92% and 95%.

• (Method Outside the Vocabulary)

  Six-Scenario Network Design

  The report combines urban, nationwide, and premium network tiers with either obtaining or not obtaining Grain spectrum to isolate the effects of network-quality targets and low-band spectrum choices on sites, cost, capacity, and time.

• (Method Outside the Vocabulary)

  Greater of Coverage Grid and Capacity Grid

  For each area, the model separately calculates the coverage sites needed to ensure signal availability and the capacity sites required to carry busy-hour traffic, then uses the larger number. Results show that all current scenarios are constrained first by coverage distance.

• Industry/Sector Analysis FrameworkVolume-price decomposition

  Decomposition of Site Count, Per-Site Cost, and Spectrum Cost

  Total investment is decomposed into the site count determined by coverage and capacity, the itemized per-site cost, and independently estimated spectrum cost, demonstrating that spectrum and engineering standards drive the total bill more than per-site pricing.

• (Method Outside the Vocabulary)

  Comparable Spectrum Pricing in $/MHz-POP

  The report compares spectrum transactions and auctions of different bandwidths and geographic scopes using price per megahertz per covered person, then uses this to estimate the potential cost of Grain spectrum and Upper C-band.

• Valuation MethodDCF

  Present Value of 15-Year Network Operating Costs

  Site rent, land, backhaul, power, and maintenance expenses increase by 2.5%-3.0% annually and are discounted at 9% over 15 years to derive long-term cash costs including capital expenditure, spectrum, and operating costs.

• (Method Outside the Vocabulary)

  Benchmark Backtesting and One-Variable Stress Testing

  The report validates the model using DISH's actual deployed sites, T-Mobile's population per tower, and a recalculation using 600MHz, then changes ten inputs individually to identify sensitivity to building penetration, cell-edge standards, spectrum, and unit costs.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• SpaceX (SPCX)

  The report's primary research subject; rated Outperform with a $248 target price, with the feasibility of independently building a U.S. terrestrial D2D network assessed.

  Strengths

  Launch, orbital data centers, and Starlink broadband constitute what the report describes as the overall positive opportunity; the satellite network and installed Starlink terminals can provide backhaul and partial traffic offload.

  Weaknesses

  Lacks low-band coverage spectrum and has limited uplink resources in its existing spectrum portfolio; D2D uplink, indoor coverage, construction capacity, and long-term capital requirements are all constraints.

  Comparison

  The nationwide base-case network's density is close to but below T-Mobile's and materially stronger than DISH's thin grid designed mainly for regulatory compliance.

  Risks

  Grain spectrum has not yet been obtained, Upper C-band activation is delayed, and changes in engineering assumptions could materially alter site counts and costs.

• American Tower (AMT)

  A communications-tower company covered by the report; if SpaceX builds independently, large-scale colocation on third-party towers and lease spending could generate business demand.

  Strengths

  Most scenarios in the report assume primary reliance on existing tower structures rather than new towers.

  Weaknesses

  The related demand depends on whether SpaceX actually builds independently and would be realized gradually over many years.

  Comparison

  Rated Outperform with a $214 target price.

  Risks

  If SpaceX chooses a partnership, reduces coverage, or delays deployment, incremental colocation demand may fall below the modeled scenarios.

• Crown Castle (CCI)

  A communications-infrastructure company covered by the report; potential exposure comes from macro-site colocation, leasing, and terrestrial-network operating expenditure.

  Strengths

  SpaceX's modeled network relies heavily on existing structures, and communications-tower rent is an explicitly included long-term cost.

  Weaknesses

  The scale, starting point, and timing of deployment remain uncertain.

  Comparison

  Rated Outperform with a $96 target price.

  Risks

  Failure to pursue an independent build or constraints on the deployment pace would weaken the potential incremental opportunity.

• SBA Communications (SBAC)

  A communications-tower company covered by the report, with the same potential benefit from colocation demand and site leasing.

  Strengths

  The vast majority of network deployments are assumed to be placed on existing towers or other existing structures.

  Weaknesses

  The report does not allocate potential SpaceX demand among specific tower companies.

  Comparison

  Rated Market-Perform with a $219 target price.

  Risks

  Demand is scenario-dependent, and construction-capacity constraints would delay realization.

• AT&T (T)

  A U.S. mobile carrier covered by the report and a comparison point for existing terrestrial-network and spectrum capabilities.

  Strengths

  Has a low-band coverage layer, and incumbent carriers' spectrum portfolios and network coverage are materially stronger than those of a new entrant.

  Comparison

  Rated Outperform with a $25 target price, both unchanged.

  Risks

  If SpaceX builds a credible nationwide network, it could increase competition in the U.S. mobile market.

• Verizon (VZ)

  An incumbent U.S. mobile carrier covered by the report and used for comparison of low-band spectrum and network capabilities.

  Strengths

  Has a 700MHz low-band coverage layer and far more low- and mid-band spectrum than SpaceX.

  Comparison

  Rated Market-Perform with a $47 target price, both unchanged.

  Risks

  The competitive impact of a potential new network depends on SpaceX's coverage quality, pricing, and deployment progress.

• T-Mobile US (TMUS)

  An incumbent carrier covered by the report and the principal benchmark for site scale, coverage, population per tower, and spectrum capabilities.

  Strengths

  Approximately 82,000 macro sites cover around 98% of the U.S. population, supported by a 600MHz low-band coverage layer.

  Comparison

  Rated Market-Perform with a $220 target price, both unchanged.

  Risks

  If SpaceX obtains Grain low-band spectrum and completes a nationwide build, it could create a more credible competing network.

• Comcast (CMCSA)

  A U.S. communications company covered by the report and included within the scope of telecommunications investment implications.

  Weaknesses

  The report's main text does not discuss its company-specific operating impact.

  Comparison

  Rated Market-Perform with a $28 target price, both unchanged.

• Charter (CHTR)

  A U.S. communications company covered by the report and included within the scope of telecommunications investment implications.

  Weaknesses

  The report's main text does not discuss its company-specific operating impact.

  Comparison

  Rated Market-Perform with a $150 target price, both unchanged.

Key data

• Total Project Cost Across Six Scenarios$46.9B-$126.5BIncludes network capital expenditure and spectrum; the report approximates the range as about $50B-$130B.
• Network Capital Expenditure Excluding Spectrum$14.3B-$78.0BRanges from the urban plus Grain scenario to the premium scenario without Grain.
• Macro-Site Requirements32,362-117,384 sitesDepends on the coverage target and whether Grain low-band spectrum is obtained.
• Nationwide Network Plus Grain Total Cost$71.5BThe report considers this the most likely of the independent-build options.
• Nationwide Network Plus Grain Scale56,705 macro sitesCovers 92% of the population and also includes 25,262 conventional small cells and 4M residential femtocells.
• Nationwide Network Plus Grain Deployment Period8.4 yearsAt the maximum construction rate, with model completion in 2037.
• Nationwide Network Without Grain$79.5B, 85,858 macro sites, 12.7 yearsCompared with the Grain scenario, it requires more sites, costs more, and takes longer.
• Impact of Grain Low-Band SpectrumApproximately 30% fewer sitesThe coverage-engine comparison declines from 88,563 to 58,423 sites; nationwide macro sites in the six-scenario table decline from 85,858 to 56,705.
• Spectrum Cost Share40%-70%Equivalent to approximately $29B-$58B, making it the most critical and difficult-to-estimate component of total cost.
• EchoStar Spectrum Portfolio65MHz, approximately $20B plus approximately $2B in interest fundingOnly 20MHz is available for uplink and 45MHz for downlink; each scenario includes approximately $21.6B of committed spending.
• Grain Band 26 Valuation$4.5B-$8.5BApproximately $0.97-$1.83/MHz-POP, providing around 10MHz of low-band coverage spectrum.
• Upper C-band Cost$6.5B-$14.2BAssumes SpaceX obtains 20-60MHz in Auction 115.
• Upper C-band TimingAuction begins April 27, 2027; available in late 2030 or mid-2031The top 75 markets can activate it no earlier than December 31, 2030, and other areas no earlier than July 1, 2031.
• Population Density DifferenceApproximately 40xAbout 940 people per square mile in urban cores versus approximately 24 per square mile in rural coverage areas.
• Highway Sites in Nationwide Plus Grain Scenario11,413 sitesEquivalent to approximately 24% more sites on top of the population-driven grid.
• Nationwide Scenario Capacity Before Upper C-band54.5M-77.3M usersThe Grain and no-Grain scenarios both target 25.2M users, with capacity headroom of 2.2x and 3.1x, respectively.
• Residential Femtocell Deployment1.5M-8M unitsThe model estimates they can offload 4%-18% of macro-network traffic, with D2D offloading another 0.5%-3.0%.
• Cost per Colocated Macro Site$258K-$364KDISH's actual unit cost is approximately $317,600.
• Model Prediction for DISH Sites22,952 sitesDISH actually has approximately 25,000, so the model underestimates by 8.2%.
• Sensitivity to Engineering Standards38.5% for tower count, 35% for capital billThe magnitude of change when requiring an additional 3dB for deep-building penetration or raising cell-edge standards.
• Unit-Cost Stress TestA 25% increase in all unit costs raises total project cost by 6.4%The impact is limited because spectrum represents about two-thirds of the base-case bill.

Impact & implications

The report believes a technical path exists for SpaceX to independently build a terrestrial network, but low-band coverage, Upper C-band capacity, construction resources, and long-term costs create substantial barriers, making a partnership the more likely mobile-business model. If an independent build actually proceeds, extensive colocated sites and 15 years of lease spending could create conditional demand for communications-tower companies and potentially add a new competitor to the U.S. mobile market. However, deployment would take many years, and the pace of benefit realization would depend on spectrum acquisition and construction progress. The report does not assess network revenue, customer acquisition, handset subsidies, retail, working capital, satellite-constellation costs, or the complete investment return.

Risks

• The fundamental technical risk for D2D remains the handset uplink, as limited transmit power, antenna gain, battery capacity, and indoor penetration losses may hinder reliable two-way communications.
• Grain low-band spectrum has not yet been contracted and faces competition from AST SpaceMobile, leaving SpaceX's ability to obtain a coverage layer uncertain.
• Upper C-band cannot be activated before late 2030 or mid-2031, so a mature network may face delayed availability of its capacity layer.
• Limited availability of new towers, construction crews, zoning approvals, and power interconnections in the United States could extend deployment to approximately 8-13 years.
• Capacity estimates are sensitive to spectral efficiency, future per-user traffic, and the downlink-to-uplink ratio; the report calls this the least robust part of the analysis.
• Changes in building-penetration and cell-edge standards could alter tower counts by 38.5% and the capital bill by 35%.
• The total project estimate excludes customer acquisition, handset subsidies, retail, working capital, and satellite constellations, and does not assess the complete network investment return.

What to watch

• Watch whether SpaceX ultimately chooses to partner with a terrestrial carrier or proceeds with an independent terrestrial network build.
• Watch the satellite partner Grain selects by November 5, 2026 and the FCC application submitted by December 4.
• Watch whether the Grain Band 26 transaction ultimately uses a purchase or long-term lease structure and whether the transaction price falls within the $4.5B-$8.5B range.
• Watch Mobile V2 satellite launch progress beginning in mid-2027 and actual downlink, uplink, indoor-penetration, and concurrent-user performance.
• Watch the Upper C-band bandwidth and cost SpaceX secures after Auction 115 begins on April 27, 2027.
• Watch whether the EchoStar spectrum transaction closes by the target date of November 30, 2027.
• Watch whether colocation rates on existing towers, new-tower construction capacity, labor, zoning approvals, and power interconnections meet the model's upper limits.
• Watch how approximately 13% growth in per-user traffic and a declining downlink-to-uplink traffic ratio affect the timing of Upper C-band requirements.