Physical AI in the field.
Telecommunications operators run one of the largest distributed physical estates in any industry and hold an approximate record of its contents. The national site counts differ by a factor of 1.76 depending on who is asking, and below ground the Common Ground Alliance's 2024 DIRT Report analysed 196,977 damage reports to buried utilities across North America, with about one damage in five traced to a facility that was unmarked or marked in the wrong place. This review states ten hypotheses about where Physical AI and logistics management are carrying value in that estate, tests each against graded evidence, and resolves each to its position on the trajectory, the constraint currently binding it, and the measured change that would move the boundary.
Ronnie Gomez, Industrial Research Fellow · The Charlot Lab
$54.7bn
US cellular opex, 2025
Against more than $10.2bn of capacity-and-coverage capex, a ratio of 5.4 to 1. That capex line is capacity and coverage only and is not comparable to a total-capex figure.
modelled input (iGR for WIA)
1.76×
gap between national site counts
CTIA counts 447,605 cell sites; WIA counts 254,850 macrocells. The two differ in reference date and in scope, and this review did not locate a published methodology from either body that reconciles them.
reported vs modelled
$21-36M
a year, drone tower inspection nationally
Deployed at carrier scale since 2016-17 and admitted by ANSI/TIA-222 as a data-collection method since January 2024, scaled from the inspection interval at vendor-quoted savings. Two constraints hold that figure where it is and both are quantified.
modelled on a modelled base
22×
less coverage per battery pair at close standoff
The same 30 minutes of usable hover covers about 14,000 square metres at a 10 m standoff and 650 at the 2.1 m needed to resolve a 0.8 mm feature. Endurance converts an area threshold into a sortie count and therefore into a price per tower.
modelled (author's arithmetic)
Ten hypotheses, ten constraints.
What the recurring work of touching the plant costs, which applications are already carrying load at national scale, and what specifically holds the rest where they are.
Any autonomous system that acts on telecom plant needs to know what is there, and the industry does not agree on the answer even in aggregate. CTIA's member survey reports 447,605 cell sites at the end of 2024. WIA's report, modelled by iGR, gives 254,850 macrocells and 158,500 purpose-built towers, and its indoor node count alone exceeds CTIA's entire site total. That spread at national level is the top of a hierarchy of registry error that runs down to individual assets, and it is why the plant model sits underneath the other nine hypotheses.
The money sits in operations. WIA and iGR put US cellular opex at about $54.7bn against more than $10.2bn of capacity-and-coverage capex, which is 5.4 times expansion capital on those figures. That capex line covers capacity and coverage only and is not comparable to a wider total-capex definition. Labour is 72% of the labour-plus-materials line for underground fiber and 67.9% of median total deployment cost on the Fiber Broadband Association's own figures. Two applications are furthest along the trajectory and both are software running on an asset record: dispatch avoidance and technician decision support carry documented load at national scale on operator-published evidence, and the constraint on the next increment there is calibration rather than capability, since an accuracy figure without a precision and recall split does not fix an operating point.
Aerial structural inspection is deployed at carrier scale and admitted by the tower standard, and two constraints hold its present national value at $21-36M a year. The regulatory one is Part 108: under 2% of BVLOS flights in the FAA's flagship BEYOND programme were flown without a visual observer through October 2024 on the DOT Inspector General's audit, and the rule reached OIRA review on 10 July 2026. The physical one is a standoff conflict computed in the review: the composite general-population exposure reference level sits 15.6 m from a filed macro sector while putting a 0.8 mm feature across three pixels needs 2.1 m, and coverage per battery pair falls by a factor of 22 across that interval. Neither contour is a flight rule, so the standoff enters as operator policy, and a scheduled sector power reduction during a pass is a procedural change under the operator's own control. The optical change that relaxes the resolution term is an entrance pupil of about 25 mm against the roughly 15 mm a 168 mm equivalent telephoto has at f/2.8, a factor of 1.7.
The inspection market, drivable.
The $21-36M figure is an arithmetic object built from four inputs, so it is packaged as an instrument rather than an assertion. Drive the tower base, the inspection interval and the per-tower saving and watch where the national figure goes across the whole plausible range, which is what tells you which input a programme would have to move. A second panel shows the per-site denominator bracket that business cases in this field pick a side of.
Telecom field economics
Two panels. The first scales drone structural inspection from the TIA-222 interval to a national figure and shows it against US cellular opex. The second divides cellular opex by macrocells and then by all nodes, and reports the bracket that results. Every default is the review's own figure and carries its grade. Runs entirely on your own device.
Open the console →
25 sourced records.
Every operator programme, vendor system and regulatory finding the review tracks, with what it is, what has actually been fielded, and the grade of the evidence behind it. A claim graded self-published comes from a party selling the thing being measured.
Twenty-five deployment records drawn from TR-2026-30's ledger of what telecom Physical AI has actually put in the field, covering aerial capture and inspection, ground and in-plant robotics, and the asset models and field systems that decide whether anyone is sent at all; grades follow the paper's convention, where verified means independently replicated or drawn from a regulator, standards body or peer-reviewed source, reported means a named analyst house or operator stated it without independent confirmation, self-published means the figure comes from a party selling the thing being measured, and modelled means it was computed rather than observed.
Aerial inspection and capture (7)
| System | Who | Evidence | What it is | Status |
|---|---|---|---|---|
| Verizon asset digital twin programme, with vHive | operator | self-published | drone-derived digital twins of cell sites, with capture and digitisation of every new build and modificationSite count is given as a range rather than a figure, and no accuracy metric, cost per twin or refresh cadence for the legacy estate has been published. The fifteen-minute disaster-assessment claim is not independently verified. | Deployed across tens of thousands of cell sites; automatic capture on every build or modification from early 2025; drone capture cutting some disaster-response site assessments from several hours to about fifteen minutes |
| American Tower and Crown Castle twin programmes | operator | self-published | drone-derived digital twins of tower structuresThe survey has no published methodology, sample or comparison set, and neither towerco discloses how many structures have been scanned or at what accuracy. | Both build drone-derived twins; American Tower reports being rated industry leader in quality of asset information in its own 2025 customer survey |
| AeroDeploy, Drone Harmony, DroniTech, Voliro, ABJ Drones | vendor | self-published | drone structural inspection of towers, sold against a climbing crewEvery figure traces to a company selling the service, and no independent, operator-audited or peer-reviewed comparison could be found. The comparison is not scope-matched, since a drone pass performs no torque checks, connector work, grounding continuity testing or hands-on remediation. On the quoted midpoints the implied saving is 59%, and the full spread of the quoted ranges runs from 25% to 77%. | Offered commercially at $350-600 per tower against $800-1,500 for a climbing crew, with drone crews completing four to six towers a day against one to two, and 30-50% overall cost savings claimed |
| AT&T drone tower inspection | operator | self-published | drone inspection of the carrier's own tower estateAn AT&T official declined to quantify the saving beyond that, and no later cost figure from AT&T has been located in public material. The absence is not itself evidence of the size of the effect. The 65,000 figure is the estate as characterised in 2016 and predates subsequent tower sales; the material is nine to ten years old. | Announced in 2016 across roughly 65,000 cell sites where AT&T has equipment, with the programme stated to break even during the initial ramp-up period |
| FAA BEYOND, PSP and IPA programmes | regulator | verified | the FAA's partnership programmes for beyond-visual-line-of-sight flight, audited by the US DOT Office of Inspector GeneralThe figures cover the partnership programmes only, so commercial waiver holders outside them are not counted, and the OIG found FAA's data validation process vulnerable to errors, having mis-categorised at least one operator's observer status. Telecom operators largely inspect under ordinary Part 107 within visual line of sight, so their flight volume never enters this dataset. | Fewer than 763 of more than 44,000 BEYOND flights through October 2024 were flown without a visual observer, about 2%, and no PSP or IPA operator flew a single flight without one. Infrastructure inspection was about 1% of more than 60,000 flights, roughly 600 flights over four years. FAA had issued 190 BVLOS waivers to 134 distinct operators as of October 2024 |
| FAA Part 108 BVLOS rule | regulator | reported | the rule that would permit routine flight beyond visual line of sightNone of the trade sources reporting this has an FAA commitment date, and OIRA timelines slip routinely. The detect-and-avoid and operator-certification requirements are not public, so the cost of compliance is unknown and could be large enough to change the per-site economics on its own. | Proposed 7 August 2025, drew roughly 3,000-3,100 comments, and reached OIRA review on 10 July 2026, with publication expected late 2026 or early 2027 and a six to twelve month transition after that |
| ANSI/TIA-222 inspection intervals | standards body | verified | the standard that sets recommended structural inspection intervals for towersTIA's TR-14 committee states explicitly that this is a recommendation rather than a mandate. Applied to 158,500 purpose-built towers it implies roughly 32,000 to 53,000 structural inspections a year, on a guyed and self-support mix that is not published. The 32,000 to 53,000 range is modelled on a tower base that is itself modelled. | Recommends major structural inspection every three years for guyed masts and every five years for self-supporting structures, plus after severe weather |
Ground and in-plant robotics (9)
| System | Who | Evidence | What it is | Status |
|---|---|---|---|---|
| Openreach MicroBORE, with Synthotech and the University of Surrey | operator | self-published | trenchless robotic drilling that bores narrow channels under pavements and driveways for FTTP dropsNo bore rate, cost per drop, failure rate or comparison against manual moling has been published, and Openreach's £15bn programme to reach 25 million premises by December 2026, about £600 per premises passed on the review's own arithmetic, is being met without it. | In trial; developed over five years by BT Group, Synthotech and the University of Surrey from lunar and Martian drilling research, with early testing at BT's Suffolk research centre confirming operational viability and broader deployment under evaluation |
| Built Robotics autonomous trenching | vendor | self-published | a retrofit that makes a standard excavator trench autonomouslyThe saving is a vendor-reported contractor figure with no named contractor or measured baseline, and no telecom fiber-build deployment of the trenching product is publicly documented. | Retrofit completed in about a day, with contractors reporting 20-30% savings |
| Built Robotics with Blattner | vendor | self-published | autonomous solar pile driving, which is the company's volume businessThe best-funded autonomous trenching company found product-market fit in open, repetitive, greenfield geometry. Telecom conduit is short-run, congested with existing utilities and constrained by surface restoration, which is close to the opposite. | Nationwide agreement announced in September 2025 |
| AT&T with Ghost Robotics | operator | self-published | quadrupedal ground vehicles commanded beyond line of sight over FirstNet using AT&T's Geocast serviceThis is a connectivity product sold to others. No fleet size, deployment count or telecom-facility use case is disclosed, and programme announcements date from 2023 with limited public updates since. The 65,000 figure is the estate as characterised in 2016 and predates subsequent tower sales; the material is nine to ten years old. | Fielded with documented use in public safety and defence, including Air Force flight-line and perimeter patrol at Tyndall AFB, rather than inspection of AT&T's own plant |
| NTT DATA data centre inspection robots | operator | self-published | facility inspection robots working inside data centresThis is the clearest at-scale ground-robot inspection programme inside telecom-owned facilities, and it sits in controlled indoor environments rather than central offices or outside plant. No current fleet count, no measured labour saving, and not independently verified. | Rolled out across Japanese sites from April 2023, with the offering moved to commercialisation |
| China Telecom with AgiBot | operator | self-published | a Robotics-as-a-Service model for aviation and logisticsAn announced programme from a state-linked entity with no independent verification of units in service, uptime or task success. If the model works it is a services line for carriers, which is a different business thesis from internal cost reduction. | Announced at MWC Shanghai in June 2026 |
| China Unicom hazard inspection robots | operator | self-published | robot deployment for hazard inspection in chemical plantsAn announced programme from a state-linked entity with no independent verification of units in service, uptime or task success. | Described at MWC Shanghai in June 2026 |
| State Grid robot programme | utility | self-published | robots for grid inspection and maintenanceAnnounced programme figures from a state-linked entity with no independent verification of units in service, uptime or task success rate, and deployment targets of this kind routinely diverge from delivery. | Announced plans for roughly 8,500 robots: about 5,000 quadrupeds, 3,000 dual-arm wheeled units and 500 humanoids |
| Contact servicing of towers | academic literature | verified | robots that physically service towers, torquing a bolt, dressing a connector or testing grounding continuityThe full citation was truncated in the compiled evidence record and must be completed before publication; no quantitative claim in the report rests on it. | Feasibility stage; no fielded system. Everything deployed at scale in telecom is non-contact sensing |
Asset models and field systems (9)
| System | Who | Evidence | What it is | Status |
|---|---|---|---|---|
| vHive | vendor | self-published | the drone-capture and digital-twin software behind the largest operator twin programmesCounts of twins generated conflate first captures, re-captures and non-telecom verticals, and no breakdown by industry or customer is published. The plant model is becoming a system of record, and concentration in the layer that holds it is a strategic exposure. | Reports generating tens of thousands of twins in a year across telecom, construction and cranes, with Verizon and Deutsche Telekom named as customers; raised a $25M Series B in April 2022 led by PSG |
| Deutsche Telekom | operator | self-published | named vHive customer, and an investor in the supplier through the Telekom Innovation PoolDT is an investor in the supplier as well as a named customer of it. | Named as a vHive customer; Telekom Innovation Pool participated in vHive's Series B round |
| Consilience Analytics | vendor | self-published | twin production processed into the Bentley stackThe count is global, spans sectors and is cumulative across repeat captures, so it is an upper bound on unique sites, cannot be set against the US tower estate as a coverage figure, and no such ratio is computed. | States it has processed over 35,000 twins since 2019 |
| Bentley OpenTower iQ, accuracy validation | vendor | self-published | automated equipment detection and dimension measurement from tower captureThe sample was four towers drawn from a group of fifty, and an accuracy rate from four observations carries a binomial interval spanning most of the usable range, so the 95% is the vendor's characterisation rather than a measurement. The comparison baseline was best-existing data, meaning RF data sheets, old structural analyses and CAD drawings, so the study measures agreement with the records whose reliability is the thing in question. The Bentley whitepaper does not state whether the 5% is of reading or of full scale. | Sponsored validation found equipment detection and dimension measurement 95% accurate and tilt and azimuth accurate within a 5% error margin, with tower height on one example reading 72 ft 3 in against 73 ft in the design documents, a 1.03% difference |
| Bentley OpenTower iQ, O&M cost model | vendor | modelled | a model of tower operations and maintenance cost with and without the twinThe per-tower figures imply a 61% cut in the stated O&M line, which the same document does not reconcile with its own 15% and 42% claims, and the three percentages are computed over different denominators that are never named. No operator has published a matched-cohort before-and-after, and the model excludes twin currency maintenance, re-capture after each site modification, and pilot and mobilisation cost. | Estimates about $5,700 per tower against $2,200 with OpenTower iQ, a $3,500 annual saving, alongside 50% reporting time savings, a 15% annual opex reduction and 42% lower ten-year TCO per tower |
| Comcast dispatch prediction | operator | self-published | a machine learning system that predicts whether a technician needs to be dispatchedNo baseline dispatch volume, no time period and no definition of what counts as an avoided roll. Accuracy without a precision and recall split is close to meaningless for an imbalanced dispatch decision where a false negative leaves a customer broken and a false positive costs a van. | Reported at about 90% accuracy and credited with saving tens of millions of dollars in avoided truck rolls |
| Comcast storm restoration | operator | self-published | alarm clustering plus tooling dispatch for mass outagesEffectiveness is not defined and the 50% claim has no stated denominator, with no third-party verification and no per-storm breakdown. The opportunity is not sized in any public source. | Trialled during the 2024 hurricane season and operational nationwide from September 2025, described as 50% more effective |
| Vodafone Field Technician Assist, with Celfocus | operator | self-published | top-three recommended actions delivered to field engineers from intervention history, built on Google CloudOperator's own figure with no baseline period, market or fault mix disclosed and no independent audit, and repeated visits is not defined in public materials. Applied to TSIA's 13% repeat rate it removes about 3.6% of total visits, which is the review's own arithmetic on a TSIA technology-services benchmark rather than a telecom field measurement. | Credited with a 28% reduction in repeated field visits |
| Vodafone remote visual assistance, with TechSee | operator | self-published | remote visual assistance that resolves faults without sending a technicianPublished by the supplier TechSee rather than by Vodafone, with no date, market, baseline or measurement period stated. | Reduced technician dispatch rate by 10%, raised first-contact resolution five points and raised NPS fifteen points |
One measurement this lab can produce.
The review nominates six measurements that would close its largest evidence gaps. Five are producible from data an operator already holds or from a field study of modest scale. This is the one that sits inside the lab's own remit.
vHive states that electromagnetic interference around telecom towers disrupts RTK positioning badly enough to force manual flights and repeat site visits, and this review did not locate error figures or a degradation curve against transmit power or frequency accompanying that claim. The physical situation is specific: the asset being modelled is a high-power RF emitter, and the sensor modelling it depends on a weak GNSS signal. The review bounds the problem at a separation of order 140 dB and an implied 103 to 108 dB of front-end rejection. A survey-grade receiver, a spectrum analyser and access to live and dark sites would produce a degradation curve against transmit power, frequency, distance and bearing. It is measurable, it sits within the lab's on-device state-estimation remit, it bears on hypotheses 1, 3 and 4, and it would convert a bounded estimate into a measured one.
Telecom operators run one of the largest distributed physical estates in any industry, and they already hold a record of what is in it. Approximate is the operative word, and it is the whole opportunity: the integration itself is tractable, and the inventory it would run against is the piece worth making trustworthy first.
One of eight, and only one of them is physics. How we read a frontier →