Telecom business model

4.3.1 Why the returns are structurally hard

Telecom looks like a dream on the P&L: 40–55% EBITDA margins, utility-grade demand, a service nobody will give up. On the balance sheet it looks like a grinder. Four forces stack up behind that.

1. Capex comes first, forever. The network must exist before the first rupee of revenue. It must cover marginal geographies to be sellable anywhere, because coverage is the product. And it must be rebuilt roughly every decade as the technology generation turns from 2G to 3G to 4G to 5G. Each cycle demands new radios, denser sites, more fiber backhaul. It often arrives before the prior generation has paid itself back. This is the capex treadmill. Capex intensity (capex ÷ revenue) runs 25–35% during rollout years and rarely falls below the high teens even in "digestion" years. Compare that with an FMCG company, which reinvests 3–4% of revenue, or an IT major at 1–2%.

2. Spectrum: the state auctions the oxygen. Radio spectrum is the finite set of frequency ranges the service physically runs on. It's government property, and the government sells it at auction. Auctions are brutally efficient at extracting the industry's future economic rent upfront. When several bidders need a scarce block to survive, bidding approaches the full expected value of owning it, and then goes past it. That overshoot is the winner's curse.

The sums involved are enormous. The 3G auctions of 2000 in Europe (~£22.5bn in the UK, ~€50bn in Germany) crippled a generation of operators. India's 2010 3G auction raised ~₹67,700 cr and its 2022 5G auction ~₹1.5 lakh cr, while the US 2021 C-band auction cleared ~$81bn. All of these are approximate, so read them for the order of magnitude. Whatever moat "efficient scale" might have granted the operators, the auction transfers a large slice of its NPV to the government before service launch. Spectrum then goes on the balance sheet as an intangible, usually paid in installments, and deferred spectrum payables are debt in everything but the label. You'll see what that does to EV when we get to valuation.

3. The product is a commodity with near-zero marginal cost. A gigabyte is a gigabyte. Number portability makes switching nearly free. And once the network is built, serving one more subscriber costs almost nothing. You've seen what that combination does in the five-forces work: in a fragmented market, price gravitates toward marginal cost, which is far below average cost. It's the airline disease with towers. The only durable rescue is market structure, meaning consolidation to 2–3 players who compete on coverage and brand and leave price alone. This is efficient scale, the telecom moat.

Why does a 50% EBITDA margin still earn a 7.5% ROIC?

4. The arithmetic of "great margins, poor returns." Watch the whole illusion in five lines (synthetic, ₹):

So you get a 50% EBITDA margin and a 7.5% ROIC against an 11–12% Indian WACC. That's value destruction at scale, in a company journalists call a cash machine. EBITDA is 50% of revenue, while the capital is 240% of revenue.

That one table explains why the sector's valuation lens must be EV-based and capex-aware, and never the P&L alone. It also explains why Phase 4's mantra, that the moat must show up as ROIC above WACC, works harder here than anywhere else you'll apply it. So when does telecom clear the bar? When structure and pricing behavior line up: a consolidated market, repaired pricing, capex intensity descending after a coverage peak, and leverage under control.

Those conditions are checkable. That's what the KPI panel is for.

4.3.2 India's three-player structure vs the US: the capital cycle, live

India ran the whole capital cycle in a decade. Before 2016 it had ~10–12 operators locked in a chronic price war, and industry ROIC was deeply negative. In September 2016 Reliance Jio launched with a nationwide 4G-only network, free voice, and data at a fraction of prevailing prices. It was a classic Greenwald attack, funded by a parent with petrochemical cash flows. Blended industry ARPU collapsed from ~₹180–200 (2015) toward ~₹100–120 (2018–19). A dozen operators became three private players (Jio, Bharti Airtel, and Vodafone Idea, the merged survivors) plus state-owned BSNL.

Then the cycle turned. Tariff "repair" rounds in Dec 2019, Nov 2021, and Jul 2024, with hikes of ~20–25% each (approximate), lifted ARPU back toward ₹200–250 for the leaders. Here the capital-cycle lesson plays out in a single sector: returns are made and destroyed by supply-side behavior, and the moment of maximum pessimism (2018's ARPU trough) was the moment supply finished exiting.

How does the US telecom market differ from India's?

The US is the mature version. Decades of consolidation left three national facilities-based players (Verizon, T-Mobile, AT&T). The market is dominated by postpaid contracts. Monthly postpaid phone churn runs under 1%, and ARPU is an order of magnitude above India's at ~$40–50 per line. Note that US firms increasingly report ARPA, average revenue per account. It's a different denominator, and you must never mix the two. Yet even in this "good" structure, returns are merely adequate. The treadmill (C-band, 5G densification, fiber) grinds on, promotional wars flare around iPhone cycles, and growth has to be scraped from fixed-wireless access and convergence bundles. So consolidation is necessary but not sufficient.

It stops the bleeding. It doesn't repeal the capex physics.

India adds its own overlay. There, ~90%+ of subscribers are prepaid, so revenue is collected in advance and customer working capital is a small mercy of the model. ARPU is among the world's lowest, while data usage per subscriber (~25–30 GB/month, illustrative) is among the world's highest. Read that one way and it's the bull case: enormous engagement, barely monetized. Read it the other way and it's the bear case, because price hikes meet a price-sensitive prepaid base that can down-trade instantly. And one player, Vodafone Idea, entered the repair era carrying the sector's heaviest regulatory burden. That brings us to AGR.

4.3.3 AGR: the dispute that reshaped a sector's balance sheets

The mechanics. Indian operators pay the government revenue-linked levies. The first is a license fee of 8% of AGR (Adjusted Gross Revenue), which includes a 5-percentage-point universal-service component. On top of that come spectrum usage charges, a further ~3–5% weighted, on older spectrum. SUC was abolished for spectrum bought in auctions from 2022 onward. Everything therefore turns on the definition of AGR.

The dispute. From the early 2000s, the Department of Telecommunications read AGR to mean all revenue, including non-telecom items like interest income, dividends, forex gains, and asset-sale proceeds. Operators contested that reading for nearly fifteen years. They paid levies on their own narrower reading while the difference compounded on paper at contractual interest plus penalties plus interest on penalties.

In October 2019 the Supreme Court ruled for the government's definition in full, retrospectively. Aggregate demands landed around ₹1.4–1.5 lakh crore across the industry (approximate), and the large majority of it was interest and penalties. Principal was the smaller part. Bharti Airtel absorbed demands in the tens of thousands of crores. Vodafone Idea's share was ~₹58,000+ cr on AGR alone, on top of deferred spectrum dues. That exceeded its market capitalization and pushed it to the edge of collapse.

An argument over the definition of revenue nearly took out a national operator.

Then the state blinked halfway. It offered staged payment timelines (to ~2031), a 2021 relief package with a four-year moratorium on AGR and spectrum installments, and the option to convert accrued interest, and later dues, into government equity. Through conversions in 2023 and 2025 the Government of India became Vodafone Idea's largest shareholder at roughly half the company. The figure is as of mid-2026, so verify the current stake. Further conversions and litigation over remaining dues were still live.

You'll want to carry this case with you permanently, because it leaves four legacies:

1. Regulatory-definition risk is real risk. A levy defined as "% of revenue" is only as safe as the definition of revenue. The AGR case is your template for reading any revenue-share regime (mining royalties, gaming levies, airport concessions). Find the definition, find the dispute, size the retrospective tail.
2. Contingent liabilities can become debt overnight. For years AGR sat in the contingent-liabilities note at amounts analysts waved past. One judgment moved it to the balance sheet. Hold on to the forensic rule: size every contingency against net worth before you dismiss it.
3. "Deferred obligations" are debt. AGR installments and deferred spectrum payables carry interest, have fixed schedules, and rank ahead of shareholders. Every leverage ratio in the valuation section includes them, whatever presentation the company prefers.
4. The state as shareholder changes the game. A government that is at once regulator, tax collector, licensor, and largest shareholder of the weakest player distorts pricing, competition, and any bankruptcy endgame. Factor that into scenario work on the whole sector. Its reach goes well beyond the affected firm.

4.3.4 The telecom KPI panel

How do you calculate ARPU and churn?

ARPU, computed properly. ARPU = service revenue for the period ÷ average subscribers ÷ months in the period. Try it on BharatCell (synthetic), which reports quarterly mobile-services revenue of ₹17,450 cr. Subscribers went from 285m opening to 291m closing, an average of 288m. ARPU = 174,500m ÷ (288m × 3) = 174,500/864 = ₹202/month.

Three habits keep this number right. Use average subscribers, because closing-base ARPU understates in growth and flatters in decline. Match the numerator by using mobile-services revenue only. Leave out equipment and enterprise lines. US reporters mix these in, and India reporters mostly don't. Know your denominator. India reports per subscriber. The US reports per line or per account, and an account may hold four lines, so ARPA ≈ 3–4× ARPU. Then there's "smartphone ARPU" versus blended.

ARPU's mix trap. ARPU is the most misread number in the sector. It's an average, so it rises when pricing improves or when poor customers leave. After India's tariff hikes, weaker operators printed rising ARPU while losing tens of millions of subscribers. Their average rose because the base shrank from the bottom, as multi-SIM users dropped their second SIM.

So always read ARPU jointly with net adds. Rising ARPU with rising subs is pricing power, the real thing. Rising ARPU with falling subs is shrinking into premiumization. That may be fine, or it may be terminal. And flat ARPU despite announced tariff hikes means the hike is not sticking, because customers are down-trading to cheaper packs. It's a listed red flag, and an early sign that pricing restraint across the market is cracking.

Churn. Monthly churn = disconnections in period ÷ average subscribers ÷ months. Back to BharatCell's quarter. Gross adds were 24m and net adds +6m. That means disconnections of 18m, and churn = 18/(288 × 3) = 2.08%/month. Churn at that level is healthy (<2.5%). Prepaid churn is structurally higher than postpaid. There's no contract and no bundled device. India's blended churn runs ~2–3%, against US postpaid phone churn of ~0.8–1.2%. Churn above 4%/month alongside subscriber losses is the crisis pairing. The base is fleeing faster than marketing can refill it, and each replacement gross add carries acquisition cost (commissions, SIM, onboarding) that churn immediately wastes.

How do you tell a premium subscriber base from a padded one?

Subscriber share vs revenue share: the quality test. TRAI publishes both, with revenue share measured via AGR. An operator whose revenue share exceeds its subscriber share owns the premium base. The reverse means a bloated, low-value base. Picture a synthetic market of 900m subs and ₹60,000 cr of quarterly industry revenue. Operator K has 350m subs (38.9% share) and ₹27,000 cr of revenue (45.0% share), for an ARPU of ₹257. Operator M has 210m subs (23.3%) and ₹9,000 cr (15.0%), for an ARPU of ₹143. K monetizes its share at 1.16× (45.0/38.9), and M at 0.64×. Revenue share is also the more trustworthy share metric across operators, because free SIMs can't inflate it. Free SIMs can inflate subscriber counts, though. The next metric exists to catch that.

VLR / active-SIM %. Every month, India's regulator publishes the share of each operator's reported subscribers actually registered on the network. VLR stands for Visitor Location Register, the switch's list of live phones. Healthy franchises run >95%, and leaders ~99%+. Historically, weak operators reported VLR in the 60–70%s. A third of their "subscribers" were dead SIMs kept on the books. Any per-subscriber metric (ARPU, EV/sub, market share) computed on a low-VLR base is fiction, so recompute it on active subs. It's the forensic revenue-quality habit in telecom form: always ask what fraction of the reported unit is real.

Why does cutting capex hurt a telecom operator?

Capex intensity and underinvestment. The treadmill means there's a maintenance level of capex, and below it the network decays: slower speeds, dropped calls, coverage holes. Share then flows to whoever kept investing, with a 12–24 month lag. Picture an operator "improving FCF" by cutting capex intensity from 25% to 12–14% while peers spend 25–30%. It is liquidating the franchise in slow motion. Vodafone Idea's history shows the anatomy. Capex starvation opens a network-quality gap. Port-outs follow. Revenue shrinks, and the capex budget shrinks with it. Read capex jointly with net adds and churn relative to peers. And remember spectrum. "Capex" disclosures usually exclude spectrum outlays, so add auction payments separately when you compute true reinvestment.

4.3.5 Valuation: EV/EBITDA, adjusted or worthless

Why EV/EBITDA? Because earnings here are polluted by enormous, policy-sensitive D&A and by leverage differences, so P/E is noise, the mirror image of IT services. EBITDA at least approximates pre-capex operating cash. But the sector demands four adjustments before the multiple means anything:

(1) Rebuild the EV in full. EV = market cap + net debt + deferred spectrum payables + AGR/regulatory dues + lease liabilities − non-core investments (e.g., listed tower-company stakes). Work it on DeccanTel (synthetic). Market cap is ₹4,20,000 cr. Conventional net debt is ₹65,000 cr, deferred spectrum ₹85,000 cr, AGR dues ₹28,000 cr, lease liabilities ₹52,000 cr. It also holds a listed tower stake worth ₹30,000 cr. EBITDA is ₹65,000 cr.

• Naive EV = 420,000 + 65,000 = ₹4,85,000 cr → EV/EBITDA = 7.5×
• True EV = 420,000 + 65,000 + 85,000 + 28,000 + 52,000 − 30,000 = ₹6,20,000 cr → 9.5×

Two full turns of EBITDA were hiding in the notes.

Any screener that shows you "7.5×" is wrong, and wrong in the same direction for every Indian telco.

(2) Respect the lease boundary: the pairing rule. Since Ind AS 116/IFRS 16, tower and site rents have left operating expenses. EBITDA is inflated by the rent add-back. Lease depreciation and interest appear below it. US GAAP's ASC 842 keeps operating-lease rent in opex, so an Indian telco's EBITDA margin isn't comparable to a US peer's without adjustment. You met the same trap in the retail playbook. That leaves two internally consistent choices. Either use reported (post-116) EBITDA with lease liabilities inside EV, or use EBITDAaL (EBITDA after lease payments) with lease liabilities excluded. DeccanTel pays ₹9,500 cr of leases, so EBITDAaL = 55,500 and EV ex-leases = 5,68,000, giving 10.2×. Never mix the pairings. Reported EBITDA with a lease-free EV flatters by construction.

(3) Cross-check against capex, always. EBITDA is pre-capex, in a sector where capex eats a quarter of revenue. DeccanTel spends ₹38,000 cr of capex, so EV/(EBITDA − capex) = 6,20,000/27,000 = 23×. Now take two telcos at the same headline EV/EBITDA, one with capex intensity of 18% and one with 30%. They aren't the same price. The second is far more expensive per rupee of actual cash. EV/(EBITDA − capex), or an explicit FCF build, is your check against the illusion.

(4) Leverage on the inclusive definition. On DeccanTel, conventional net debt/EBITDA = 65,000/65,000 = 1.0×, which looks pristine. Include spectrum + AGR + leases and you get 230,000/65,000 = 3.5×, above the <3× line. It's the same company. All that changed is what you were willing to count. Apply the threshold to the inclusive number, and watch its direction. Net debt/EBITDA rising while spectrum dues mount means the treadmill is winning. It's a red flag in its own right, and it's the entry ramp to the death spiral: dues → less capex → share loss → less EBITDA → higher ratio → repeat.

One cross-standard footnote is worth real money. US GAAP treats FCC spectrum licenses as indefinite-lived intangibles and tests them for impairment instead of amortizing them. Indian operators amortize spectrum over the license period. So a US telco's EBIT is structurally flattered relative to an Indian peer's at identical economics. That's one more reason to compare at the EBITDA layer, with the adjustments above. It's also why you read D&A policy in the notes. Typical ranges, illustrative as of mid-2026: Indian leaders trade at ~8–12× adjusted EV/EBITDA on a growth-and-repair story, and US majors at ~6–7.5× as mature, dividend-anchored names. A per-subscriber cross-check (EV ÷ active subs) works as a sanity metric within one market. Across markets with 20× ARPU gaps it tells you nothing.

Anchors, one from each market. These are illustrative, modeled on ~FY24/FY25 scale, so pull the actual filings. Bharti Airtel has consolidated revenue of ~₹1.5–1.7 lakh cr and India mobile ARPU of ~₹245–250 after the 2024 hike. Its consolidated EBITDA margin is ~52–54% (post-Ind AS 116), and net debt including leases is ~₹2–2.2 lakh cr, or ~2.0–2.3× EBITDA. Bharti is the premium name of the repair story, with revenue share comfortably above subscriber share. Jio (within Reliance) has the largest base (~470–490m) at an ARPU of ~₹200–210, and it's still the price-setter. Vodafone Idea is the cautionary tail. Its ARPU is the lowest of the three and its subscriber base has been shrinking for years. AGR + spectrum obligations dwarf market EBITDA, and the government owns about half the company.

Verizon has revenue of ~$135bn and an EBITDA margin of ~35–36%. Verizon reports under ASC 842, so remember the boundary before you compare that 36% with Bharti's 53%. Net debt is ~$150bn+, or ~2.5–3.0×, and capex runs ~$17–18bn (~13%, mature-market digestion). Postpaid phone churn is ~0.9%/month. Investors own the equity for its dividend, and it's priced like the regulated utility it economically resembles.

4.3.6 Telecom red flags