India needs 1.32 million chargers by 2030. The grid doesn't know yet.

A 45x build problem nobody is sized for

The Confederation of Indian Industry pegs the 2030 number at 1.32 million public charging stations. That figure is aligned to a 30% EV share of new vehicle sales by the end of the decade.

The current count, reported widely in the Indian industry press at the start of 2026, is about 29,000 public chargers. That number reflects a six-fold expansion since 2022, when the country had roughly 5,000. Fast growth on a small base.

To hit 1.32 million on schedule, India has to install on the order of 400,000 chargers per year, every year, from now until 2030. The PM E-DRIVE scheme, notified in October 2024, commits ₹10,900 crore total, of which ₹2,000 crore is earmarked for charging infrastructure. The scheme supports the equivalent of about 2,000 stations directly. Private capital has to do the rest. The math doesn't close on public money alone.

This is the chart everyone shares on the way to a panel discussion. It is also the chart that misses the operating problem hidden inside it.

The ratio gap is real, but the wrong frame

India sits at roughly one public charger for every 235 EVs. Pick whichever benchmark you want. The country is at least an order of magnitude behind the rest of the world.

The reflex on seeing that chart is to copy the American playbook: more chargers, faster, with subsidies, on the highways. That reflex is wrong because Indian mobility doesn't look like American mobility.

India's mobility shape is two-wheelers and autos

From DD News reporting on 2025 EV sales, the segment mix is dominated by small vehicles.

Two-wheelers were 57% of EV sales. Three-wheelers added another 35%. Combined, they're 91% of the Indian EV market. That isn't a temporary stage on the way to a Tesla future. It is the structure of how Indians move.

The implication for charging is concrete. The American picture, where a sedan pulls into a 250 kW DC fast charger off an interstate, is the wrong reference frame. India's load shape looks more like a million Ola scooters slow-charging at home, two million autos topping up at depots overnight, and a relatively thin layer of DC fast chargers along the Golden Quadrilateral.

That mix produces a different grid problem. Two-wheelers and three-wheelers do most of their charging at locations that aren't on industrial feeders. They draw modest peak power but they draw it often, on residential and small-commercial connections, in places the discom was not historically watching at minute granularity. A 3 kW two-wheeler charger plugged into a tier-3 city neighborhood feeder is not the kind of load India's distribution network was designed to track.

In other words: 1.32 million chargers, weighted toward slow and shared, distributed across the country, on a grid that has historically been visible at quarterly billing resolution.

Why this is not a hardware problem

The reflex when people see "we need 1.32 million chargers" is to think about manufacturing, supply chain, real estate, electrical engineering. Those things matter. They are not the binding constraint.

The binding constraint is the control surface.

A charger, considered alone, is mostly a power electronics box with a tariff lookup and a connector. The hard part is the layer above it. Who issued the command to draw 22 kW? Was it the user, the operator, the cloud agent, or a stale local cache? What policy decided the session was allowed? Did the meter actually record what the charger reported? When the bill comes, can anyone reconstruct the sequence?

Today, in most Indian deployments, the answer is: the operator's cloud has a log, the discom has a different log, the user has an app screen, and none of them can be independently checked against the other two. This works at 29,000 chargers because nobody is auditing at scale. It does not work at 1.32 million.

Anything that doesn't achieve unsubsidized market-competitiveness within seven years of starting to scale is not worth subsidizing.

Khosla's frame applies sideways here. PM E-DRIVE will end. The 400,000-per-year build rate continues or it doesn't, depending on whether the operating economics work. Operating economics break the moment dispute volume scales with charger count. Every disputed bill that takes a human to resolve is a tax on the network. The network that wins at 1.32 million chargers is the one where disputes are settled by replaying signed evidence, not by a customer-service queue.

State variance is wider than the headlines

The national AT&C loss figure (the discom industry's headline number combining theft, billing errors, and physical losses) hides a 35x spread across states. From the PFC 2023-24 discom performance report:

DGVCL in southern Gujarat, the country's best, lost 1.31% of the energy it bought. Nagaland Power Department, the country's worst, lost 47.11% of what it bought. Same regulatory framework, same protocol stack, same hardware vendors, vastly different operating reality.

This matters for chargers because the 1.32 million build is national. A charging operator can't pick which state's grid they plug into. If their billing trail has to live with the underlying discom's audit hygiene, then every operator setting up a depot in Nagaland is inheriting a 47% loss environment by default, regardless of their own software.

A useful goal for the proof layer is to make the operator's billing trail independent of the underlying discom's audit hygiene. Signed evidence at the charger, signed evidence at the meter, hash-chained between them, and the operator can settle disputes without depending on a discom's log that may not exist.

PM E-DRIVE pays for boxes, not control

PM E-DRIVE's ₹2,000 crore for charging infrastructure covers capital subsidy for the physical EVSE plus upstream electrical work. Up to 100% of cost in some categories. That's a real subsidy and a useful one.

What it doesn't pay for is the software layer that decides what every box is allowed to do. The scheme operational guidelines treat the EVSE as a unit of capex, not as a node in a distributed control plane. There's no line item for adapter software, no line item for signing infrastructure, no line item for evidence-pack export. Those are operator costs. They scale with charger count. They are the operating expense that determines whether the network breaks at scale.

The ICEV-era analog is the toll plaza. Building a toll booth is cheap. Running a national toll booth network that can audit each transaction, settle across operators, and survive disputes is a different problem entirely. FASTag and NPCI's plumbing made that scale work. Charging needs the equivalent, and it isn't built yet.

The meter problem under the chargers

The Indian power sector has been trying to fix its measurement problem for thirty years. The AT&C loss number is the headline metric, and the trajectory has been moving the right direction, with a recent regression.

The recent improvement is real, but the PFC's 2023-24 report shows losses ticking back up to 16.12%, with accumulated discom losses now exceeding ₹6.9 lakh crore. The recovery is fragile.

The RDSS scheme, running 2021-22 through 2025-26, targets 12 to 15% national AT&C losses and a zero ACS-ARR gap. The single biggest lever inside RDSS is the prepaid smart meter rollout: 250 million meters across the country, replacing what was a quarterly-reading, often-estimated, sometimes-tampered analog stock.

The rollout is behind. As of March 2025, about 5% of the 250 million target has been installed. Of those, only 1.9% are operating in prepaid mode.

This is the system that the charging infrastructure has to ride on. Smart meters that mostly aren't there yet, on a network with ₹6.9 lakh crore of accumulated discom losses, with a measurement history of disputed readings, all about to be asked to validate a 45x growth in distributed dispatchable load.

A reasonable person reads that paragraph and updates downward on the probability that 1.32 million chargers, as currently planned, produce a settlement-clean billing system by default.

What a charge session actually looks like

This is the control flow most public-charging deployments quietly run today. The active stage rotates so you can see each step in turn.

Three parties exchange state during a charge session. The charger reports kWh delivered, the meter reports kWh imported, the cloud reports the policy that allowed the command. In a healthy deployment they roughly agree.

In a contested deployment, they diverge. The user's app shows one number. The operator's dashboard shows another. The discom's meter reading shows a third. Each system has its own log. None of them is signed. None of them can be replayed against the others.

Today this is annoying. At 1.32 million chargers, it is structurally broken.

A dispute walkthrough, with real numbers

Take a small EV depot in Pune. Six AC chargers, average session 4.2 kWh, 30 sessions per day per charger. The depot does 756 kWh per day, about 22,680 kWh per month, billed to the operator on a commercial tariff at roughly ₹9 per kWh, so about ₹204,000 per month.

Month 4, the discom bill jumps to ₹252,000. A 24% spike. The discom claims meter import was higher than what the operator's charger logs say. The operator's dashboard shows the same 22,680 kWh from charger sessions. The discom says their meter recorded 28,000 kWh. The difference is 5,320 kWh per month, about ₹47,880 of disputed revenue.

What does the operator do?

Today, in 2026, the answer is: file a dispute with the discom, wait six to twelve weeks, hire a third-party meter test if it gets contentious, and hope the meter test outcome matches their own logs. The operator's logs are not cryptographically signed, so the discom doesn't have to accept them as evidence. The discom's meter readings are not signed either, so the operator doesn't have to accept those.

This is a single depot. Multiply by 1.32 million chargers. The dispute volume at scale is not a customer service problem. It is a market-failure problem.

The proof-layer fix is structurally simple. The runtime at the depot signs every event it sees: every kWh increment on the meter, every charger command, every policy decision. The events go into a hash-chained ledger. When the discom bill arrives, the operator hands over an evidence pack covering that month: the canonical bytes of every meter event, the chain head at the start and end of the period, the signatures, the policy decisions. The discom can run the public verifier against their own meter logs and recompute. Either the operator's chain is intact and the discom's reading is wrong, or vice versa. The dispute resolves in hours, not months.

Tariff design implications

ToU (time-of-use) tariffs are spreading across Indian discoms. DERC has approved ToU in Delhi. BESCOM is piloting it in Bengaluru. The intent is to flatten the load curve by making peak hours expensive and off-peak hours cheap, so EV chargers and residential storage shift demand.

This is good policy. It also requires the runtime to know, with sub-second confidence, which tariff window each kWh fell into. A 10-minute clock drift between the charger and the meter can move a kWh from off-peak (₹6 per unit) to peak (₹12 per unit). At a million sessions a day, those edge cases add up. The dispute volume from tariff-boundary disagreements is the next AT&C loss waiting to be named.

The fix needs three things: timestamped meter reads at second-or-better resolution, signed policy bundles that include the active tariff window, and audit trails that the operator can produce on demand. None of those three things are in the current EVSE spec.

What India Stack did, and what energy still needs

DPI has the power to dramatically improve the lives of citizens and transform governance.

Nilekani's point applies to energy more than people notice. India Stack worked because UPI, Aadhaar, and the account aggregator framework agreed on what a transaction was and who signed it. Energy doesn't have that. Energy has 80 utilities, a dozen metering vendors, three protocol families (Modbus, OCPP, DLMS/COSEM), and no shared notion of a signed event.

The Government of India did announce an India Energy Stack initiative in 2024. The framing was right. The execution is at the policy-framework stage. The protocol layer, the signing layer, the verifier layer: those are the parts somebody has to build. Whoever builds them sets the long-term operating economics of the EV charging market.

What needs to exist underneath

A signed runtime sits between the protocol layer (Modbus, OCPP, DLMS, MQTT) and the policy layer. Every event that comes off a device gets canonicalized, hashed, signed with a key the runtime owns, and appended to a hash-chained ledger. Every command going the other way passes through a policy gate that produces a signed accept or reject. The ledger is the system of record. The dashboard is a view onto it.

When a dispute happens, the operator doesn't have to trust their own log. They hand over an evidence pack: the canonical bytes, the chain head, the policy decisions, the signatures. Any reviewer with the public verifier and the operator's public key can recompute the chain and either confirm or reject the claim.

The interesting property is what happens when the signature fails. The chain doesn't proceed silently. The rejection becomes evidence. A cloud agent that tries to draw 50 kW from a site capped at 40 kW does not just get a polite error. It gets a signed receipt of its own bad attempt, which sits in the ledger forever.

This is the layer JouleBridge is building. The full architecture, the canonicalization rules, the Ed25519 signing, the evidence-pack format, all of that is the JouleBridge canonical paper, which goes deeper than this market piece. The point of this piece is upstream of that: the market is going to need this layer whether JouleBridge builds it or someone else does. The 1.32 million target makes it a forcing function.

Who else is building this

The honest answer in mid-2026 is: not many. Existing energy software companies are mostly dashboards. They visualize what the operator already knew, prettily, with charts. That market is crowded.

The audit-and-proof layer is where the gap is. International EV charging networks have ad-hoc internal solutions. Indian discoms have nothing yet. The India Energy Stack announcement is good policy theater but no specification. The closest analogs are financial settlement systems: NPCI for payments, NSDL for securities, CCIL for forex. Those are exactly the right reference architectures if you want to know what mature looks like.

There's no reason the answer has to come from any one company. The bet underneath JouleBridge is that the proof layer is to energy what UPI was to payments: boring, infrastructural, and economically dominant once it exists.

Closing

The 1.32 million number is going to dominate the next four years of EV-in-India coverage. Most of that coverage will be about where the chargers go and who's funding them. Both important questions.

The question worth watching is the one underneath: when the disputes start, and they will, what evidence will the operator hand the consumer? What will the discom hand the regulator? What will the cloud agent be allowed to do without a signed receipt?

The boxes get installed. The grid follows whoever can prove what happened.

Sources

• Confederation of Indian Industry, Charging Infrastructure for Electric Vehicles report. Referenced via IBEF industry brief.
• India's EV sales cross 2.3 million in 2025, market share rises to 8%, DD News.
• PM E-DRIVE scheme official portal and operational guidelines.
• NITI Aayog handbook on EV charging infrastructure, 2023.
• PFC 2023-24 discom performance report summary, Power Line Magazine.
• National AT&C losses 2020-21 to 2021-22, PIB.
• Smart Metering in India 2025: RDSS, AMISP, AT&C losses, NES India.
• Smart Metering in India: A work in progress, Prayas Energy Group.
• Gujarat AT&C loss reduction context, DeshGujarat.
• India Energy Stack initiative, Mercom India.
• Vinod Khosla quote from his Khosla Ventures energy-investment talks 2014 onward.
• Nandan Nilekani DPI quote from public addresses 2023-2024, reported by Business Standard and YourStory.