What It Will Take for India to Reach 100 GW of Nuclear by 2047

It is Independence Day, and the number everyone is repeating is 100 GW.

One hundred gigawatts of nuclear power by 2047, the centenary year. It is a good number. It is also about eleven and a half times what we actually have. The installed fleet as of this month is 8.78 GW across 24 reactors, which works out to roughly 3% of India's electricity, against coal's 70%.

So the question I keep coming back to is not whether the target is inspiring. It is: what is the actual engineering path from 8.78 to 100, and where does it break?

I want to work through that properly. Not the policy-brief version, but the one with fuel cycles, capacity factors, construction schedules and supply chains in it. And I want to spend real time on fusion, because the 2047 conversation here is almost entirely a fission conversation, and I think that is the single largest strategic error in the plan.

Let me say the encouraging thing first, though, because it is true and it is recent. On 6 April 2026 at 8:25 PM, the Prototype Fast Breeder Reactor at Kalpakkam went critical. After decades of a programme that a lot of serious people had written off, India formally entered Stage 2 of Bhabha's three-stage plan. That happened four months ago. It is the strongest evidence in fifty years that this country can finish extremely hard nuclear engineering.

It also took 23 years from sanction. Both of those facts matter, and I am going to keep both of them in view.

Because here is the thing that makes this urgent rather than merely ambitious. Twenty-one years sounds like a long time. It is not. A nuclear project sanctioned today, built on our current record, connects to the grid somewhere around 2038. That leaves room for roughly one and a half build cycles between now and 2047. Not three, not five. One and a half. Every reactor contributing to that number has to be substantially designed and sited within about the next five years, or it is not a 2047 machine at all.

The competition is not waiting for us to work that out. China connects reactors in about 60 months and is building roughly thirty at once. American fusion startups are applying for grid interconnections. Our timelines were set for a world where India was building for itself, on its own clock, with nobody else's schedule to worry about. That world is gone.

The arithmetic of the gap

Start with what is actually sanctioned, funded, and pouring concrete.

The pipeline gets to roughly 54 GW. The remaining 46 GW has no committed builder, no sited project, and in several cases no licensed design. That is not really a criticism of the roadmap so much as a description of what the roadmap is: half a plan, half an invitation.

TERI puts the total capital requirement at ₹23 to 25 lakh crore, call it $280 to $300 billion over two decades. For scale, that is roughly the entire annual central government capital expenditure, sustained, for a decade, on one sector.

And it has to be built by an industry that currently completes reactors in about twice its planned schedule.

Why construction, not physics, is the binding constraint

Look at the PHWR-700 programme, which is India's best indigenous product and the workhorse of that 54 GW.

Kakrapar 3 poured first concrete in November 2010 with a planned 66-month build. It achieved criticality in July 2020 and grid connection in January 2021. Ten years against five and a half. Kakrapar 4 followed in February 2024, Rajasthan 7 connected in March 2025, Rajasthan 8 is expected this year.

Here is the part that should reframe how you read those numbers. India's sixteen indigenous PHWRs have averaged roughly 80% capacity factor over the last five years. Once these machines are running, they run well. The operating organisation is genuinely good. NPCIL is not bad at nuclear power, it is bad at nuclear construction schedules, and that is a different disease with different medicine.

The medicine that has been prescribed is fleet mode: ten 700 MWe units sanctioned as a block at Kaiga, Gorakhpur, Chutka and Mahi Banswara. Standardised design, bulk-ordered long-lead forgings, one supply chain amortised across ten units instead of re-tendered for each. This is the correct instinct. It is roughly how South Korea got APR-1400 build times down, and how China now finishes units in about 60 months.

Fleet mode only works if the design is genuinely frozen, though. Every design change after first concrete propagates through ten units of procurement, and the schedule discipline evaporates. Our institutional habit of continuous design improvement is admirable in a research organisation and fatal in a fleet build.

The fuel problem nobody wants to lead with

This is the number that worries me most, and it surfaced in a Parliamentary panel report just last week.

India's future nuclear fleet will need about 5,400 tonnes of uranium oxide per year. Domestic production covers roughly 30% of that.

The reason is geology, not effort. Indian ore at Jaduguda and Tummalapalle runs about 0.03 to 0.1% U, against 1% or more, sometimes 15% or more, at Canadian and Kazakh deposits. You are moving thirty times the rock for the same fuel. No amount of mining policy fixes an order-of-magnitude grade deficit.

So the balance comes from Kazakhstan, Russia, Uzbekistan and Canada. The Uzbekistan long-term contract runs to 2026, which is this year. Building 100 GW of predominantly natural-uranium PHWRs means locking in imported fuel for a fleet that is the backbone of the grid, from suppliers subject to sanctions regimes we do not control.

There are three honest answers to this, and I think India needs all three:

Enrichment. The BSMR-200 uses slightly enriched uranium, which improves burnup and fuel utilisation per tonne of ore substantially. SEU-fuelled PHWRs are a real efficiency multiplier on a constrained supply.
The breeder cycle. This is what PFBR is for. A fast breeder producing more fissile material than it consumes turns a uranium-constrained programme into a plutonium-and-thorium-driven one. Stage 2 is not a prestige project, it is the fuel security answer.
Thorium. India sits on one of the world's great thorium endowments in the monazite sands running down the Kerala coast from Chavara to Kasaragod. PFBR is designed to carry a Th-232 blanket, transmuting it to U-233 for Stage 3. The AHWR-300 design exists at BARC and the thorium fuel fabrication technology exists.

The three-stage programme has been mocked for sixty years as a plan perpetually one stage away from delivering. In April, Stage 2 stopped being theoretical. The thorium logic that used to look like national romanticism now looks like the correct answer to a fuel constraint that will otherwise cap 2047 well short of 100.

SMRs: right instinct, wrong clock

The Nuclear Energy Mission put ₹20,000 crore behind at least five indigenous SMRs operational by 2033.

The flagship is BSMR-200, a 200 MWe pressurised heavy water reactor from BARC and NPCIL running slightly enriched uranium with passive safety, at a programme cost of ₹5,960 crore, with lead units sited at Tarapur and a scale-up path to 300 MWe. Alongside it sits the Bharat Small Reactor, a 220 MWe PHWR derivative aimed at captive industrial loads and retiring coal sites, where the grid connection, cooling water and switchyard already exist. Dropping a BSR onto a decommissioned coal site is honestly one of the highest-leverage moves available anywhere in Indian energy.

Now the clock. BSMR-200 construction is estimated at 60 to 72 months after administrative and financial sanction. That is five to six years for a small modular reactor, which is roughly what a large PHWR is supposed to take, and it defeats the entire premise. The economic argument for SMRs is factory serial production: build the reactor in a shop, ship modules, assemble on site in 24 to 36 months, and let unit 20 cost 40% of unit 1 through learning-curve effects.

An SMR built like a bespoke large reactor is just an expensive small reactor. India does not yet have the module fabrication yards, the serial licensing pathway, or the order book to make the learning curve real. Committing to a hundred-unit order book, even conditionally, is what would let a manufacturer justify building the yard. Five demonstration units by 2033 will not.

The law changed, and that is bigger than it sounds

In December 2025 Parliament passed the SHANTI Act, for Sustainable Harnessing and Advancement of Nuclear Energy for Transforming India. It repeals both the Atomic Energy Act of 1962 and the Civil Liability for Nuclear Damage Act of 2010, replacing them with a single statute.

Two things in it genuinely matter.

Private and foreign participation is now permitted across generation, fuel fabrication up to notified enrichment thresholds, component manufacturing, and reactor operations including SMRs. Reliance, Tata Power, Adani Power, JSW Energy, Hindalco and Jindal Steel & Power have all signalled interest.

The liability regime is capped. CLND 2010's supplier-liability provision, Section 17(b), the clause that let operators recover from suppliers and that effectively froze foreign vendor participation for fifteen years, is gone. SHANTI replaces it with a five-tier cap indexed to reactor thermal power, topping out at 300 million SDR, about $430 million, per incident.

You can argue about whether $430M is adequate for a severe release. The Bulletin of the Atomic Scientists has, at length, and the concern is not frivolous. But the practical effect of the 2010 regime was that essentially no foreign reactor got built here for fifteen years. A capped, insurable, quantified liability is what makes a project financeable.

SHANTI also grants statutory independence to the AERB, lifting the regulator out from under the Atomic Energy Commission. Until now we had the awkward arrangement of the regulator reporting into the same structure as the operator. Statutory independence is not the same as functional independence, though. AERB will need its own budget line, its own recruitment authority outside the DAE cadre, and enough licensing engineers to review a dozen novel designs in parallel. The regulator is now on the critical path of the 100 GW target, and it is currently staffed for a programme a tenth the size.

The capital problem, stated plainly

Opening the sector legally and funding it are different things, and this is where I think the gap is widest.

India's entire deeptech ecosystem raised about $1.65 billion in 2025. American deeptech raised roughly $147 billion in the same year. That is more than 80x. China did about $81 billion. Indian climate-tech VC actually fell, from $1.17B in 2024 to roughly $657 million in 2025.

Now hold those next to the fusion numbers from earlier. Commonwealth Fusion Systems, one company, has raised about $4 billion. That is more than twice what all of Indian deeptech raised last year, put together. Helion's single June round of $465M is around 70% of India's entire climate-tech venture funding for 2025. The ₹20,000 crore Nuclear Energy Mission, which is a genuinely serious commitment by Indian standards, works out to roughly $2.3B spread across five or more SMRs through 2033. One American fusion startup raised nearly twice that for one machine.

The underlying number behind all of this is R&D intensity. India spends about 0.64% of GDP on R&D. China spends about 2.4%. In 2023, roughly 5% of Indian startup funding went into deeptech, against 35% in China. We are not short of engineers or ideas. We are short of the patient capital that lets an engineer spend eight years on a machine that might not work.

This is the part private participation actually solves, and it is why SHANTI matters more than it is being given credit for. Reliance, Tata Power, Adani Power, JSW, Hindalco and Jindal have balance sheets, project management depth, and something NPCIL structurally lacks: a commercial reason to finish early. A public sector unit that finishes a reactor two years late absorbs a schedule slip. A private developer that finishes two years late absorbs two years of debt service on an idle asset. That asymmetry is worth more to the 2047 target than any amount of exhortation.

Three things would turn the legal opening into actual capital:

Underwrite the offtake. Nobody finances a twenty-year asset against a merchant power price. What made ARC financeable was Google and Eni signing offtake agreements before a single component was installed. India has exactly the same instrument available and is barely using it: long-term power purchase agreements from the industrial and data centre load that is coming online anyway. An Indian data centre operator signing a fifteen-year PPA with an SMR developer does more for this target than another mission statement.

Fund the risky middle. The gap that kills Indian deeptech is not seed money and it is not the eventual project finance. It is the decade in between, where you have a design and no revenue. That is precisely what the DOE milestone programme covers in the US, and what the India Deep Tech Alliance's $1B commitment is groping toward. A sovereign co-investment vehicle that matches private rupees into licensed nuclear and fusion ventures, paying on milestones rather than on effort, would change the risk calculus for every Indian VC currently sitting this out.

Let Indian suppliers sell to the world now. L&T built the ITER cryostat. That capability should be earning revenue from every fusion programme on earth this decade, not waiting for a domestic reactor in 2060. Export orders fund the learning curve that a domestic programme later inherits for free. This is exactly what China did with CRAFT, except China built the demand at home first because it could afford to.

Now the part India is not doing

Everything above is fission, and fission alone probably cannot get to 100 GW by 2047. The pipeline reaches 54, SMRs are on a slow clock, and the fuel is imported. But there is a second reason to care about fusion, which is that it is where the rest of the world is putting its most aggressive capital and its best engineers, and the industrial base it needs is being built right now, this decade, without us in it.

Here is where everyone actually stands.

What the American startups did

Roughly $10 billion of private capital has gone into fusion globally, and the structure of it is as interesting as the amount.

Commonwealth Fusion Systems leads, with about $4 billion raised, roughly a third of all private fusion capital, after another $1B round in July 2026. Their bet was narrow and correct. The fusion triple product scales roughly as β·τ·B⁴, so performance goes as the fourth power of magnetic field strength. ITER's low-temperature niobium-tin magnets top out near 12 T, which is precisely why ITER has to be 16.4 m across and cost $20B and up. CFS bet on REBCO high-temperature superconducting tape to reach around 20 T, and with a 4th-power scaling law that field increase buys an enormous reduction in machine size. SPARC is under 5 m across.

I wrote about SPARC outpacing ITER exactly a year ago, and the thing I got most right was that the magnet bet would compound. Status as of this month: SPARC is roughly 75% complete. First HTS magnet installed in January 2026, all 18 toroidal field magnets due in by the end of this summer, first plasma expected later this year and Q>1 in 2027, with peer-reviewed predictions in the Journal of Plasma Physics saying they clear it with margin.

And they are already building the commercial machine. ARC is a roughly 400 MWe plant in Chesterfield County, Virginia, the Fall Line Fusion Power Station, targeting grid power in the early 2030s. They hold the world's first Conditional Use Permit for a commercial fusion plant, have signed offtake with Google and Eni, and became the first fusion company to enter the PJM interconnection queue. That last one is the tell. Applying for a grid interconnection is not a physics milestone, it is what a company does when it believes it is selling electrons.

Helion took a completely different path: field-reversed configuration, pulsed, with direct electrical recovery. The expanding plasma pushes back against the magnetic field and induces current in the coils, so you skip the steam cycle and the Carnot penalty entirely. In February 2026 their Polaris machine became the first privately developed device to demonstrate measurable D-T fusion, at 150 million °C. They raised $465M in June 2026 at a $15.5B valuation, and hold a Microsoft PPA for 50 MW.

Then TAE (over $1.3B, beam-driven FRC, chasing aneutronic p-B11), Pacific Fusion (a $900M Series A, among the largest first rounds in venture history, on pulsed magnetic inertial confinement), and Zap Energy (sheared-flow-stabilised Z-pinch, no external magnets at all). I went through why these confinement approaches diverge so sharply in more detail last year, and the short version is that nobody has to be wrong for this to work. Several of these bets can pay off independently.

The structural lesson here is not that private capital is magic. It is the DOE Milestone-Based Fusion Development Program: the government pays on demonstrated technical milestones, not cost-plus contracts or annual institutional grants. Money arrives when a machine does a thing. That single design choice is why eight American companies are racing rather than eight American labs publishing.

What China did

China did not run a startup race. It built a factory town for fusion.

CRAFT, the Comprehensive Research Facility for Fusion Technology at Hefei, is a 40-hectare, 20-facility, roughly $570 million campus hosting EAST, BEST and the planned CFETR, and it just finished final testing on the world's largest superconducting fusion magnet, a 400-tonne class component.

The machines are delivering. EAST held plasma at 100 million °C for 1,066 seconds in January 2025, a long-pulse record that matters more than peak-performance records, because steady-state operation is the actual engineering problem for a power plant. BEST finishes assembly in 2027 aiming at a genuine burning plasma. CFETR completed engineering design in 2020, begins full construction late this decade, and targets first electricity to the grid by 2035.

Notice how different that is in kind from the American approach. CRAFT is not primarily a physics facility. It is a facility for building the things fusion plants are made of: magnets, tungsten plasma-facing components, cryogenic plant, divertor cassettes, tritium handling, all co-located with the machines that use them. China is treating fusion as an industrial supply chain problem that happens to have a physics component. Given where the real bottlenecks turn out to be, that may be the more sophisticated read.

Where India stands

We are not absent from fusion. We are misallocated within it.

The Institute for Plasma Research in Gandhinagar runs ADITYA-U and SST-1, India's first superconducting tokamak. SST-1 has achieved plasma of around 650 milliseconds against a design goal of up to 1,000 seconds. Set that next to EAST's 1,066 seconds achieved and the gap is about three orders of magnitude in the single parameter that matters most for a power plant.

India is also one of seven ITER members, contributing about 9% in kind, and the contributions are genuinely world-class. The ITER cryostat, the largest stainless steel vacuum vessel ever built, was fabricated by L&T at Hazira. We also deliver in-wall shielding, cryolines, cooling water systems, ion cyclotron and electron cyclotron RF sources, and the diagnostic neutral beam.

Sit with that for a second. Indian industry can build the single largest precision vacuum structure in the history of fusion. The manufacturing capability is not hypothetical. It is delivered, installed, and operating in France.

And then the roadmap. IPR's 2025 plan proposes SST-Bharat, a superconducting tokamak leading to a demonstration reactor by 2060.

Twenty-five years behind China. Thirty behind CFS. On a technology where we already manufacture tier-one hardware for someone else's reactor.

That is the gap I would spend Independence Day thinking about. Not because 2060 is a stupid date, it is a perfectly reasonable date for a conservative, publicly-funded, ITER-derivative programme. But because it is a date set by the funding model, not by the physics or the industrial capability. We priced our fusion ambition off institutional grant cycles while everyone else repriced theirs off a materials breakthrough.

The bottlenecks that decide fusion, and our strange advantage in two of them

If fusion gets industrialised, three supply chains gate it. India's position in each is instructive.

REBCO tape. Global fusion-grade production is roughly 1,500 km per year. A single CFS reactor needs over 5,400 km. The entire world builds about a quarter of one machine's worth of tape annually. Making it means layer-by-layer deposition of exotic oxides onto a moving metal substrate with defect tolerances in microns, across kilometre lengths, with no dropouts, because one bad centimetre can quench a magnet. China holds substantial manufacturing strength here, which makes it a geopolitical chokepoint as much as an industrial one. CFS understood this well enough to spin its magnet technology into a separate business line, signing a supply partnership with Realta Fusion in April 2026 explicitly to create enough aggregate demand to justify scaling the tape supply chain.

That is an open door. HTS tape production is a thin-film manufacturing problem: precision deposition, in-line metrology, yield engineering at scale. It is closer to semiconductor and display manufacturing than to reactor engineering, and it has to grow by more than an order of magnitude within a decade. There is no incumbent to displace, because the incumbent barely exists.

Lithium-6. Every D-T reactor breeds its own tritium from lithium in the blanket, and you need enriched Li-6 to do it. The historical route was the mercury-based COLEX process, which is environmentally indefensible and effectively unavailable now. A 2026 arXiv paper put it bluntly: lithium enrichment threatens to curb fusion deployment. And Li-6 is dual-use, it is a thermonuclear weapons material, so scaling it carries non-proliferation weight no purely commercial actor can resolve alone.

Tritium. A D-T plant burns roughly 55.6 kg of tritium per gigawatt-year. The entire world civil inventory is on the order of 25 kg, and it decays at 5.5% a year. Every plant has to breed more than it burns, with a tritium breeding ratio above about 1.05 to cover losses and the startup inventory for the next plant.

Here is the strange thing. The world's civil tritium comes overwhelmingly from heavy-water reactors, because neutron capture on deuterium produces it as a byproduct. India operates the world's second-largest fleet of pressurised heavy water reactors.

Sitting inside our existing fission programme, largely unremarked, is one of the planet's few structural tritium production bases. Plus the heavy water separation expertise, the lithium chemistry from the same industrial base, and decades of tritium handling experience from PHWR operation. India is one of a small handful of countries that could credibly supply the startup inventory the global fusion industry is going to need in the 2030s.

Our realistic entry into fusion is not building the fastest tokamak. It is becoming indispensable to everyone who is. Tritium and lithium from the PHWR industrial base, precision fabrication from the Hazira lineage, and a serious national push into HTS tape manufacturing. Sell into the world's fusion build-out, capture the learning, and let SST-Bharat pull forward from 2060 because the supply chain around it got built first.

So what actually has to happen

Pulling it together, here is what the 100 GW number requires that we are not currently doing. I have tried to order these by how soon the decision has to be made rather than by how much they matter, because on a 21-year clock those are almost the same thing.

Freeze the design and defend the schedule. Fleet mode is the right answer and it is already sanctioned. It fails the moment the tenth unit's drawings differ from the first. Ten identical PHWR-700s built on schedule would do more for 2047 than any new announcement.

Solve fuel or cap the ambition. 5,400 tonnes a year at 30% domestic coverage is not a footnote, it is the ceiling. That means SEU fuelling to stretch every tonne, a breeder programme treated as fuel-security infrastructure rather than a research prestige item, and thorium taken seriously now that Stage 2 is real.

Give SMRs an order book, not a demonstration. Five units by 2033 does not build a factory. A conditional hundred-unit commitment does. And coal-site brownfield deployment for the Bharat Small Reactor is the cheapest 20 GW available anywhere in the plan.

Pay on milestones. The highest-leverage institutional import from the US is not private capital, it is the DOE milestone-based structure. Fund Indian nuclear and fusion companies on demonstrated technical achievement. It changes who applies, how fast they move, and what failure costs.

Get private balance sheets onto sites, not just into press releases. SHANTI made private participation legal in December. Eight months on, the interested corporates are still mostly interested. What converts interest into concrete is a licensed standard design, a sited plot, and an offtake contract. Whoever hands a private developer those three things first will find out that Indian industry builds faster than Indian government does, which is the entire bet SHANTI was placing.

Build the patient capital. An 80x deeptech funding gap is not closed by exhortation. It closes when there is a sovereign co-investment vehicle matching private rupees on milestone terms, when insurance products exist for construction risk on a capped-liability reactor, and when a fund manager can point to one Indian nuclear exit. Somebody has to be first, and the state is the only actor that can make first cheap enough for anyone else to follow.

Staff the regulator like it is on the critical path, because it is. Statutory independence for AERB is done. Functional independence, meaning its own budget, its own cadre, enough licensing engineers to review a dozen novel designs concurrently, is not. Every month of licensing queue is a month off a 21-year schedule.

Move the fusion date. 2060 is a funding decision wearing a physics costume. A milestone-funded programme anchored on the tritium and lithium advantages we already have, plus an HTS tape manufacturing push, could plausibly bring a demonstration into the 2040s. Inside the window where it contributes to 2047 rather than commemorating it. And the payoff compounds well beyond electricity: I have argued before that cheap fusion turns desalination and carbon removal into line items on a utility bill, which for a country facing the water stress we are facing is not a side benefit. It might eventually be the main one.

Why I am hopeful anyway

I want to end where I started, because I think the pessimistic reading of all this is wrong.

Four months ago, a 500 MWe sodium-cooled fast breeder that had been in development since 2003, through two decades of delays that made it a punchline, went critical at Kalpakkam. Rafael Grossi called it out from the IAEA. India became one of a very small number of countries operating a fast breeder at all, and did it with an indigenous design, indigenous sodium technology and indigenous fuel.

That machine is the proof of the thing that actually matters here. The constraint on Indian nuclear was never talent, never physics, never manufacturing capability. The ITER cryostat settles the manufacturing question permanently. The constraint was institutional: schedules that slipped without consequence, a liability law that locked out capital, a regulator sitting inside the organisation it regulated, and a funding model that paid for effort instead of results.

In the last eighteen months, three of those four changed. SHANTI repealed the liability regime and opened the sector. AERB got statutory independence. Fleet mode replaced one-off construction. The fourth, milestone-based funding, is a decision someone could make this year.

100 GW by 2047 is not a forecast. On the current trajectory it is closer to 60, and anyone being honest should say so. But it is a genuinely achievable number if fuel security gets solved through the breeder and thorium route, if SMRs get an order book instead of a pilot, if private balance sheets get sites and offtake instead of encouragement, and if fusion is treated as an industrial supply chain we can dominate rather than a science project we will eventually get around to.

What I keep coming back to is that none of those are twenty-year decisions. They are decisions about the next two or three years, which then take twenty years to show up. The reactors that will be running in 2047 are being argued about in committee rooms right now, and the ones nobody argues for in this window simply will not exist. That is the whole urgency. Not that 2047 is close, but that the window to influence 2047 is nearly shut, and it closes quietly, without anyone announcing it.

Bhabha sketched the three-stage programme in the 1950s, in a country with almost no industrial base, and was mocked for it. Stage 2 went critical in April.

He was not working with better tools than we have. He was working with a clearer sense that the decisions in front of him were the ones that mattered, and that waiting was itself a choice. The plan was never the problem. Finishing it was. And on that, at least, the evidence has recently changed.

Vande Mataram.

Sources: