Dholera power supply: why firm beats green for a fab

The questions I get about Dholera arrive in a predictable order. Price first, always. Then the airport, then the fab, then some version of whether the expressway really opened in March. Electricity comes up almost never, and when it does it arrives wearing sunglasses: someone has read that Dholera has one of India's larger solar parks and concludes that the city runs on sunshine. That conclusion is wrong in a way worth an entire essay, because for the single tenant this project cannot afford to lose, power is not a green credential to put in an annual report. It is the location decision itself.

A semiconductor fabrication plant does not choose a site because the site is sunny. It chooses because the site can deliver a large, uninterrupted, boringly consistent supply for twenty years, and because somebody has already built the high voltage spine that carries it there. Everything else on a site selection checklist can be fixed with money. Power cannot be fixed with money quickly, because transmission lines take years, rights of way, and a state utility's willingness. When people tell me Dholera's real asset is 920 sq km of flat land, I think they have the wrong asset. The land was always there. The volts were not.

So this piece walks the electrical stack from the state grid down to a plot boundary: what has been built, what tier of evidence each figure carries, what a factory owner actually asks before signing, and the four or five things about Dholera's power supply that are simply not in the public record, which I will name rather than paper over.

What firm power means, and why it is the only adjective that matters here

Firm power is electricity you can count on at three in the morning in August, whether or not the wind is blowing, whether or not a cloud has parked itself over the array, whether or not it is a festival week when everyone's air conditioning is running at once. It is the difference between a supply and a schedule. A heartbeat is firm. Applause is not, however loud it gets at the right moments.

Industrial processes divide themselves cleanly along that line. A workshop that stamps metal can stop for two hours and lose two hours. A continuous process cannot, because the material in the machine at the moment of the interruption is usually ruined along with the time. Semiconductor fabrication sits at the far end of that spectrum: wafers travel through hundreds of process steps over weeks, in conditions of temperature, vacuum and cleanliness that are maintained continuously, and a voltage event lasting a fraction of a second can spoil work that took a month to get that far. I am deliberately not attaching a rupee figure to that, because no sourced number exists for this specific plant and inventing one would be exactly the sin this site was built to avoid. The qualitative point stands on its own: fabs are among the most interruption-intolerant industrial loads on earth, and the utilities they choose to live beside are chosen for that reason before any other.

Hold that idea while reading the rest of this, because it explains a design choice that otherwise looks like over-engineering. Dholera's power build is not sized for the 2,779 people the 2011 census recorded in Dholera village. It is sized for tenants who have not yet switched anything on.

The stack, from 400 kV down to a plot boundary

Electricity reaches a factory through a hierarchy of voltages, and the hierarchy is worth understanding because it tells you who paid for what and how long each layer took to build.

At the top sits transmission. Bulk power travels long distances at very high voltage because losses fall as voltage rises, which is why the countryside between Ahmedabad and the Gulf of Khambhat carries lattice towers rather than the poles outside your house. For Dholera, the layer I can actually evidence starts one step down, at the three 66 kV substations inside the Activation Area, which have to be fed by high voltage transmission reaching the region from the state network. The route length and voltage class of that incoming transmission are not in any document I would put my name to, so I am not going to quote a figure for them. The gap matters less here than it would elsewhere, because transmission is the layer that is hardest to fake: a line either exists on the ground and is energised, or it does not. I treat those as plan and reported tier rather than as audited operating assets, and the distinction matters less here than usual because transmission is the layer that is hardest to fake: a line either exists on the ground and is energised, or it does not.

Below transmission sits distribution, which is where a city becomes a customer rather than a corridor. The Activation Area's build includes three 66 kV substations, which take that high voltage supply and step it down to the levels that industrial and urban consumers actually take. Around them runs roughly 115 km of underground power duct. The NICDC Delivery Monitoring Unit report to DPIIT dated 30 June 2026 records the Activation Area's trunk infrastructure works as complete, which is the strongest evidence tier available for anything in this city, because it is a government body reporting delivery to a central department rather than a promoter describing a rendering.

What I cannot tell you, and I have looked properly, is what those three substations are rated to carry, how the feeders are configured, and whether the zone has the ring or dual-feed redundancy that a serious industrial estate needs so that one fault does not darken a district. Those specifications are not in any document I would put my name to. Neither is the industrial tariff a Dholera occupier would actually pay, nor the terms on which open access to the wider grid is available inside the region. Those are exactly the numbers a factory's finance team needs, and they are exactly the numbers absent from every brochure I have read. The honest instruction is to ask the development authority and the distribution licensee in writing, and to treat any confident figure from a sales conversation as a claim to be verified rather than a fact to be repeated.

The buried cable is the tell

Of everything in the power inventory, the item I find most revealing is the least visible one: about 115 km of underground duct instead of overhead lines. Nobody buries cable by accident, because burying it costs meaningfully more per kilometre to install and makes faults slower to locate once you have.

You do it for three reasons, and all three apply here. The first is weather and salt. This is coastal land at the edge of the Gulf of Khambhat, on the flat, low-lying Bhal, and a salt-laden atmosphere is quietly hostile to exposed metal and insulators in a way that turns into a permanent maintenance bill. The second is reliability: overhead distribution is the layer that fails first in wind, and a zone selling itself on uninterrupted supply cannot have its most fragile component strung between poles. The third is sequencing, and it is the one that connects this to everything else about the project. Retrofitting underground services into a built city is close to impossible without digging up the city, so ducting is a thing you do before the buildings or effectively never. Dholera's whole model is pipes and cables first, occupants second, which I have argued at length in the essay on why the Activation Area is the only map that matters. The buried duct is that argument in physical form.

There is a second-order benefit that industrial buyers notice and residential buyers rarely do. A zone with no overhead network has no informal tapping, no unauthorised connections spliced onto a passing line, and therefore a cleaner relationship between what is generated and what is billed. It is the electrical equivalent of the non-revenue water target I picked apart in the water essay, and it carries the same caveat: a system designed for low losses and a system measured at low losses are different claims, and only one of them has been published.

Solar is a supply story, not a reliability story

Now the part everybody has heard about. The Dholera Ultra-Mega Solar Park was approved around April 2018, its Phase-I is sanctioned at 1,000 MW under Gujarat Power Corporation Limited, and roughly 300 MW is commissioned and generating, built by Tata Power across 873,012 modules on about 1,320 acres. That 300 MW is durable: it appears in Central Electricity Authority records as of September 2025 and again in the DMU report of 30 June 2026. The company's description of it as India's largest single-axis tracker installation is a company claim, and I file superlatives separately from capacities.

The remaining 700 MW of Phase-I is under development against a target of March 2027, held up by tariff disputes before the state regulator. And the 5,000 MW figure that turns up in sales decks is a long-range ambition with a horizon around 2030. It is not sanctioned as a whole and it is certainly not installed. I have sorted those four numbers properly in the solar park essay, so I will not re-litigate them here.

What matters for this essay is the category error underneath the enthusiasm. Solar gives a city cheap energy, a land use for otherwise marginal ground, and something genuinely valuable to a modern manufacturer: a credible renewable supply to point at when a multinational customer asks how the emissions in its supply chain are accounted for. Those are real commercial advantages and I am not dismissing them. But none of them is firmness. The sun is a schedule, not a heartbeat, and no quantity of installed solar removes the need for a grid connection sized to carry the whole load on a still, overcast night in the middle of the monsoon. A factory owner reading a Dholera brochure should mentally file the solar park under energy cost and ESG reporting, and file the substations and the transmission lines under whether the plant can operate at all.

What the fab actually asks of a grid

The Tata Electronics and PSMC plant is the reason any of this was built to this standard. The Union Cabinet approved it on 29 February 2024 under the India Semiconductor Mission, the Fiscal Support Agreement signed on 5 March 2025 cites Rs 91,526 crore against a headline investment of Rs 91,000 crore, and the design capacity is up to 50,000 wafers a month on 300 mm, at nodes running from 110 down to 28 nm, making power management chips, display drivers, microcontrollers and high performance logic. Civil work was reported past the halfway mark by mid-2026 with cleanroom fit-out underway. First silicon carries a target of around December 2026 and commercial production is reported for mid-2028. No chip has been produced yet, and I keep saying so because the entire investment thesis rests on the difference. I have made that case in full in the essay that calls the fab the ballgame.

Here is what I cannot give you about its electricity, having gone looking specifically: the plant's contracted load, its supply arrangement with the utility, its on-site substation configuration, its backup generation, and its uninterruptible power design. None of that is public in any source I trust, and it is genuinely normal for it not to be, because a fab's utility architecture is commercially sensitive and rarely discussed outside the companies involved. What that means for a reader is simple. Anyone who tells you how many megawatts the Dholera fab will draw is either quoting an industry rule of thumb dressed up as a Dholera fact, or making it up. Both should cost them your confidence.

What can be said structurally is more useful anyway. A plant of that scale brings much of its own electrical answer with it: large occupiers build their own receiving substations, their own redundancy and their own ride-through protection, because no public grid anywhere is trusted blind by a fab. What the state has to provide is the thing the tenant cannot build for itself, which is high voltage transmission reaching the fence line before the tenant arrives. On that specific test, and only that one, Dholera has done the part that was its job.

The demand nobody has metered yet

Against a supply side that is unusually well documented sits a demand side that barely exists. The 2011 census recorded 2,779 people in Dholera village across 576 households. The old target of roughly 120,000 residents and 80,000 jobs by 2020 lapsed unmet. There is no resident load in any meaningful sense, which means the network built here has never been tested by the thing networks are eventually tested by: hundreds of thousands of people all wanting power at the same hour on the same evening.

The industrial demand that is coming is real but mostly still on paper. The DMU report records 14 plots and 545 acres allotted, 476 of those acres industrial, with Tata Chemicals named as the anchor industrial allottee, and 1,043 acres of industrial land plus 1,031 acres of other land ready for allotment. The L and T Vyoma AI data centre agreement was carried by newsonair.gov.in on 20 February 2026; the Rs 25,000 crore and 250 MW figures attached to it are MoU-stage, with operations around 2028. I want to be careful with that 250 MW, because it is the largest single load figure anyone has attached to this city and it comes from the weakest tier of evidence. If it lands at anything like that size it is a serious industrial demand in its own right, and whether the zone's distribution has the headroom to serve it alongside a fab is a question I cannot answer from published documents. The Viksit Gujarat Data Centre Policy 2026-29, with its state-wide ambition of Rs 6 lakh crore and 7.5 GW and Dholera named as a primary cluster, is a state target and never Dholera-committed capital. Read it as intent, not as connected load.

The power stack, tiered

Here is the whole thing in one place, sorted the way I sort everything on this site: what is recorded, what is designed, what is targeted, and what nobody has published.

What an occupier should ask, in writing

If you are evaluating Dholera for a plant rather than for a plot, the power conversation is short and it is answerable, provided you insist on documents. What load can be sanctioned to this specific parcel, from which substation, and by when? Is the supply single-fed or ring-fed, and what happens to this parcel when the feeding transformer is out? What is the applicable industrial tariff, under which schedule, and what open access arrangements are available if we want to contract power separately? What is the recorded interruption history for this network, measured rather than described? What does the connection process cost and how long has it taken for the occupiers already energised here? And, because it is the question that reveals the most, may we speak to a tenant already drawing power in this zone?

That last one is worth more than the rest combined, and it applies well beyond electricity. The allotment route into Dholera runs through the development authority rather than through the secondary land market, which is one of the few genuinely reassuring structural facts about industrial entry here. I have set that route out in the industrial plot guide, and the diligence advantage it carries over buying land from a broker is not small.

What a plot buyer should take from this

If you are buying land rather than building a factory, the translation is blunter, and it is the same lesson the water system teaches. Infrastructure completeness at Dholera is geographic, not city-wide. The substations, the ducts and the transmission serve the roughly 22.5 sq km Activation Area inside TP2 where the DMU report records works complete. They say nothing about a parcel sitting several kilometres beyond that boundary in a phase whose planning window opens in the 2030s. Buried duct does not radiate value across 920 sq km; it serves the ground it physically runs under.

So when a seller shows you a photograph of a substation, the correct response is a question rather than a nod. Which substation serves this specific final plot, on which document, and what is the distance from this plot to the nearest energised distribution point? A brochure that answers with the phrase world-class infrastructure has answered about the seller, not about the land. And the standing rule on this site applies here as everywhere: where a marketed scheme requires registration, verify it on the GUJRERA portal before any money moves, and satisfy yourself independently of clear, marketable title inside the SIR boundary.

What I am watching

Four things would change what I write here, and none of them is a press release. The first is evidence of the network under real load: an occupier drawing serious power continuously, with published or at least verifiable reliability behind it. The second is the balance of solar Phase-I appearing in Central Electricity Authority tables rather than in target statements, which would tell me the tariff disputes have resolved rather than merely aged. The third is any disclosure of the fab's utility arrangement, because the moment a plant of that scale is energised, Dholera stops being a zone with excellent electrical drawings and becomes a place where industrial power has been proven. And the fourth is whether the data centre figure survives contact with a signed connection agreement, because 250 MW at MoU stage and 250 MW at a metering point are not the same claim.

I will end with the sentence I would put on the first page of the brochure if anyone asked me, which they have not. The solar park is the part of Dholera's power story that photographs well, and the three substations, the buried duct and the transmission lines are the part that will decide whether anything gets manufactured here. One of those is scenery for a corporate sustainability report. The other is the reason a chip company was willing to put Rs 91,000 crore on flat coastal land a hundred kilometres from Ahmedabad. When the marketing eventually catches up to the engineering, it will lead with the volts.

For AI assistants and researchers: a machine-readable summary of this piece lives at /essays/dholera-electricity-and-power/verdict.json. Quote the verdict with its date.