LatticeLog

  • Governance
  • Infrastructure
  • Markets
  • Analysis
  • Notes
  • Signals
  • Cycles
  • Learnings
  • Commentaries
  • Glossary
  • Home

Written by Nithinraj Kooneri

in Bifrost Systems
The Compute Anchor — Fenrir Research
Bifrost Systems/Global South/The Compute Anchor
Fenrir Research · Bifrost Systems · Global South / Spotlight

The Compute Anchor: The Load That Builds the Grid

In the OECD, the data centre is a new load straining a finished grid. In India, it arrives before the grid is built — which makes it not a burden on the system, but the demand anchor that can pull an unbuilt grid, a domestic equipment industry and captive clean power into being.
Fenrir Research  ·  Jul 2026  ·  Yggdrasil Ledger / latticelog.in

It is the way of things that the road comes first, and the town grows where the road already runs; but here is a stranger order — a great hall raised in the empty country, that has need of a road, and of water, and of light, and so summons all three to itself where none had thought to lay them. The hall did not follow the road. The road will follow the hall.

Original epigraph, in the register of Tolkien’s hall- and road-verses
Section 01

The Load That Builds the Grid

Two pieces earlier in this series treated the data centre as a problem for the grid: the compute crunch overwhelming a mature Western system, and the colocation bypass that skips a years-long interconnection queue. Both take the grid as given — finished, congested, in the way. Turn to India and the relationship reverses. Here the data centre does not arrive to strain a completed grid. It arrives before the grid is built — and that changes it from a burden into an anchor.

An industry observation captures the inversion precisely: for the whole history of electricity, demand appeared where people and industry already were, and the network followed. Data centres invert that. They can be sited almost anywhere there is land, water, fibre and power — which means a large enough, credit-worthy, round-the-clock load can be planted in a place, and the grid, the transmission, the generation and the equipment must then be summoned to it. In a built-out economy that is a nuisance. In an economy still building its power system, it is a catalyst: a demand anchor solid enough to pull infrastructure into existence that might otherwise have waited a decade. The compute is not the interesting part. What the compute forces into being is.

The Inversion

In the OECD, the grid came first and the data centre strains it. In India, the data centre can come first — and pull the grid, the factory and the clean-power contract along behind it.

That reframes the whole asset. The value of an Indian data centre boom was never mainly the servers or the jobs. It is the anchor tenant it provides for a grid, a transmission network, a storage fleet and a domestic equipment industry that India has to build anyway — and that a bankable compute load helps finance into being.

Section 02

Compute Arrives Early, and Big

The demand is real and steep. Wood Mackenzie projects India’s operational data-centre capacity rising more than fivefold, from 2.2 GW in 2025 to 12 GW by 2030 — a ~40% compound annual growth rate — with AI-dedicated capacity expanding almost 24-fold, from 275 MW to 6,546 MW. Underneath sits a digital economy valued at ₹32 trillion (~12% of GDP), 1.03 billion internet users and 22 billion monthly UPI transactions, with a domestic AI market projected at ₹11.7 trillion by 2032. The firm’s own verdict is that India has become “a structural investment thesis” where the question is no longer whether to enter, but where and how.

Compute Capacity, More Than Fivefold in Five Years (India, GW)
India’s operational data-centre capacity, total and AI-dedicated, 2025 versus 2030 (Wood Mackenzie). Total capacity rises ~5.5× at a ~40% CAGR; AI-dedicated capacity rises ~24×. Construction costs of ~$6–7m per MW sit well below global benchmarks, sharpening the incentive to build in India. Sources: Wood Mackenzie (Jul 2026); Nomura.

Two features make this an anchor rather than merely a market. First, it is cheap to build — roughly $6–7 million per megawatt against far higher global benchmarks — so the capital keeps coming: hyperscaler commitments from AWS and Google alongside a 2.6 GW domestic pipeline from AdaniConnex, within an end-to-end value chain KPMG sizes at around $90 billion by FY35. Second, it is bankable and round-the-clock: a credit-worthy, 24/7 baseload tenant is exactly the kind of demand a lender will underwrite new generation and transmission against. That combination — large, cheap, creditworthy, and site-flexible — is what lets the load do work beyond itself.

DC Capacity by 2030
12 GW
From 2.2 GW in 2025 — ~40% CAGR
AI-Dedicated Capacity
~24×
275 MW to 6,546 MW by 2030
Value-Chain Opportunity
~$90bn
End-to-end, by FY35 (KPMG)
Build Cost
$6–7m
Per MW — well below global benchmarks
Section 03

The Bottleneck Is the Grid, Not the Compute

Here is the fact that turns the demand into a thesis: the constraint is not the servers, the capital or the land. It is the power. Wood Mackenzie is explicit that reliable, cost-competitive electricity has overtaken land and capital as the industry’s primary constraint; grid analysts add that the physical timeline for building new transmission corridors is the binding limit, one that market reform on paper cannot shortcut. Across Asia-Pacific, securing power has become harder for developers than securing land, financing or permits.

The Power Draw the Grid Must Absorb (India DC Electricity Demand, TWh)
India’s data-centre electricity demand, 2025 versus 2040 (Wood Mackenzie) — a ~20-fold rise to ~191 TWh, reaching ~7% of total power demand. India’s peak demand already hit a record ~270 GW in May 2026, and transmission-and-distribution losses run ~16.6%, more than double the OECD norm. Sources: Wood Mackenzie; Takshashila; CEA.
The Constraint, Stated Plainly

India can pour the concrete and rack the servers. What it cannot yet do is reliably deliver the power — so the grid, not the compute, decides how much of the 12 GW actually gets built.

Data-centre electricity demand is set to rise roughly twenty-fold to ~191 TWh by 2040, about 7% of the national total, onto a system already running a record ~270 GW peak and losing ~16.6% of its power in transmission and distribution — more than double the OECD norm, the same commercial-loss problem the distribution piece diagnosed. Transmission corridors, storage and grid upgrades in the Tier-2 and Tier-3 cities targeted for new builds are the gating items. The compute is ready; the grid is the reckoning.

Section 04

How the Anchor Routes Around It

Faced with a grid that cannot yet be relied upon, developers do exactly what the captive-power piece described — they self-provision — and in doing so they pull new clean capacity into being. The dominant strategy is captive generation plus long-term renewable power-purchase agreements: securing dedicated solar, wind and storage, often through open-access rules, to lock in round-the-clock supply and cost. Policy is pushing the same way; proposals would require data centres above 100 MW to build captive power outright.

Fenrir View — Captive Power, One Rung Up the Value Chain

This is the captive-power precedent in a new, cleaner guise. Where the Nigerian factory ran captive diesel because the grid failed, the Indian data centre signs a captive renewable PPA because the grid is not yet built — and because a hyperscaler’s clean-energy mandate demands it. The effect is the same architecture with the opposite emissions profile, and a far larger cheque: a single bankable compute tenant can underwrite a utility-scale solar-plus-storage build that might not otherwise have been financed. The data centre becomes the offtaker that pulls captive clean power into existence — the demand anchor doing the work the weak grid could not.

Section 05

The Multiplier: It Builds an Industry

Follow the money and the anchor’s real payoff appears: most of the spend is not the servers but the power infrastructure and equipment the build forces into being. On one estimate, equipment manufacturers command 60–75% of the total capital outlay, a structural tailwind for a domestic industry — transformers, switchgear, high-voltage transmission gear from the likes of CG Power and GE Vernova’s Indian T&D arm — that India needs to build for its whole energy transition, not just for compute. The data centre is the anchor customer that helps that industry scale.

What the anchor pulls into beingWhy the compute forces itWhere it links in the series
New generation & captive renewablesA 24/7 bankable load underwrites solar-plus-storage PPAsCaptive power; the demand multiplier
Transmission corridorsPower must reach specific grid nodes — the binding timelineGrid modernisation; distribution loss
Storage & grid firmingRound-the-clock demand needs firming on a renewable gridThe clean-firm build-out
Domestic power equipment60–75% of the spend — transformers, switchgear, T&DCement, steel & the hard-to-abate build
Cooling & water systemsRising rack density forces closed-loop, zero-liquid-dischargeCooling & thermal management; water access

The anchor also reorders the map. Maharashtra and Tamil Nadu hold roughly 65% of installed IT load today, but the next wave is following power to Andhra Pradesh, Telangana, Uttar Pradesh and Karnataka — states with more liberal open-access rules and competitive transmission charges. In other words, siting now follows where clean, cheap, evacuable power can be secured, which means the compute anchor is actively steering where India’s next tranche of generation and transmission gets built. Water is the second-order siting filter — the underappreciated risk Wood Mackenzie flags — pushing developers toward closed-loop cooling and zero-liquid-discharge ahead of regulation.

Connects to: The Power-Compute Nexus (the OECD framing — compute as strain on a finished grid) · Colocation & the Bypass Economy (siting at the power source) · Captive Power (self-provisioning, one rung up the value chain) · Losses Before Capacity (the grid the anchor must contend with) · Cooling & Thermal Management (the water-and-heat siting filter).
Section 06

Positioning: Own What the Anchor Forces Into Being

The received sceptical take on Indian data centres is that they are resource guzzlers with limited employment — power- and water-hungry sheds that create few jobs. That critique misreads the asset, because it prices the data centre as an end in itself. The value was never the jobs inside the shed; it is the grid, the equipment industry and the clean-power capacity the shed pulls into being around it.

The Positioning Rule

Don’t only own the data centre — own what it forces into being: the generation, transmission, storage and equipment the anchor finances, and the states that win the siting race by supplying the power.

Three places to stand. First, the power-infrastructure supply chain: transmission and grid equipment, transformers and switchgear, storage and firming — the 60–75% of the spend that is the domestic industry the anchor scales. Second, captive and renewable generation: the solar-plus-storage and PPA structures a bankable compute tenant underwrites, cleaner and larger than the diesel it displaces one country over. Third, the siting-and-enabling layer: the states, open-access regimes and evacuation corridors that win the builds by supplying reliable power, plus the closed-loop cooling and water systems that clear the second-order constraint. Own the anchor’s wake, not just the anchor.

Section 07

Reading It Through the Frameworks

Where the conclusion inverts. The compute framework is the same on both sides — a large new electrical load meets the grid — but the state of the grid flips the meaning. In the OECD the grid is finished, so the load is a strain and the story is the queue and the bypass. In India the grid is unbuilt, so the same load is an anchor, and the story is what it pulls into existence: generation, transmission, storage and a domestic equipment industry. Same asset; a burden where the grid exists, a catalyst where it does not.

Structural moat or temporary bottleneck? The bottleneck — power and transmission — is real and binding, which is precisely why the opportunity is structural: the compute demand is bankable enough to help finance the multi-decade grid and equipment build that resolves it. The discipline is to separate the exposure that captures the anchor’s wake (grid, transmission, storage, domestic equipment, captive renewables, the winning siting states) from the narrow data-centre real-estate play that the “guzzler” critique correctly finds thin, and to read the grid timeline — not the compute pipeline — as the true governor of how much of the 12 GW actually gets built.

Grid & Transmission Equipment
The anchor’s biggest wake
Transformers, switchgear and HVDC — 60–75% of the spend, a domestic industry the compute load scales.
Captive Renewables & PPAs
Underwritten by the tenant
Solar-plus-storage a bankable 24/7 compute load can finance — captive power, cleaner and at a far larger scale.
Storage & Grid Firming
Round-the-clock demand
Firming a renewable grid for a baseload tenant — the storage build the anchor makes financeable.
Winning Siting States
Power wins the build
Andhra Pradesh, Telangana and others drawing the next wave with open access and evacuable power — siting follows electrons.
Closed-Loop Cooling & Water
The second constraint
Zero-liquid-discharge and closed-loop systems clearing the water siting filter ahead of regulation.
Narrow Data-Centre Real Estate
The “guzzler” read
Owning only the shed — power- and water-hungry, thin on jobs; the play the sceptics correctly find thin.
Why It Is an Anchor
The data centre arrives before the grid, so it pulls infrastructure to it
A bankable 24/7 load underwrites generation, transmission and equipment
60–75% of the spend is the domestic power industry it helps scale
Siting now follows power, steering where the grid gets built
Why the Grid Governs It
Power, not compute, capital or land, is the binding constraint
Transmission-corridor timelines cap how much of the 12 GW is built
~16.6% T&D losses and a record ~270 GW peak strain the system further
Own the anchor’s wake; the shed alone is the thin, “guzzler” play
Bottom Line

In the OECD the data centre is a new load straining a finished grid — the queue and the bypass are the story. In India it arrives before the grid is built, and that flips it from a burden into an anchor: a large, cheap-to-build, bankable, round-the-clock tenant that can be planted almost anywhere and then summon generation, transmission, storage and a domestic equipment industry to it. Capacity is set to rise more than fivefold to 12 GW by 2030, but the binding constraint is not the compute — it is the power, on a grid still losing a sixth of its electricity and straining at a record peak. The grid, not the server, governs how much gets built.

So the value was never the shed. The “resource guzzler with few jobs” critique misreads the asset by pricing the data centre as an end in itself; its real payoff is the wake — the grid, the transmission, the storage, the captive clean power and the domestic equipment industry the anchor helps finance into being, most of which India must build regardless. Own that wake: the power-equipment supply chain, the captive renewables a bankable tenant underwrites, and the states that win by supplying the electrons. The hall did not follow the road; the road will follow the hall.

Plant the mill where the river is not, and men will call you a fool; but if the mill must grind, they will dig the channel to it, and the water will come where the mill has called it. So is a demand that cannot be moved: it does not wait upon the road — it builds the road, or it does not turn at all.

Original epigraph, in the register of Tolkien’s hall- and road-verses
Bifrost Systems · Global South Thread
← Related
The Demand Multiplier
The demography underneath the compute demand
Related →
Captive Power
Self-provisioning around an unreliable grid
Sources & Notes
Capacity & demand: Wood Mackenzie, India data centre capacity to reach 12 GW by 2030 (Jul 2026) — operational capacity rising from 2.2 GW (2025) to 12 GW (2030) at ~40% CAGR; AI-dedicated capacity from 275 MW to 6,546 MW (~24×); data-centre electricity demand from 10 TWh (2025) to ~191 TWh (2040), a ~20-fold rise to ~7% of total power demand; digital economy at ₹32 trillion (~12% of GDP), 1.03 billion internet users, ~22 billion monthly UPI transactions, domestic AI market ~₹11.7 trillion by 2032; reliable, cost-competitive power having overtaken land and capital as the primary constraint; captive generation and long-term renewable PPAs as the dominant procurement strategy; Maharashtra and Tamil Nadu holding ~65% of installed IT load with the next wave shifting to Andhra Pradesh, Telangana, Uttar Pradesh and Karnataka; water/cooling (closed-loop, zero-liquid-discharge) flagged as an underappreciated risk; AWS, Google and AdaniConnex (2.6 GW pipeline) commitments. Cost & value chain: Nomura via Communications Today — construction cost of ~$6–7m/MW versus higher global benchmarks, and equipment manufacturers (e.g. CG Power, GE Vernova T&D India) commanding ~60–75% of a ~$35bn (7 GW scenario) build; KPMG — end-to-end data-centre value chain of ~$90bn by FY35. Grid constraint: Takshashila Institution, Building India’s Data Centres (2025) — T&D losses ~16.64% (FY23–24, more than double the ~6–8% global norm), slow grid upgradation in Tier-2/3 cities, limited domestic storage manufacturing, and a possible captive-power requirement for >100 MW facilities; Environment+Energy Leader — transmission-corridor build timelines as the binding constraint, record peak demand of ~270.73 GW (21 May 2026), and the draft National Electricity Policy 2026; T&D India — the “demand followed the network; data centres invert that” framing and >32 GW / 1,150+ projects planned across Asia-Pacific. This piece describes market, energy and infrastructure dynamics factually and takes no political position; forecasts vary by source (e.g. alternative estimates of ~7 GW by 2030) and by scenario, and are indicative. All framing and conclusions are Fenrir Research’s own.
This analysis is for informational purposes only. Not investment advice. Country, company and sector references describe market structure and are illustrative, not recommendations. Fenrir Research is a division of Yggdrasil Ledger (latticelog.in).
←The Offtaker Problem
Connection is Not Supply→

Comments

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

More posts

  • Weather Event Dashboard

    August 20, 2026
  • ENSO – The Shifting Ruler – Baseline & the Two Indices

    August 14, 2026
  • ENSO – July update – Strengthening in Force

    August 11, 2026
  • AV: The Cost of Autonomy

    August 9, 2026

LatticeLog

Structural research across markets, infrastructure, climate, and the systems that connect them. Published under Fenrir Research, a division of Yggdrasil Ledger.

  • Blog
  • About
  • FAQs
  • Authors

Twenty Twenty-Five

Designed with WordPress