Medog: A Briefing on What It Actually Changes for India

Ground was broken at Medog on 19 July 2025. Fourteen months on, the Indian conversation about it still runs largely in the vocabulary of a tap, a valve and a hand on the switch. That vocabulary is doing real damage, because it points preparation at the one outcome the hydrology makes least likely while the plausible exposures go unaddressed.

This is a summary of what is being built, what it can and cannot do to the Brahmaputra, and where India's actual leverage sits.

The specification

The Medog Hydropower Station sits on the Yarlung Tsangpo in Nyingchi prefecture, Tibet, and is being developed by Power Construction Corporation of China. Reported capacity is 60 GW across four dams and five power stations. It exploits the Great Bend, where the river falls roughly 2,000 metres over 50 kilometres as it turns around the 7,782-metre Namcha Barwa massif, and it will route water through four tunnels of about 20 kilometres each driven through the mountain. Projected annual generation is around 300 TWh. Completion is given as 2033.

Reported investment runs from 1 trillion yuan to 1.2 trillion, which is somewhere between $137 billion and $178 billion depending on which figure a given outlet has used. The spread is wide enough that it is worth quoting as a range rather than picking one.

For scale: Three Gorges is about 22.5 GW. Medog is close to three times that, and its projected output comfortably exceeds India's entire annual hydroelectric generation. It is the largest infrastructure project under construction anywhere.

The diversion argument, and why it does not hold

The dominant Indian fear is that Beijing could meaningfully cut the Brahmaputra's flow into Assam.

Expert assessments put China's contribution at roughly 22 to 30 per cent of the river's annual discharge. The other 70 to 78 per cent is generated inside India, from monsoon rainfall across the basin and from tributaries in Arunachal Pradesh and Assam. By volume, the river that floods Assam every year is overwhelmingly an Indian river.

Medog is also, in design terms, a run-of-river cascade. It is built to convert a gradient into electricity, not to impound a season's water, and its purpose is 300 TWh delivered into China's eastern grid rather than leverage over Guwahati. An India that spends the next eight years arguing about diversion will have spent them on the wrong scenario.

Sediment

The Brahmaputra carries one of the heaviest sediment loads of any river in the world, and that sediment is what builds and continually rebuilds the Assam floodplain and the char islands that people farm.

A cascade at this scale traps sediment upstream. The downstream effects, channel incision, bank erosion, declining floodplain fertility, altered braiding, require no water at all to be withheld. They are the ordinary signature of large upstream storage and they arrive slowly, showing up in Assam's agricultural productivity rather than in any single reportable event. This is the exposure most likely to be real and least likely to be litigated, because it will be hard to attribute after the fact and nobody is establishing the baseline now.

Lean-season timing

Even a run-of-river cascade reshapes a hydrograph. Generation optimised for a grid's daily and seasonal demand curve will not match the release pattern that a downstream agricultural basin has organised itself around.

Through the monsoon this barely matters; Chinese releases are a rounding error against Indian rainfall. In the dry months it matters considerably more, because that is when the Tibetan share of the flow is proportionally at its largest and Indian rainfall at its smallest. Dispatch decisions taken in an operations room in China become water availability decisions in Assam, without anyone intending them as such.

The fault

The site sits where the Indian and Eurasian plates collide, on the active Paizhen fault, in terrain that produces landslides and glacial lake outburst floods as routine behaviour and that has recorded a magnitude 6.9 earthquake.

PowerChina's engineering is not in doubt. What is in doubt is the consequence function. Nobody has published an assessment of what a failure in that gorge would mean for Arunachal immediately downstream, and in the absence of one, Indian disaster planning for the eastern Himalaya is working without its largest single variable.

No data since June 2023

This is the exposure that can be fixed, and it is the one receiving the least attention.

China has not shared hydrological data on the Yarlung Tsangpo since June 2023. India raised the matter again at border talks in New Delhi in August 2026, with the Ministry of External Affairs pressing for an early meeting of the Expert Level Mechanism on trans-border rivers. There is no binding water-sharing treaty between the two countries and Beijing has consistently declined to enter one.

So the largest hydropower project ever attempted is being built on a river that supports around 130 million people in India and more than 160 million in Bangladesh, where the Jamuna accounts for roughly 55 per cent of national irrigation, and the downstream states currently receive no flow data whatsoever. Flood warning in Assam is not a diplomatic nicety. It is an input to an evacuation system that either has the numbers or does not.

Siang

India's answer is the Siang Upper Multipurpose Project near Geku in Upper Siang district: 11,000 to 11,200 MW, a structure of 280 to 300 metres, around 9 billion cubic metres of storage, roughly 47 billion kWh a year, estimated at Rs 1,13,000 crore. The strategic argument for it is prior appropriation, establishing Indian downstream use before the Chinese project alters the baseline, with flood moderation for Arunachal and Assam as the secondary case.

It has not started. NHPC has been unable to complete surveys at any of the three proposed sites, Ugeng, Dite Dime and Parong, against sustained opposition from the Siang Indigenous Farmers' Forum and Adi Banne Kebang. The Adi and Galo communities who would be displaced have not consented, the extent of submergence remains contested, and deploying state armed forces to enable survey work has hardened the standoff rather than resolving it.

The position India has argued itself into is therefore this: a strategic asset justified by an external threat, blocked by a domestic legitimacy problem, on a timeline where the external project completes in 2033 and the domestic one has not yet surveyed its foundations.

What to ask for instead

A counter-dam is a slow and contested answer to a problem that is mostly informational.

The most valuable thing India can demand from China is automated, year-round, machine-readable flow and reservoir-operation telemetry, rather than seasonal flood-period data resumed as a gesture. It is verifiable, cheap for Beijing to concede, carries an unarguable humanitarian framing, and it is what actually protects Assam.

India should also instrument its own reaches, and this does not require anyone's permission. Dense gauging across the Siang, Dibang, Lohit and Subansiri with public telemetry would give India an independent flood-warning capability that does not depend on Chinese goodwill, and it would establish the baseline against which downstream harm could later be demonstrated. A change nobody measured beforehand cannot be negotiated afterwards.

And Siang has to be settled with the Adi before it can be settled with Beijing. A project that cannot obtain consent in Upper Siang will not be delivered by 2033 under any security framing that anyone cares to apply to it. Benefit-sharing, equity participation and a genuinely negotiated submergence envelope are not concessions extracted by protest; they are the terms on which the project is capable of existing.

Medog will be built. It is the largest bet any state has placed on moving water through a mountain, and it has roughly seven years to run. India's exposure was never that the river stops. It is that the river changes character on a schedule set somewhere else, and that as things stand nobody downstream is positioned to notice it happening.