Landlord of the AI Economy: Mongolia Bets Half Its Grid

Mongolia signed ten memoranda of understanding covering 863 megawatts of prospective data centre capacity at the seventeenth Conference of the Parties to the UN Convention to Combat Desertification in Ulaanbaatar on 25 August 2026, roughly 54 per cent of the country's entire installed generating capacity of about 1.6 gigawatts. The initiative, called "Sovereign by Design" and subtitled Mongolia's Green AI Data Centre Initiative, was opened by Prime Minister Nyam-Osor Uchral, with the Asian Infrastructure Investment Bank and other international partners presenting up to two billion dollars in financing for eligible green and renewable infrastructure projects. One memorandum, announced the following day, was signed with NOVVA Group by managing director Vincent Wen and Minister of Digital Development, Innovation and Communications Nomin Chinbat, covering energy-efficient data centres, AI-ready infrastructure, renewable energy and storage, site readiness, and policy consultation.

Landlord of the AI Economy: Mongolia Bets Half Its Grid The venue was a United Nations conference about land turning to dust. The product on offer was electricity. On 25 August 2026, inside the Host Country House at the seventeenth Conference of the Parties to the UN Convention to Combat Desertification in Ulaanbaatar, Mongolia's government convened an event it called “Sovereign by Design”, subtitled Mongolia's Green AI Data Centre Initiative. The Prime Minister, Nyam-Osor Uchral, opened it. His Deputy Prime Minister, Togmid Dorjkhand, pitched investors on the arithmetic of cheap power. His Minister of Digital Development, Innovation and Communications, Nomin Chinbat, told the room something more interesting than a price. “We offer more than just natural resources,” she said. “We offer safety and data sovereignty.” By the time the twelve-day conference wound down, Mongolia had signed ten memoranda of understanding covering 863 megawatts of prospective data centre capacity. The initiative was not a solo effort. The Asian Infrastructure Investment Bank and other international partners have joined it, with up to two billion dollars in financing presented as available for eligible green and renewable infrastructure projects. One of the ten memoranda, announced the following day, was with NOVVA Group, described in its own materials as a global clean energy infrastructure platform enabling the AI economy. It was signed by Vincent Wen, the company's managing director, and by Nomin Chinbat, and witnessed by the Prime Minister himself. Its scope covers energy-efficient data centres and AI-ready infrastructure, renewable energy and storage including solar photovoltaics, wind and battery systems, site readiness, and policy consultation. That is a lot of nouns and not many verbs. But hold the scepticism for a moment, because the easy reaction to this story is the wrong one, and the interesting reaction takes considerably more work. The Number That Should Stop You Mongolia's entire electricity system has an installed generating capacity of about 1.6 gigawatts. That figure, from the Stockholm Environment Institute's 2024 policy overview, is not controversial. So the 863 megawatts of data centre capacity Mongolia signed up at COP17 amounts to roughly 54 per cent of the entire installed capacity of the country. Not 54 per cent of spare capacity. Not 54 per cent of planned additions. Fifty-four per cent of every turbine, panel and coal-fired boiler currently connected to the Mongolian grid, committed in a fortnight, at a conference about desertification, in documents that are almost certainly not legally binding. This is the fact around which everything else in this story orbits, and it is worth sitting with. Why Mongolia Is Not Being Naive Start with the strongest possible version of Mongolia's case, because it is genuinely strong, and because the reflex to patronise small states pursuing industrial policy is one of the least attractive habits in technology commentary. Mongolia is a landlocked country of roughly 3.5 million people wedged between two nuclear-armed great powers, with essentially no third option for physical trade. Somewhere north of 90 per cent of its exports go to China, overwhelmingly coal and copper, and China supplies around 70 per cent of its imports. Its electricity grid is physically interconnected with the Russian Siberian system at 220 kilovolts, and it imported 22.3 per cent of its electricity in 2023 from China and Russia combined. When your lights depend on two neighbours who are also your only customers, “sovereignty” is not an abstraction to be deconstructed in a seminar. It is the daily condition of your existence. Mongolia's third neighbour policy, formulated in the early 1990s, has been a decades-long attempt to manufacture optionality where geography offers none, through partnerships with the United States, Japan, South Korea, India and the European Union. Digital infrastructure is the purest possible expression of that doctrine. Fibre and compute are the only exports Mongolia can ship without asking Beijing or Moscow for permission to cross a border with a train. The resource case is real too. Mongolia's combined wind and solar potential is estimated at around 2,600 gigawatts, more than a thousand times its installed capacity, and Ulaanbaatar is the coldest capital on earth, which means free cooling for much of the year and a thermodynamic advantage over Texas or Singapore. Dorjkhand's pitch was power below four cents per kilowatt hour against fourteen to sixteen cents in the United States, grid connection inside 24 months, and returns near 18 per cent. He also offered a line that doubles as a national brand: “we are a very cold country, eight months a year.” And there is a strategic niche that is smarter than it first appears. Mongolia is explicitly courting customers who want to serve the Chinese market while keeping their data physically outside Chinese jurisdiction. That is a real and growing commercial need, and Mongolia's geography makes it one of very few places that can plausibly offer it. Most importantly: everything Mongolia is doing here is exactly what development economists, multilateral lenders and Western governments have spent thirty years telling countries like Mongolia to do. Move up the value chain. Stop exporting raw rock. Attract foreign direct investment. Use your comparative advantage. Diversify away from extractive dependency. To turn round now and tell a dryland economy that it should not build compute because the electricity would be better spent elsewhere would be to pull the ladder up with unusual brazenness. The Global North built its compute freely, on coal and gas, without asking anyone's permission, and it is still building it. So the question is not whether Mongolia should want this. The question is whether what was signed at COP17 delivers it. Thirty-Two Countries Own the Machine Room Here is the structural reality Mongolia is trying to escape, and it is documented with unusual clarity in a paper posted to arXiv in March 2026 and revised in July by Amirpasha Mozaffari and colleagues at the Barcelona Supercomputing Center, with co-authors at Imperial College London and the Catalan Institution for Research and Advanced Studies. The paper is ostensibly about weather and climate information, but its infrastructure section is the most useful short description available of who owns the AI economy. Its central finding is blunt: “only 32 countries globally host the hyperscale AI data centres necessary for robust cloud-based deployment.” Thirty-two. Out of 193 UN member states. Mongolia is not among them. The authors go further, and their argument is precisely the one Mongolia's ministers are making, arrived at from the opposite direction. Frontier models, they note, are developed almost exclusively through collaborations between national public computing centres in the Global North and East Asia and a handful of multinational firms including Google DeepMind, Nvidia, Microsoft and Huawei. This concentration “effectively gatekeeps the development of foundation models, restricting it to entities with access to exascale computing and specialised engineering talent.” The consequence, in their words, is “a risk of technological dependency, where low-income nations are relegated to being consumers of models that are not generated for them as the main target by infrastructure they cannot audit, control, or adapt to their local contexts.” Without what they call compute sovereignty, meaning “the ability to train and fine-tune models on sovereign infrastructure, nations lack the agency to prioritise their specific regional climate vulnerabilities.” Note the phrase. Compute sovereignty. A team of climate scientists in Barcelona, writing about forecasting inequality, independently reached for the same word Mongolia's government put on a banner in Ulaanbaatar. That convergence is not a coincidence. It is evidence that the diagnosis is correct. The same paper supplies the resource figures that make the diagnosis uncomfortable. Data centre energy demand is projected to reach 4.5 per cent of global electricity generation by 2030. Global AI water withdrawal is projected at 4.2 to 6.6 billion cubic metres by 2027. Those are the entry costs to the club of 32. They are why the club has 32 members. What 863 Megawatts Does to a 1.6 Gigawatt Country The second paper in this story gives us a tool for testing Mongolia's plan rather than merely admiring it. Posted to arXiv in March 2026 by Danbo Chen and colleagues at Ohio State University, Zhejiang A&F University and Meta Platforms, “Concentrated siting of AI data centers drives regional power-system stress under rising global compute demand” builds what the authors call an AI-energy coupling framework, combining language-model analysis of corporate, policy and media disclosures with quantitative energy-system modelling to 2030. Its findings confirm the concentration thesis from the supply side. North America, Western Europe and Asia-Pacific together account for more than 90 per cent of projected compute capacity. Consumption by the six leading firms is projected to rise from roughly 118 terawatt hours in 2024 to between 239 and 295 by 2030, about 1 per cent of global power demand on its own. The International Energy Agency projection they cite has global data centre consumption more than doubling from about 415 terawatt hours in 2024 to roughly 945 by 2030, with the United States and China responsible for almost 80 per cent of the growth. AI infrastructure, the authors write, is “evolving from a marginal digital service into a structural component of power-system dynamics.” The useful part is their metric. The Power Stress Index is the ratio of data centre electricity demand to total regional generation capacity. Above 0.25 is their high-stress threshold, indicating grid vulnerability. Ireland, the worst case in their dataset, approaches 0.5, which the authors observe implies AI-related consumption absorbing nearly half of local generation. Oregon exceeds the threshold. Virginia, Nebraska and Washington sit between 0.20 and 0.30. Diversified systems such as Texas, Japan and Australia stay below 0.10. The authors did not run Mongolia. So let us run it, transparently, using their published formula and independently verified Mongolian data. Mongolia's net electricity generation in 2024 was 8.83 terawatt hours, up 6.6 per cent on 2023. National consumption was 10.29 terawatt hours, the gap filled by imports. Now take the 863 megawatts of signed data centre capacity. At a conservative 60 per cent utilisation, that load consumes 4.54 terawatt hours a year, a PSI of 0.51 against national generation. At a more realistic 70 per cent, it is 5.29 terawatt hours and a PSI of 0.60. At 80 per cent, which is unremarkable for AI training infrastructure, it is 6.05 terawatt hours and a PSI of 0.68. Read that against the paper's own benchmarks. At the low end of plausible utilisation, Mongolia matches Ireland, the most stressed grid in the study. At the middle and upper end, Mongolia exceeds it by a wide margin, landing somewhere no jurisdiction in the dataset occupies. Even measured against total national consumption including imports, the figures run from 0.44 to 0.59. This is the analytical heart of the matter, and it is not a rhetorical flourish. In Virginia, a hyperscale campus is a large marginal load on a very large system. In Mongolia, 863 megawatts is not marginal. It is structural. It would be, at a stroke, the defining feature of the national power system, larger in energy terms than most of what the grid currently does. Every subsequent decision about Mongolian electricity, tariffs, dispatch, winter reserve margins, import contracts with Russia, would be taken in the shadow of that load. The obvious rejoinder is that this is the entire point. The initiative is explicitly about building new renewable generation alongside the compute, so the denominator grows with the numerator. Good. That rejoinder is correct, and it is also precisely the thing an MOU cannot guarantee. The Grid That Already Struggles to Get Through Winter Because the Mongolian grid is not a blank sheet waiting for gigawatts of Gobi solar. It is a system already under acute strain. Fossil fuels generated 91.1 per cent of Mongolia's electricity in 2024, some 8.04 terawatt hours, with coal alone at 7.93. Renewables managed 8.9 per cent, about 0.79 terawatt hours, despite making up a considerably larger share of installed capacity, which tells you how much sits idle or curtailed. The country's target under its Vision 2050 strategy is 30 per cent renewable capacity by 2030, and it is not obviously on track. The Central Energy System, which supplies more than 70 per cent of the country, is built around Soviet-design combined heat and power plants running baseload, interconnected with the Russian grid. Those plants do double duty: they heat Ulaanbaatar, in a city where winter temperatures fall well below minus 20 Celsius. Utilisation of ageing CHP units during winter peak hours has been reported above 90 per cent, and peak capacity shortage is the acknowledged urgent problem of the Mongolian energy sector. Mongolia imports up to 300 megawatts from Russia to help cover it. Now layer the air pollution problem on top. Households in the ger districts, largely migrants from a countryside emptied by successive dzud disasters, burn coal for heat because they lack grid connections. In the depths of winter, PM2.5 concentrations in Ulaanbaatar have been recorded at levels many multiples of World Health Organisation guidance, and the city has ranked among the most polluted on earth. The 2019 raw coal ban and the switch to briquettes cut average winter concentrations substantially, by around half against the 2016 to 2017 baseline, which was a real public health achievement won at real political cost. Put these together and a sharp equity question emerges, one with nothing to do with foreign companies. There are Mongolian households burning coal indoors because they are not connected to the grid, in the same country now offering four-cent electricity and 24-month grid connections to foreign compute. If the new renewable build follows the data centre load, and the transmission gets built to the server halls rather than to the ger districts, Mongolia will have used its renewable endowment to decarbonise someone else's inference workload while its own citizens keep lighting stoves. That is not an inevitability. It is a design choice, exactly the kind that phrases like “sovereign by design” are supposed to govern. Sovereignty Has a Definition and an MOU Is Not It So what would sovereign actually mean? The word is doing enormous work in this story, and it is worth pinning down. The concept, in its current form, was popularised less by political theorists than by a chip vendor. Jensen Huang has been evangelising sovereign AI since at least 2023, telling the World Governments Summit in Dubai in February 2024 that “every country needs to own the production of their own intelligence” and that a national model “codifies your culture, your society's intelligence, your common sense, your history.” It is a compelling argument. It is also, as critics note with some relish, one that ends with nation states becoming the next category of customer after the hyperscalers. The Center for a New American Security has done the most rigorous work on whether the rhetoric survives contact with balance sheets. Its Sovereign AI Index, authored by Pablo Chavez, Vivek Chilukuri and Ruby Scanlon and updated in August 2026, tracks 185 sovereign AI projects worldwide, defining sovereign AI as a government-backed initiative tied explicitly to national strategic interests and backed by material public investment in domestic compute, models or data. The findings are sobering for anyone taking the word at face value. More than 60 per cent of tracked projects disclose at least one foreign partner, and four-fifths of those partnerships involve a United States company. Nvidia alone supplies the GPUs for 45 per cent of all tracked infrastructure projects. The money is even more concentrated than the compute: the Middle East and East Asia account for over 80 per cent of disclosed investment, the UAE and Japan alone represent nearly two-thirds, and the top ten spenders account for roughly 90 per cent. The index's own summary of everyone else is that “the gap between sovereign aspiration and capacity remains vast.” This is the pattern Mongolia is entering. Indonesia's first sovereign AI factory runs on Nvidia silicon through Indosat. Kenya's ambitions run through Cassava Technologies, again on Nvidia. India's programme, the Gulf states' programmes and Europe's all sit on the same foundation. The sovereignty on offer is real but partial: jurisdictional control over data at rest, genuinely worth having, coexisting with dependency on a foreign accelerator supply chain, foreign software stacks, foreign update cycles and foreign export-control regimes. Set against that, here is what “sovereign by design” would need to mean if it were an engineering specification rather than a slogan. Ownership of the physical assets, or a defined path to it, rather than a long-term lease of Mongolian land and Mongolian electricity to a foreign balance sheet. Control of the model layer, meaning the capacity to train and fine-tune, not merely to host someone else's inference. Domestic skills, meaning Mongolian engineers operating the facility rather than a rotating expatriate crew. Legal jurisdiction over the data, which is the one dimension Nomin Chinbat explicitly claimed and the one most plausibly deliverable. And grid control, meaning the state retains the authority to curtail a data centre before it curtails a hospital in January. Against that specification, what does an MOU deliver? A memorandum of understanding is generally not legally binding. It documents intentions, expectations and responsibilities before a formal contract exists. Whether any particular MOU binds anyone depends on its wording and the parties' conduct rather than the label, and the published materials for the NOVVA agreement do not specify its binding status either way. What is publicly known is that it covers four workstreams, one of which is “policy consultation”, which is to say that the foreign infrastructure developer has been given a seat at the table where Mongolia writes the rules governing foreign infrastructure developers. That is not scandalous. It is entirely normal, and expertise has to come from somewhere. But it is the opposite of a footnote in a discussion about sovereignty, and it deserves naming rather than glossing. What Can and Cannot Be Verified About NOVVA Group The brief here is to be precise rather than insinuating, so let us be precise. NOVVA Group is a real company with a verifiable, if short, public record. Its releases carry a Hong Kong dateline. Its founder and chief executive is named as Steven Liu. Vincent Wen, who signed in Ulaanbaatar, is its managing director. Its own boilerplate describes it as “a global AI-enabling energy infrastructure platform that originates, finances, builds, and operates bankable clean energy assets across Southeast Asia and Latin America.” In June 2026 it announced the acquisition of the 120 megawatt-peak San Jose Solar Power Plant in Bukidnon, Mindanao, in the Philippines, expected to generate over 200 gigawatt hours annually, and it has referenced a Colombian solar portfolio. That is a genuine renewable developer with genuine assets. It is also a company whose disclosed operating portfolio is measured in the low hundreds of megawatts of solar, now signing a memorandum covering data centres, wind, solar, battery storage and national policy consultation in a country where a single hyperscale campus would reorganise the entire grid. No data centre operating track record appears in its public materials. That is not an accusation. It is an observation about scale and specialism that any counterparty ought to make out loud. One further point of hygiene, because the name invites confusion. There is a separate, unrelated American company called Novva Data Centers, founded in 2020 by Wes Swenson, headquartered in Utah, backed by CIM Group, operating a 180 megawatt flagship campus in West Jordan with waterless cooling and facilities in Colorado and Nevada. Searches surface both entities together. No evidence in the public record connects the Hong Kong-based NOVVA Group to the Utah-based Novva Data Centers, and readers should treat them as distinct unless either says otherwise. Why the Desertification Venue Is the Story, Not the Joke It would be easy to file the setting under irony. A conference about drought produces a deal about server farms. Cue the knowing headline. That reading is too cheap, and it misses what the venue actually reveals. COP17 ran from 17 to 28 August 2026 under the theme “Restoring Land. Restoring Hope.”, bringing together the convention's 197 parties, and it convened during the UN International Year of Rangelands and Pastoralists 2026, declared by the General Assembly at Mongolia's own initiative. The UNCCD's executive secretary, Yasmine Fouad, framed it as the “COP of implementation”. The substantive fight was over a global drought framework left unresolved at COP16 in Riyadh, with the African Group and others pushing for a legally binding protocol and other parties preferring something non-binding. Parties did not agree one. The framework was deferred again, to COP18. What the conference produced instead, on its closing day, was a finance portfolio of 1.3 billion dollars for land restoration and drought resilience, of which 644.5 million dollars was identified as new money. The annual financing gap it was announced against is estimated at around 278 billion dollars. Delegates confronted, at the same time, a gap between 1.5 to 2 billion hectares of degraded land and restoration pledges covering roughly 1 billion, against around 1.8 billion people affected by drought. Hold that alongside the Mongolian numbers. Seventy-seven per cent of Mongolian territory is affected by desertification and land degradation. The country has warmed at more than twice the global average, with average temperatures up around 2.46 degrees over eight decades. Nearly all of Umnugovi province in the South Gobi, which is precisely where the best solar resource sits, is moderately or severely degraded. A World Bank assessment found Southern Gobi groundwater reserves sufficient for roughly a decade. Successive dzud events, the compound summer drought and winter freeze that kills livestock at scale, wiped out around a third of the national herd across 2001 and 2002 and more than 13 per cent in a recent event, driving the rural migration that built the coal-burning ger districts in the first place. Now the sharp version of the observation. The conference did produce money, and it should be credited with it: 1.3 billion dollars of portfolio against a 278 billion dollar annual gap is not nothing, even if the arithmetic is unkind. What it did not produce was the thing parties came to Ulaanbaatar to settle. There is no drought protocol. There is a deferral to COP18. Set beside that, the most concrete private investment outcome of a twelve-day United Nations conference on desertification was 863 megawatts of data centre memoranda. And the parallel structure is exact, which is now a matter of record rather than of prediction: the conference could not agree a binding instrument on drought, and the deals signed on its sidelines were non-binding instruments on compute. Ulaanbaatar in August 2026 was, in both rooms, a festival of stated intentions. That is not hypocrisy. It is a description of how climate adaptation and digital development have been fused into a single policy narrative, in which hosting compute becomes a legitimate climate-resilience strategy because it monetises a renewable endowment that would otherwise sit in the desert doing nothing. Mongolia is not smuggling an industrial policy into a climate venue. It is making the entirely coherent argument that a dryland economy losing its pastoral base needs a new economic base, and that its sun and wind are the only endowment it has that China cannot buy on a rail wagon. The water question is where that argument has to be tested rather than accepted. Siting energy-intensive compute in a dryland economy is not automatically absurd, because cooling technology is a choice, not a constant. Evaporative cooling in a hyperscale facility can consume on the order of 1.5 million litres a day. Closed-loop and chip-level liquid systems can eliminate evaporative water use almost entirely; Microsoft's closed-loop deployments have been reported to cut more than 125 million litres per facility per year. The Utah Novva campus, unrelated as it is, markets waterless cooling precisely because it sits in a desert. The trade-off is real, since dry cooling costs energy where water cooling costs water, but in a cold country with a stranded renewable resource, the energy penalty is the cheaper currency to pay. What matters is whether waterless cooling is contractually mandatory or merely aspirational. Nothing in the public description of the Mongolian memoranda says which. The Jobs Nobody Counts Before Signing One more piece of evidence ought to be in the room before any of these memoranda become contracts, and it concerns what hosting compute does for a host economy. The most careful recent work is by Dany Bahar and Greg Wright, published through Brookings in May 2026 and updated that August. Studying counties that received data centres, they find the data processing sector grows 56 per cent over the first decade and telecommunications 43 per cent, which sounds transformative until you see the absolute numbers: roughly 100 to 200 jobs in a typical county. Wages remain unchanged. House prices rise 2 to 5 per cent. They are explicit that naive estimates which fail to account for pre-existing growth trends overstate the effect. The fiscal side is worse. Virginia's data centre sales tax exemption cost an estimated 1.6 billion dollars in the 2025 fiscal year alone. In hyperscale counties, incentives amount to around 2 per cent of construction investment. In colocation counties, they run to about 62 per cent of total investment, meaning the largest subsidies flow to precisely the facilities that generate the fewest jobs. Other work cited alongside it suggests the net county-level employment effect can be close to zero once sectoral reshuffling is accounted for. Apply that to Mongolia. If 863 megawatts eventually gets built, it might employ several hundred people directly. Against that, it would consume more electricity than the country currently generates, occupy the transmission build-out for a decade, and anchor tariff and dispatch policy around foreign-owned load. The construction phase would be substantial and genuinely valuable. The steady state would be a handful of buildings full of machines that employ very few Mongolians and pay their way mainly through the tax base, the electricity sales and whatever the country negotiates on top. Which is why what gets negotiated on top is the whole ballgame. The Conditions That Would Turn This Into Sovereignty If Mongolia wants “sovereign by design” to be an engineering requirement rather than a conference banner, a fairly specific list would have to appear in the contracts that follow these memoranda. None of it is exotic. All of it has precedent somewhere. Grid additionality, written as a condition precedent. No data centre load energises until the corresponding renewable generation and storage are commissioned and delivering, verified by metered output rather than by nameplate capacity or by unbundled certificates bought elsewhere. This is the single most important clause, because without it the Power Stress Index arithmetic above stops being hypothetical and starts being a winter emergency. A curtailment hierarchy with statutory force. In a system where peak CHP utilisation already exceeds 90 per cent in winter, the state must retain an unambiguous, non-compensable right to shed compute load before it sheds heat or households. Data centres are unusually good candidates for this, since training workloads can be paused in ways that hospitals cannot. Domestic offtake carved out at the source. A defined share of the compute reserved at cost for Mongolian public institutions, universities and firms, denominated in GPU-hours rather than goodwill. This is the difference between hosting the machine room and having access to it, and it is what the Barcelona researchers identify as the precondition for fine-tuning models on a country's own climate vulnerabilities rather than importing forecasts built for somewhere else. For a country losing a third of its herd to dzud events, that is not abstract. Equity and a path to ownership. Sovereignty over an asset you do not own is a contradiction. Structures exist for this: state co-investment, build-operate-transfer with a defined handover, or a sovereign stake alongside the developer. A ninety-nine-year land lease with a foreign balance sheet on the other end is not one of them. Here the Asian Infrastructure Investment Bank's involvement genuinely helps Mongolia's hand, and it deserves saying rather than skipping over. Multilateral development bank money arrives with procurement standards, disclosure obligations and governance conditions attached, and it puts a counterparty in the room whose interests are not identical to the developer's. That is a materially different proposition from a pure foreign-balance-sheet lease, and it is the most plausible answer yet to the question of who funds the additional renewable generation the additionality clause would require. It is not the whole answer. Two billion dollars stated as available is not two billion dollars committed and disbursed, and availability of finance settles nothing about sequencing. Whether the generation is commissioned before the load energises remains a matter for the contracts, not the communiqué. Water accounting written into the permit. Mandatory closed-loop or dry cooling, published withdrawal and consumption figures audited annually, and an absolute prohibition on drawing from Southern Gobi aquifers that a World Bank assessment already put on a roughly ten-year clock. In a country where 77 per cent of the land is degraded, water disclosure is not an ESG nicety, it is the licence to operate. Skills transfer with numbers attached. Not “capacity building” in a recital, but a specified count of Mongolian engineers trained and employed in named roles by a named date, with financial consequences for missing it, and a route into the model layer rather than only the facilities layer. Jurisdiction stated explicitly. If the pitch is that data stored in Mongolia stays outside Chinese and Russian reach, then the governing law, the location of arbitration, the ownership chain of the operating entity and the treatment of foreign government access requests all have to be pinned down in public. The claim of data sovereignty is the most valuable thing Mongolia is selling and the easiest thing to quietly undermine in a schedule. Publication. All of the above, disclosed. Ten memoranda were signed in a fortnight covering more than half the country's installed capacity, and the public record consists largely of press releases. Parliament, and the ger district households who will live with how the transmission gets built, are entitled to see the terms. What This Tells Us About Who Gets to Play Mongolia's ministers were making an argument that the evidence supports. Thirty-two countries host the infrastructure. Three regions hold more than 90 per cent of projected compute. The top ten sovereign AI spenders account for around 90 per cent of disclosed investment. That is not a market outcome anyone should be relaxed about, and a country trying to force its way into that club with the one endowment it has is doing something rational and, on the merits, defensible. But the terms of entry are the point. Mongolia is being offered membership of the AI economy in the role of landlord: providing the land, the wind, the sun, the cold air and the regulatory accommodation, while the accelerators, the models, the customers and the margin remain somewhere else. That is a real economic relationship and it may well be worth having. It is not what the word sovereign means, and the gap is where the next twelve months of contract drafting will decide the outcome. The encouraging thing is that the gap is closeable through instruments Mongolia already controls. Additionality clauses, curtailment rights, domestic GPU-hour allocations, water permits and equity stakes are all within the gift of a state with something scarce to sell, and Mongolia does have something scarce to sell. Its leverage is highest now, before the concrete is poured and the load is on the system. What was signed in Ulaanbaatar was not sovereignty. It was an option on sovereignty, written in non-binding language, at a conference that could not agree binding language about drought either. Ten days on, none of the memoranda signed at that event has been reported as converting into a definitive contract. They remain exactly what they were on the day they were witnessed: statements of intent. Options expire. This one runs on the same clock as the Southern Gobi's groundwater and the country's ability to build 863 megawatts of new clean generation before it is asked to switch on the load. Sources and References 1. PR Newswire, via The Manila Times, “NOVVA Group and Mongolia's MDDIC sign MOU on clean-powered AI infrastructure, supporting Mongolia's 'Sovereign by Design' green AI data centre initiative”, 26 August 2026. https://www.manilatimes.net/2026/08/26/tmt-newswire/pr-newswire/novva-group-and-mongolias-mddic-sign-mou-on-clean-powered-ai-infrastructure-supporting-mongolias-sovereign-by-design-green-ai-data-centre-initiative/2412402 https://www.manilatimes.net/2026/08/26/tmt-newswire/pr-newswire/novva-group-and-mongolias-mddic-sign-mou-on-clean-powered-ai-infrastructure-supporting-mongolias-sovereign-by-design-green-ai-data-centre-initiative/2412402 2. Euronews, “Mongolia positions itself as Asia's next data centre hub”, 25 August 2026. https://www.euronews.com/2026/08/25/mongolia-positions-itself-as-asias-next-data-centre-hub https://www.euronews.com/2026/08/25/mongolia-positions-itself-as-asias-next-data-centre-hub 3. Amirpasha Mozaffari, Amanda Duarte, Lina Teckentrup, Stefano Materia, Gina E. C. Charnley, Lluís Palma, Eulalia Baulenas Serra, Dragana Bojovic, Paula Checchia, Aude Carreric and Francisco Doblas-Reyes Barcelona Supercomputing Center, Imperial College London, ICREA , “The Rise of AI in Weather and Climate Information and its Impact on Global Inequality”, arXiv:2603.05710v2, 5 March 2026, revised 29 July 2026. https://arxiv.org/abs/2603.05710 https://arxiv.org/abs/2603.05710 4. Danbo Chen, Zijun Zhou, Yongyang Cai, Jiahong Qin, Ani Katchova and Lei Chen Ohio State University, Meta Platforms, Zhejiang A&F University , “Concentrated siting of AI data centers drives regional power-system stress under rising global compute demand”, arXiv:2604.06198v1, 13 March 2026. https://arxiv.org/abs/2604.06198 https://arxiv.org/abs/2604.06198 5. Stockholm Environment Institute, “Solar and wind power in Mongolia: 2024 policy overview”, October 2024. https://www.sei.org/wp-content/uploads/2024/10/solar-and-wind-power-in-mongolia-2024-policy-overview-sei2024-046.pdf https://www.sei.org/wp-content/uploads/2024/10/solar-and-wind-power-in-mongolia-2024-policy-overview-sei2024-046.pdf 6. Asian Development Bank, “Unlocking Mongolia's Rich Renewable Energy Potential”. https://www.adb.org/news/features/unlocking-mongolias-rich-renewable-energy-potential https://www.adb.org/news/features/unlocking-mongolias-rich-renewable-energy-potential 7. Low Carbon Power, “Mongolia Electricity Generation Mix”, 2026. https://lowcarbonpower.org/region/Mongolia https://lowcarbonpower.org/region/Mongolia 8. Worldometer, “Mongolia Electricity”, 2024 data. https://www.worldometers.info/electricity/mongolia-electricity/ https://www.worldometers.info/electricity/mongolia-electricity/ 9. World Energy Council, “World Energy Trilemma: Mongolia Country Profile”, February 2025. https://www.worldenergy.org/assets/downloads/Asia Trilemma MONGOLIA Profile Template.pdf https://www.worldenergy.org/assets/downloads/Asia Trilemma MONGOLIA Profile Template.pdf 10. Pablo Chavez, Vivek Chilukuri and Ruby Scanlon, Center for a New American Security, “Sovereign AI Index”, April 2026, updated August 2026. https://interactives.cnas.org/reports/sovereign-ai-index/ https://interactives.cnas.org/reports/sovereign-ai-index/ 11. Dany Bahar and Greg Wright, Brookings Institution, “New evidence on data center employment effects”, 4 May 2026, updated 10 August 2026. https://www.brookings.edu/articles/new-evidence-on-data-center-employment-effects/ https://www.brookings.edu/articles/new-evidence-on-data-center-employment-effects/ 12. NVIDIA, “NVIDIA CEO: Every Country Needs Sovereign AI”, remarks by Jensen Huang at the World Governments Summit, Dubai, 12 February 2024. https://blogs.nvidia.com/blog/world-governments-summit/ https://blogs.nvidia.com/blog/world-governments-summit/ 13. UNCCD, “UNCCD 17th session of the Conference of the Parties COP17 ”, 17 to 28 August 2026. https://www.unccd.int/cop17 https://www.unccd.int/cop17 14. Down To Earth, “UNCCD COP17 opens in Ulaanbaatar with a single demand: enough declarations, turn land pledges into money and action”, August 2026. https://www.downtoearth.org.in/climate-change/unccd-cop17-opens-in-ulaanbaatar-with-a-single-demand-enough-declarations-turn-land-pledges-into-money-and-action https://www.downtoearth.org.in/climate-change/unccd-cop17-opens-in-ulaanbaatar-with-a-single-demand-enough-declarations-turn-land-pledges-into-money-and-action 15. World Resources Institute, “STATEMENT: COP17 Advances Finance but Falls Short on Global Drought Action”, August 2026. https://www.wri.org/news/statement-cop17-advances-finance-falls-short-global-drought-action https://www.wri.org/news/statement-cop17-advances-finance-falls-short-global-drought-action 16. PR Newswire Asia, “NOVVA Group acquires 120 MWp Philippines solar project, anchoring its AI-era power platform in Southeast Asia”, 19 June 2026. https://www.prnewswire.com/apac/news-releases/novva-group-acquires-120-mwp-philippines-solar-project-anchoring-its-ai-era-power-platform-in-southeast-asia-302805075.html https://www.prnewswire.com/apac/news-releases/novva-group-acquires-120-mwp-philippines-solar-project-anchoring-its-ai-era-power-platform-in-southeast-asia-302805075.html 17. Novva Data Centers Utah, USA , “About Us”. https://www.novva.com/about/ https://www.novva.com/about/ 18. United Nations Development Programme, “Saving the Gobi Desert and Mongolian steppes from the dzud will also save lives and livelihoods”. https://www.undp.org/blog/saving-gobi-desert-and-mongolian-steppes-dzud-will-also-save-lives-and-livelihoods https://www.undp.org/blog/saving-gobi-desert-and-mongolian-steppes-dzud-will-also-save-lives-and-livelihoods 19. United Nations Development Programme Mongolia, “For a Thriving Mongolia: The Fight Against Biodiversity Loss, Climate Change, and Desertification”. https://www.undp.org/mongolia/stories/thriving-mongolia-fight-against-biodiversity-loss-climate-change-and-desertification https://www.undp.org/mongolia/stories/thriving-mongolia-fight-against-biodiversity-loss-climate-change-and-desertification 20. The Diplomat, “Toxic Winter: The 'Slow Violence' of Air Pollution in Mongolia”, December 2022. https://thediplomat.com/2022/12/toxic-winter-the-slow-violence-of-air-pollution-in-mongolia/ https://thediplomat.com/2022/12/toxic-winter-the-slow-violence-of-air-pollution-in-mongolia/ 21. IQAir, “Ulaanbaatar among top 10 most polluted cities in the world”, 19 November 2025. https://www.iqair.com/newsroom/ulaanbaatar-among-top-10-most-polluted-cities-in-the-world-11-19-2025 https://www.iqair.com/newsroom/ulaanbaatar-among-top-10-most-polluted-cities-in-the-world-11-19-2025 22. Congressional Research Service, “Mongolia”, In Focus IF10926, 12 February 2025. https://www.everycrsreport.com/files/2025-02-12 IF10926 ec6374a0a1dec1eb1dc79d1f4b0a98345772ab84.html https://www.everycrsreport.com/files/2025-02-12 IF10926 ec6374a0a1dec1eb1dc79d1f4b0a98345772ab84.html 23. Geopolitical Monitor, “Mongolia's 'Third Neighbor': Balancing between China, Russia, and the U.S.“. https://www.geopoliticalmonitor.com/mongolias-third-neighbor-finding-balance-between-china-russia-and-the-u-s/ https://www.geopoliticalmonitor.com/mongolias-third-neighbor-finding-balance-between-china-russia-and-the-u-s/ 24. Environmental and Energy Study Institute, “Data Centers and Water Consumption”. https://www.eesi.org/articles/view/data-centers-and-water-consumption https://www.eesi.org/articles/view/data-centers-and-water-consumption 25. Uptime Institute, “Water is local: generalities do not apply”. https://journal.uptimeinstitute.com/water-is-local-generalities-do-not-apply/ https://journal.uptimeinstitute.com/water-is-local-generalities-do-not-apply/ Tim Green UK-based Systems Theorist & Independent Technology Writer Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk https://smarterarticles.co.uk , challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship. His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities. ORCID: 0009-0002-0156-9795 https://orcid.org/0009-0002-0156-9795 Email: tim@smarterarticles.co.uk mailto:tim@smarterarticles.co.uk Listen to the free weekly SmarterArticles Podcast https://www.smarterarticles.fm