The Great Data Centre Gold Rush
A.I. Boom to Climate Disaster?
Across history there have been many resource booms that have reshaped our world and driven people to extreme, perhaps excess, activity. Gold, Diamonds, South Sea Spices, Oil, Data, and now the latest great resource rush has begun: Compute. Everyone wants it, no-one has enough of it, and if you can get your hands on the means of producing it, it promises to make you rich beyond your wildest dreams of avarice thanks to the ever increasing appetites of the world's Artificial Intelligences.
Worldwide right now we are witnessing record expansion in data centre construction driven by expansion in the use and training of A.I. In 2024 data centre deals alone reached a record $57 billion dollars (USD). Rather than a few large contracts this number include a huge growth in smaller contracts, indicative of a broad expansion in the market and this is only speeding up. Meta have announced this year they will spend $60-65 billion in A.I. related capital expenditure, Microsoft has signalled their intent to spend $80 billion whilst the newly announced Stargate project in the United States claims it will spend $500 billion over the next four years, with $100 billion up front to begin. No-one wants to be left behind in this race, just the largest tech players will together spend more than $200 billion over the next year, each one looking to outspend or out-innovate their competition. A true arms race.
A.I. continues to dominate the technology news cycles every week as demand for GPU's, and the chips that power them, continues to grow by leaps and bounds - far outstripping supply. There is competition not only at the corporate level but among nations too: Data centres are now considered critical national infrastructure and even, in America this week, announced as critical national security assets. Make no mistake, this is no mere economic boom but a complete paradigm shift reshaping our world and laying the foundations for a new age: the age of Artificial Intelligence.
The Driving Force: AI and its Insatiable Demand
The key issue here is the completely unheard of explosive growth of Data Centres across the globe. Artificial Intelligence, the potential of them and the desire for them, is the driver and primary catalyst of an incredible surge in demand for compute time, pushing even the large cloud data centres to their limits. Since ChatGPT burst on to the scene several years ago and made OpenAI a household name the race has gathered momentum, the competition has gathered, and costs are growing. A.I. use, performance and size have all grown exponentially year on year ever since and are fast outstripping the existing asset resources of even the largest, wealthiest companies; leading directly to this coming investment rush. According to estimates, A.I. models are growing at a speed of roughly 3x per year with their computational requirements said to be "nothing if not intense."
"Ultimately, this all points to the theme that is quickly coming to define 2025: Big Tech sees AI as the most existential technology of the coming era and will keep spending like hell to make sure OpenAI doesn’t completely run away with it." - Alex Heath, The AI spending frenzy is just getting started
Everyone wants to be king of the hill and every few months brings a new model release, a new press release, a new frontrunner. Sam Altman and OpenAI are the ones to beat having had the first mover advantage for a few years now, but there is a lot of elbowing at the top and arguably ChatGPT isn't currently the top model: they have some impressive claims for GPT-5, their next version, however it is not currently available to the public with no official announcement of when it will be. In the meantime, Zuckerberg has stated that Meta's AI will become "the leading assistant serving moire than a billion people," and that their Llama 4 model would become the "leading state of the art model", very explicitly throwing down the gauntlet to Altman and other would-be champions. There are no shortage of contenders, with OpenAI, Meta, Anthropic, DeepSeek, Google and xAI all with models that beat the field in particular benchmarks.
It's not only big tech who are at each other's throats either - the nations of the world aren't far behind. A.I. supremacy is fast becoming the new millennium’s "Space Race" as the U.S. and China face off across the globe. The U.S. is attempting to cripple China's capacity by cutting off its supply of tensor-chips, the core component in GPUs used for training A.I.'s and China has responded through innovation, with a Chinese company (DeepSeek) releasing a top-level competitive A.I. model at Christmas that was trained for a tiny fraction of the cost and resources as those of the big tech companies. They followed up with a release of a reasoning model this week that sits comfortably with much larger, well funded competitors, underlining perhaps that trade wars and economic coercion may not be enough to allow the U.S. to rest easy.
While these two giants face off other nations are not sitting idle either. The buzzword this week has been "Sovereign A.I.", the idea that for national security reasons a country should have its own home-grown A.I. to ensure it is trained in a "non-biased" way; and is no threat from the perspective of its host country. Whilst there is a touch of the kind of naked flag-waving that has been so popular lately to this, there may be something to the idea; try asking the DeepSeek model about Taiwan or Tienanmen Square for instance for an example. Whether you think this calls for sovereign A.I. within your own country, or a wide variety of models trained around the world to allow for diverse perspectives and opinions, really depends on your own personal philosophy.
The U.K. have thrown their hat in the sovereignty ring this week leaving no room for ambiguity as to their intentions. After receiving the 50-point economic "AI Opportunities Action Plan", Prime Minister announced they would be adopting all 50 recommendations aimed at making the UK an A.I. leader, with a broad focus including building local infrastructure and potentially enabling U.K. groups to train models to compete with those owned by the big tech companies.
For a few years this has been an expanding economic boom, and for many a worrying "hype bubble", however as it gains critical mass it appears to be showing all the signs of a transformative shift in global power dynamics the likes of which we haven't seen since the industrial revolution or the original personal computing boom of the late 20th century.
Environmental Consequences: A Global Challenge
While big tech races ahead and governments and economists wet their lips in anticipation, alarm bells have been sounding across the world as the data centre rush starts to really take hold. Whilst touted as a strong sign of economic health and progress, the dirty secret of data centres is their massive, often hidden and fundamentally unsustainable environment costs. While politicians rush to claim responsibility for bringing new development to their constituencies and everyone up to the Prime Minister of the U.K. and President of the U.S. are claiming victory with huge, impressive infrastructure deals and investments, the true costs are yet to be calculated.
It begins with power. Data centres consumed 460 TWh of electricity in 2022 and this is set to double by 2026, to consume potentially more than 1,000 TWh by 20261. This is equivalent to adding the entire power consumption of a developed country such as Sweden or Germany to the world power grids in just four years.
This is driven almost entirely by the compute requirements of A.I. According to DBTA, current estimates suggest that data centres account for 3% of global electricity consumption, a huge percetage already that will potentially rise to 10% by 2030. That's 10% of the entire world's electricity usage. The primary cause appears to be a combination of new data centre construction and exponential growth in power requirements for the chips that power them. Whilst growing in power and speed, these chips have also been shooting ahead in power requirements; the latest Nvidia AI chips consume up to 300% more power than their predecessors. Electricity makes up 46-60% of total datacenter operating expenses; the growth of A.I. and the associated data centre gold rush is great news for power generators and distributors, but for the world? The statistic that the governments are not so quick to tout is one released by Morgan Stanley whose recent study predicts that greenhouse gas emissions from data centres will be three times higher between now and the end of the decade than they would have been if pre-A.I. trends had continued.2
Electricity isn't the only story however so lets talk water. In 2015 global data centre water consumption was 738 million litres a day, which rose to 840 million litres in 2021. The explosive growth of data centres will result in an equivalent growth in water usage; a mid-sized U.S. data centre uses approximately 1.14 million litres of water daily, equivalent to the water consumption of 100,000 homes. About 25% of this is used directly (primarily for cooling) and the other 75% is an indirect cost of the electricity generation and water treatment attributed to data centres on the grids. With the projected growth, therefore, Bluefield research predicts data centre water usage will top 1.7 billion litres daily by 2030, more than double the 2017 consumption.3 In 2022 Microsoft alone consumed 17.5 million litres a day, according to their own environmental report. By 2027, global A.I. driven demand for water is expected to account for 4 to 6.5 billion litres of water withdrawn from world supplies, more than 4-6 times the total annual water withdrawl of the whole of Denmark.
What has been the tech industry reaction to these realities? The key efforts to date have mostly involved purchasing carbon offset credits. The problems with these are too numerous to go into here, however they have been well documented elsewhere; rampant corruption, money laundering, false accounting and according to IACC, up to 90% of carbon credits are potentially worthless in terms of helping the environment4 or anything beyond the PR of the purchasing company - a practice known as green washing.
Despite these known issues, it appears to be the tech industry's primary "solution" for their carbon problem, with Meta committing to a deal to purchase up to 3.9 million carbon credits from BTG Pactual, a deal rumoured to be worth about $16 million, and Microsoft purchasing 8 million carbon credits from BTG Pactual in June 2024. The effectiveness of carbon offset credits as a PR strategy appear to be waning however as Google shifted strategy away from mass carbon offsets in 2023, and Microsoft this week announce their new carbon plan. Both companies moved to a similar strategy: carbon removal credits. Google committed to $35 million to carbon removal credits, contracted for 62,500 tCL2e of carbon removal and invested $200 million in a carbon removal market development fund. Microsoft this week made a new agreement for 3.5 million tons of carbon removal credits over 25 years, in additional to a previous agreement for 3 million tons over 15 years. The new agreement is worth approximately $200 million and involve Re.green working with landowners in Brazil to restore degraded land by replanting native species.
These schemes appear designed primarily for green washing purposes and don't go nearly far enough. Not only are the schemes themselves unreliable and likely to do very little to actually offset the carbon but the scales of investment are vastly inequitable. $200 million from Microsoft might seem like a lot of money until you realise that is $200 million over 25 years (about $8 million a year), whilst they plan to spend more than $80 billion this year alone on A.I. related infrastructure, making this a miniscule investment. In the meantime, Microsoft's carbon emissions have increased by nearly 30 percent since 2020 and Google's have increased by more than 48% since 2019; a rise that will only accelerate as more and more money is pumped into A.I. data centres.
Green washing and PR campaigns simply aren't enough, if this revolution is going to go ahead (and it is absolutely going to go ahead) we need a paradigm shift. We need system changes to power sources for power-hungry utilities such as data centres and better consumption practices to increase sector sustainability or we're going to be crushed between AI progress and environmental devastation.
There is hope however, recently this message seems to be getting through and we are seeing more commitments. For instance:
From Microsoft:
Carbon Negative by 2030
$1 billion direct investment in carbon removal technologies.
Water positive by 2030
Zero waste by 2030
All data centres to be powered 100% from renewables by 2030
From Google
Net Zero Emissions by 2030
24/7 Carbon Free Energy (all locations) by 2030
$30 million funding for projects accelerating technology advances in climate action.
Water replenishment (120%) across all its locations by 2030. It’s worth noting that Google locations have been associated with water scarcity due to their usage in the past.
Meta, Amazon and other big tech companies have made similiar commitments showing the intention, at least at some level of the companies, of working towards fixing this issue; at minimum a recognition that the problem exists. This new arms race however threatens to make the success of these problems impossible.
Power Demands: A Race for Resources
Investment in data centres is going reach new record heights this year and likely continue to do so year on year as the A.I. race heats up. We have discussed the effect this has on power requirements, the additional pollution and water supply damage this causes, but a question remains: where exactly is this power going to come from?
The numbers are extremely clear, globally there isn't anywhere near enough power generation to support the sort of data centre growth planned for the rest of this decade. An entire industry that was already extremely power hungry doubling it's consumption over just a few years? The impact of that on existing infrastructure, designed for much lower requirements, is obvious. Along with data centre construction tech companies are already recognising that they have to invest in power generation expansion as well if they wish to meet the schedules they are setting for themselves. Despite the risks involved the ultra-competitive nature of an arms race pushes all other considerations aside and the idea of slowing expansion is unlikely to be seriously considered, if it's considered at all.
As this realisation slowly sinks in world wide the tech companies have come out to meet it with solutions that, if not "green" at least show they are considering these issues. Microsoft for instance has partnered with Brookfield Renewable Partners to add renewable capacity for the US and European data centres and is financially backing the restart of Three Mile Island nuclear plant, with nuclear more often these days being seen as a potentially much cleaner "flexible" source of energy generation than traditional fossil fuel plants.
A key difference (and difficulty) with many renewable energy sources is the lack of this flexibility - solar for instance provides a variable amount of power based solely on its access to sunlight at any given time. Addressing this problem and making sure power is available when its needed requires considerable management and the use of extensive power storage options such as batteries, or pumped hydro. A traditional coal-fired plant, on the other hand, produces a known and stable quantity of energy so long as it has access to fuel and by retaining additional turbines can be prepared to "ramp up" how much it generates in response to a crisis or surge in demand if required.
Google has announced significant investments in solar energy towards powering their future data centres, acknowledging that the existing power grid will not be sufficient to support the energy demands of A.I. Meta and Google, along with several other companies, have investigated the use of exciting new nuclear technologies in the form of small modular reactors and have projects well underway to trial them for powering their data centres. Meta has partnered with Sage Geosystems to expand geothermal energy use to provide carbon-free power to data centres, and Amazon purchased a nuclear powered data centre in Pennsylvania from Talen Energy. As mentioned above Amazon, Microsoft, Meta and Google have collectively all targeted net-zero emissions for their companies before 2030 and committed to exploring alternative energy sources.
All good so far, so what’s the rub? In short, the incompatibility of these companies A.I. ambitions and their stated environmental targets. As mentioned, Morgan Stanley has predicted data centre emissions could triple by 2030, potentially emitting 2.5 billion tonnes of CO2 due to the energy demands of AI technologies. The reported emissions increases alone make it near impossible for companies like Microsoft and Google to meet their targets. Worse than that, the combined emissions of the race across the world is more than likely going make it impossible for many nations to keep to their emissions targets as well; something many countries were already struggling with; and countries like the U.S., China and the U.K., by placing so much emphasis on the national interest aspect of A.I. development, have effectively telegraphed they have no intentions of reigning in this rampant carbon growth either.
Part of the reason for this is despite the “100% renewable” aspirations of these companies and their explorations of alternative energy sources, many of the proposed solutions are tested yet (such as modular small reactors) or take a considerable amount of time to build (Modular or regular nuclear reactors, hydro-electric and geothermal power plants, etc). The first modular Nuclear reactors are scheduled to begin generating until 2030 at the absolute earliest and that's assuming there are no delays due to the experimental nature of this technology. As mentioned, the investments in A.I. data centre expansion aren't happening in 2030 but right now, and the power requirement increases aren’t going to wait until 2030; there needs to be a nearer term method of providing this power. Of course, there is.
Before the AI boom the historical trend was downwards on traditional fossil-fuel powered power plants such as coal-fired plants, especially in the U.S. and Europe. The current trend? A resurgence in coal-fired stations as planned decommissions are delayed or cancelled to meet increasing demands and a surge in demand for fossil fuels; the mining of which is itself a highly emissive activity. In addition recent announcements around data centre investments in the U.S. have revealed that there are as many as 80 planned new gas-fired power plants over the next few years, specifically to meet the demands of data centre expansion. Entergy, for instance, has announced a $3.2 billion plan to build three new gas plants totally 2.3 GW of power generation, specifically to serve Meta's AI data-centre.
The A.I. boom, whatever other effects it has, has rolled back decades of work transitioning away from fossil fuels and significantly reduced the chances of meeting climate targets that were already slipping significantly. There will no doubt be a great deal of effort to make it look, on paper at least, as if the big tech companies meet their goal of completely renewably powered data centres by 2030 however a paper win wont be enough in this case. If every piece of renewable generation they can manage were redirected to serving the needs of big tech it’s possible we could meet their stated goals; but if the hole this makes in the rest of the grid requires 80 new fossil-fuel powered stations and the cancellation of the decommissioning of the worlds dirtiest energy sources it’s still a net loss for the environment overall. What is needed is an acceleration of the transition from fossil fuels to carbon-free power, what we are seeing is the opposite.
Sustainability and Innovation: A Path Forward
"The emissions reduction plans clearly fall short of what is required... the Net Zero promises by these companies is fake." - UN Secretary-General Antonio Guterres
There is immense potential offered to our global society, our economies, productivity, and our lives in general by the continued innovations in A.I. If it even partially meets the promises made by the techno-utopians we are in for an exciting and fascinating few decades from here. Despite the potential for immense good however we cannot ignore the costs. The disparity of investment between expansion of carbon-emitting assets and investment in combatting these issues is concerning and unsustainable - when the environment is damaged it effects everyone but the benefits of this new technology is not as well distributed. It is a moral imperative that those doing the most damage, and accruing the most benefit, should be held most responsible for solving the problems as well.
Green washing has been a sad reality of corporate life for decades now, ever since the rise in environmental awareness that started in the late 20th century. We are far past the point where this can be accepted as a reasonable "effort" of these companies to account for the damage their activities do. Sustainability cannot be allowed to remain a purely PR term, useful for sidestepping blame or judgement for carbon pollution.
"The practice of making misleading claims about the environmental benefits of a product or service is widespread and often shields harmful practices from scrutiny." - John Pabon, Author of "The Great Greenwashing".
Despite the investments we have discussed already, Microsoft has also recently been called out and accused of gross hypocrisy and green washing recently when it was revealed that the company had identified fossil fuel industries as a key growth target for its AI and Cloud Services. These services would be used to aid exploration and production of new fossil fuel resources, making these industries more efficient at their production - and as a side effect, helping the economic sustainability of one of the most polluting industries on the planet. This raises an interesting question: Even ignoring all of the concerns over the validity of carbon credits and other issues, can your business truly claim to be carbon-neutral if it builds and sells technology that helps their customers pollute more efficiently?
The big tech companies do not have to accept the role of villains in our global climate issues, there are opportunities for solutions ripe for the taking; some of them emerging from the very work these companies are focused on. AI, as Microsoft pointed out, could be used to drive efficiencies in fossil fuel production, however it could be just as easily applied to solving this same problem - and is being used for exactly that purpose in a wide variety of innovative ways:
- AI combined with satellite imagery is currently being used to monitor and map deforestation impacts, with company Space Intelligence having already mapped over 1 million hectares across 30 countries. This allows accurate measurement of deforestation rates and forest carbon storage, as well as making it possible to both plan reforestation projects and detect illegal deforestation projects.
- The startup Greyparrot has developed an AI system that enhances waste recycling processes by analysing waste processing facilities, identifying materials that can be recovered and potentially reducing methane emissions from landfills.
- The Ocean Cleanup organisation is using AI to detect plastic pollution in oceans, creating detailed maps to enable efficient clean-up operations.
- The company Sipremo, in Brazil, is using AI to predict climate change disasters and help governments and businesses prepare for potential impacts.
- Google's DeepMind has applied AI algorithms to optimise energy use in its data centres, achieve a 40% reduction in cooling energy consumption, directly reducing its own carbon footprint.
- AI is aiding in the development and optimisation of renewable energy sources such as wind and solar power by predicting energy output and improving maintenance schedules.
This technology, as worrying as its impacts are, contains within it the best chance for slowing and potentially repairing climate impacts we have yet seen both directly and as an enabler of the development and optimisation of other technologies. What is needed is direct investment and true prioritisation of these issues by the same companies who have the moral responsibility, the technology, and the money to make a real difference.
It would be childish to expect these companies to change course purely based on it being the right thing to do, some sense of civic responsibility or even possibly shame. It's 2025 and our business and political leaders have long since abandoned such notions as relics of times passed, sadly. Thus we have to look to these same politicians, who are at least nominally driven by a need for votes, to push for regulation on these industries with regards to environmental impacts. Laws and regulations already exist to hold large corporations responsible for their actions but they are weak and full of loopholes. A concerted effort is needed to rectify this; perhaps another area in which this new technology could be useful? Sadly, given the gold-rush mentality taking hold in the tech industry, the nature of any arms race justifying short-cuts to achieve victory, and the current political state in the U.S., it seems extremely unlikely the legislative will exists to make these changes where they would do the most good.
Navigating the Future of Data
This is happening. The genie doesn't go back into the bottle, the toothpaste into the tube, we cannot uninvent A.I technology and realistically we shouldn't want to. Our technologies have not always been good news, for us or the planet, but our ability to conceive, build and use them is what has brought us up to this point from mere survival and could take us further still. Over the next few years we are going to see a massive increase in computing capacity to continue developing this technology, with all its associated costs and benefits. Our role over the rest of this decade has to be to make sure this investment becomes a step up to a new level of quality of life, wealth, and happiness for humans everywhere - and not the final mistake that leads to our own destruction.
This is the choice we now face, melodramatic as it sounds. Two paths stand before us. One of these paths is that of stewardship - using these new powers to attack the problems facing our planet and make a real difference, to rise above our inherent selfishness and tribal instincts and try to make real progress towards a sustainable, post-scarcity future. For the first time in decades, it really feels as if there may be a path, however treacherous, that could take us there.
"If we are victorious in one more battle with the Romans, we shall be utterly ruined." - Pyrrhus of Epirus, after the Battle of Asculum (279 BC)
Or we could take a path sadly more familiar to all of us. That of short-sighted gain, a focus on zero-sum competition and ever skyrocketing profits that would try to win the world even as it burns down around us. This is where we risk heading if we allow the A.I. revolution to become about petty egos and record profits instead of about uplifting all of humanity. In a time when we appear to be re-embracing all of our worst instincts, tribalism and xenophobia, nationalism and racism, as if we have all collectively forgotten where this path had lead us to so many times before, we are facing a challenge to rise above exactly these impulses. At a time when trust in collective action is low, we need it more than ever, as collectively may be the only way to influence the enormous wealth and power of the companies that control these technologies and the people that control the companies. We should never forget that Citizens United decision aside, Corporations are not people, they are groups of people who have crafted a fine art of weaponising collective behaviour to an efficiency level that neither union or socialist ever truly managed. If we are going to learn from them then perhaps this is the lesson to take.
No matter skin colour, nation of birth or wealth, we all collectively have a stake in life on this planet and our global, if fractured, society. The responsibility of progress rests with all of us and it will be all of us together who will ultimately judged for the decisions we make.
"The test of our progress is not whether we add more to the abundance of those who have much; it is whether we provide enough for those who have too little." - Franklin D. Roosevelt