The life and death of Microsoft’s Moonshot
RIP to Microsoft’s moonshot. 2020 – 2025. Born in an era of ‘climate pledges‘, months after the 2019 global climate strikes. Found dead in May 2025, crushed under big tech’s generative “AI” stampede.
For five long and beautiful years, Microsoft has tortured the ‘moonshot’ metaphor when describing its climate goals. In its landmark 2020 announcement, VP Brad Smith said:
"Reducing carbon is where the world needs to go, and we recognize that it’s what our customers and employees are asking us to pursue. This is a bold bet — a moonshot — for Microsoft. And it will need to become a moonshot for the world"
It was big talk, but big talk was common at the start of this decade. After the signing of the Paris Climate Agreement in 2015, high-emitting companies and countries realised you could deploy grandiose Obama-style speechifying without really having to demonstrate real, immediate emissions reductions. The moon landing is a favourite in this category of intense over-promising. Norway’s failed adventure in CCS was presented in the 2000s as the ‘moon landing‘ project. Google’s ‘moonshot factory’, the now-awkwardly-named X, is closing down without having solved climate change.
In 2021, after seeing mild emissions reductions and some purchased carbon removal, the company said:
"Using our moonshot analogy, I think of it this way – if our goal is to get to the moon by the end of this decade, this is the equivalent of sending an astronaut into orbit around the earth. It puts us on the right path, but we have a long journey ahead"
The astronaut is in space! Good news. Now let’s take a big sip of coffee and check the status in May 2024:
“In 2020, we unveiled what we called our carbon moonshot. That was before the explosion in artificial intelligence......so in many ways the moon is five times as far away as it was in 2020, if you just think of our own forecast for the expansion of AI and its electrical needs"
Oh no, the moon moved five times further away. I hate it when the moon does that.
Here’s what actually happened: Microsoft built the rocket, left the atmosphere, and shot off in precisely the wrong direction. Then blamed the moon for suddenly being too far away, as they intentionally hurtled away from it.
The company’s decision makers have actively chosen to massively expand the energy they directly consume (mostly in data centres), and the energy they indirectly consume (mostly manufacturing devices and building new data centres). These numbers have risen massively in recent years, and that means the company’s emissions have risen too, garnering headlines last year about the company missing its targets.
In February 2025, Microsoft offered an ‘update‘:
"In 2020, Microsoft leaders referred to our sustainability goals as a “moonshot,” and nearly five years later, we have had to acknowledge that the moon has gotten further away. However, the force creating this distance from our goals in the short term is the same one that will help us build a bigger, faster, and more powerful rocket to reach them in the long term: artificial intelligence (AI)"
This contains a tacit admission the company’s AI-specific investments are a rocket that’s blasting them in the wrong direction – but also that the rocket will become so powerful they can turn around and zoom back towards the moon. In case you were thinking ‘that sounds like hyperbole’, just relax. This is what they put at the end of that paragraph:
"This is not hyperbole"
Fast forward to May 2025. The moonshot has met a sad and subtle end: Microsoft’s latest update unceremoniously dumps the metaphor, unmentioned in the blog post or the full report. We have lost all communication with the rocket, and it’s being scrubbed from the databases.
We have come back down to Earth with a far simpler analogy:
As we remain focused on sustained progress towards our 2030 goals, it has become clear that our journey towards being carbon negative is a marathon, not a sprint.
Okay, it’s a marathon now. But the same problem exists: on a journey towards a goal, you should be heading towards the goal, not away from it (whether that’s the moon, or a finish line). This is really the absolute basics of running a marathon.
For every step the company takes towards the end, they take another 500 backwards. Microsoft is rocketing / marathon-running / whatever-ing in the wrong direction.
It is clear that however climate factors into the company’s decisions, the anxious chasing of maximum generative content infecting every piece of software takes precedence every time. That is a race the company is truly invested in: the frantic, senseless and hyperactive competition to inject generative text, images and videos into every single piece of software in existence, in the hope that something sticks, and that the thing that sticks saves an industry that stopped inventing new and useful things a decade ago. An industry that stopped inventing useful things a decade ago is resorting to environmentally ruinous anxiety-driven growth mania. Let’s dig into the details.
Power and glory
Big tech consumes a lot of electrical energy. Data centres provide cloud services (like emails, file storage, video calls or streaming) but in recent years, they’ve increasingly been used to both train and operate generative machine learning models, which output text, images, and video based off vast amounts of scraped content, mostly used without permission. These processes tend to use different chips: devices that run far hotter, and consume much more energy than traditional chips in data centres. Very soon, machine learning will surpass Bitcoin mining as a driver of data centre expansion.
The latest update1 from Microsoft shows that the electricity consumption of the company continues to rise. It isn’t linear – the amount by which it rises is rising, too. That is to say: its growth is accelerating. Meta and Microsoft in particular are on an absolute tear:
In Microsoft’s latest report the company brags about how it has ‘only’ seen a 23% increase in emissions, relative to a 168% increase in energy use, and a 71% increase in revenue growth. We’ll come back to emissions, but implicit in that statement is the fact their energy consumption is growing far faster than their revenue:
For the company to make a million bucks in 2020, it burned through 75 megawatt hours of energy. To make a million bucks in 2024, it had to burn through 122 megawatt hours: a 60% increase.
It also matters where their facilities consume electrical energy. The vast majority of Microsoft’s power consumption is concentrated in the United States:
Which is important, because American power grids tend to be heavily reliant on fossil gas. When you multiply the amount of power consumption by the average emissions intensity of the grids on which Microsoft consumes power, you get ‘location based emissions’ – a rough but reasonable estimate of the emissions relating to that power consumption. In 2024, it was just under ten million tonnes of carbon dioxide equivalent.
Microsoft does not include this in its headline emissions data – it uses an adjusted figure known as “market based” emissions. In short: Microsoft claims its actions were the reason some renewable energy came into existence, and generated energy. In claiming that for itself, it also claims the actual electrical energy as its own: clean, and zero emissions. It all depends on whether the company really did cause the renewable energy to exist.
I know this is confusing and I am sorry, but please stick with me, because it’s important.
Microsoft can claim to have caused new renewable energy additions in two ways: either by buying certificates generated by clean energy facilities, or entering into power purchasing agreements with new or existing sites.
In practice, the ‘certificates’ option makes little sense. Buying an old certificate that was generated by a hydro plant built in the 1980s on the other side of the world is dodgy: there’s no chance your purchase helped incentivise that hydro plant. You didn’t cause climate action, and it’s wrong to claim that renewable power as your ‘own’. It has a lot in common with the problems around carbon offsetting. The ‘contracts’ are a bit better, because there’s more chance it helped bring renewable energy online, but not much better.
To demonstrate why this matters so much, here are a bunch of ‘actual’ (location, purple) vs adjusted (‘market’, green) data for some large tech companies:
Microsoft has shifted away from the heavily criticised certificates towards doing deals (‘power purchase agreements’, or PPAs) with clean power facilities. These deals are much more likely to be a thing that helps bring a wind or solar farm into reality. Now, only 22% of the company’s ‘market based’ reductions came from certificates:
PPAs, however, exist on a big fuzzy spectrum. Can Microsoft really claim to have been responsible for a wind farm generating power if it already exists? As analytics firm NewClimate Institute wrote2:
"While this shift to PPAs is positive, it is not a silver bullet for reducing electricity-related emissions: long-term contracts for local PPAs are more likely to provide effective support for increasing renewable capacity in a grid, but the causality regarding the additionality of this support is complex and uncertain"
The company’s nuke deals are a great example of this:
"Indeed, Microsoft has recently signed agreements with existing nuclear plant operators in the USA and Canada to purchase RECs on an hourly basis (Microsoft, 2022c; Constellation, 2023) and signed a PPA for a fusion power plant that is scheduled to start operating in 2028 (Helion, 2023). Buying RECs from existing nuclear power plants does not drive the development of additional zero-carbon energy capacity"
Microsoft has also promised to use a significantly more accurate method for matching its power demand to renewable energy output (‘hourly matching’), but this only garners one brief mention in the report, and they share no details on progress.
This is a big part of how companies that are seeing their electrical energy consumption skyrocketing (away from the moon) take the edge off that trend. Microsoft highlight the discrepancy between their emissions and energy increases because they’re good at ‘adjusting’ the data. But that is purely contractual, rather than a measured physical change in real emissions.
The supply chain question
As Microsoft acknowledge in their own report, there is a massive supply chain flowing into the company (think: steel used in a data centre) and flowing out of it (think: the power consumption of a Surface laptop).
Microsoft report far more of these ‘scope 3’, indirect emissions than most other major tech companies do – and with more detail. But read the footnotes, caveats, and adjustments, and you start to recognise a few worrying things.
The first is the use of something Microsoft calls ‘management criteria’. A few scope 3 categories are calculated using a unique methodology. Here’s an example: instead of calculating the lifetime emissions of every product sold in that year, they calculate the 1-year emissions of every product sold prior. Here’s the ‘adjusted’ figures against the ‘raw’ (again, the ‘adjusted’ numbers the ones used in the headline emissions figures) for all scope 3 emissions:
In 2024, the difference between the two methods isn’t massive. And the ‘unadjusted’ numbers above look…okay? The company’s scope 3 emissions seem to have fallen for the first time in a long time. But when you break it down into the categories, a lot of weird things emerge.
Category 2 – ‘capital goods’ – is rising fast thanks to the company’s continued heavy investments into building new data centres. Their investments in green steel and green concrete are clearly dwarfed by the massive scale of their build-out.
But emissions from the use of sold products have plummeted, and category 1 emissions have fallen slightly. Why? Are these due to material reductions in energy consumption (such as Xbox’s energy saving features), or are Microsoft being told by third parties that they’ve bought renewable energy certificates?
One estimate comes from a 2024 report from Stand.Earth, which found that most of Microsoft’s suppliers mostly rely on fossil fuels, and where they claim to be using renewable energy, most of that comes from the highly suspect certificates rather than any less-bad claims of renewable sourcing. I would bet there’s a good chance Microsoft is simply accepting the claims of supplier’s ‘cleanliness’ here, and reproducing them as if they count as real climate action and falling scope 3 emissions.
Tellingly, for Scope 3 Category 3 (the emissions that relate to extracting, processing and using fuels for energy, like methane dumping for fossil gas), Microsoft oddly stopped reporting unadjusted (‘location-based’) figures starting in last year’s report. In fact, every scope 3 category is ‘market based’ – with no view of what the unadjusted (‘location based’) numbers actually look like.
As good as Microsoft is at disclosing more supply chain information than most, it is pretty clear that there is still a lot being hidden here – things that very materially impact the overall picture the company is presenting on its emissions. All we can say for sure is that even the “unadjusted” numbers for scope 3 are actually adjusted – and that the real impacts are higher than the company is reporting. Again: taking the edge off the severity of the company’s direction as it races frantically towards higher consumption of fossil fuelled energy.
The carbon removals bubble
Microsoft don’t count carbon offsets against their total headline emissions numbers – this is good. But Microsoft talks about carbon removal a lot. It features prominently on their website, in this sustainability report and across social media and in interviews. There is a new headline about the company’s carbon removal contracts seemingly every few weeks.
While their contracts with renewable energy deliver electrical energy rapidly after signing, there doesn’t seem to have been any material delivery of ‘contracted’ carbon removal, with the space in between the amount delivered and amount contract widening each year:
Microsoft promises to ‘remove’ all of its operational emissions since 1970 – and if the reported contracts in this report are fulfilled and the companies actually remove and permanently store carbon, that’ll be fulfilled comfortably.
Here’s the problem: while the carbon removal industry has been going wild with contracts, pre-purchases and trading, only a small fraction of those future contracts have actually been delivered as real permanent removal of carbon from the atmosphere. All the hype and hysteria of big tech combined with the physical challenges of removing carbon from the atmosphere are combining to create what looks alarmingly like a removals bubble3.
This could be accepted as the expected circumstances of an early-stage atmospheric clean-up project. But the world’s biggest polluters, fossil fuel lenders, and worst-emitting tech companies are the ones lining up to buy contracts, and carbon offsets from undelivered removals are being traded and claimed against emissions. That makes this industry look way more like CCS than solar.
bsky.app/profile/rishpardikar.…
You can see this same pattern of massive contracting and under-delivery reflected in global carbon removals (partly because Microsoft comprises such a wildly huge proportion of global pre-purchasing):
What actually happens if Microsoft’s all-in bet on carbon removal does not eventuate? It has clearly skewed its efforts towards emitting first and removing later, rather than reducing greenhouse gas emissions at the source. If the bubble pops, the company will simply be left a decade of absurdly steep emissions growth.
Post-Moonshot Microsoft
Microsoft claim that their emissions rose 23%, from 2020 to 2024. That alone is not a good look, but if you take away the certificate / contractual adjustments to emissions (at least, the ones we know about), and the company’s own ‘criteria’ for scope 3, and you get something more like a 55.4% increase:
This becomes even more stark when you compare the latest data to the chart4 of projected emissions reductions and carbon removal depicted in the 2020 report, when the moonshot was announced:
What are we even doing here? There is no relationship between this panicky software goliath going all-in on energy-intensive generative slop and the hard physical reality of what an explosion in energy demand means for fossil fuel combustion.
The industry is pivoting to claiming their “AI” investments will create climate solutions that undo all the damage and solve the climate crisis. It is the same break-first-fix-later wishful thinking as we see in high-emitting carbon removal investments, or in geo-engineering. Microsoft obscures what proportion of its energy usage goes to what type of applications of “AI” because the reality is most of it probably relates to the Copilot-generating-spam type, not the AI-predicting-the-weather type.
Microsoft also intentionally excludes its sales of machine learning software services to the fossil fuel industry from its reports. The “Enabled Emissions” campaign, helmed by two former Microsoft sustainability folks, explains how Microsoft once bragged openly about helping oil and gas companies figure out how to extract significantly more fossil fuels.
The moonshot era is done. While Microsoft is keeping its climate targets on paper, it treats them as functionally non-existent. Nothing will override the panicked expansion of data centre power consumption, and nothing will truly neutralise the increase in fossil fuel combustion that occurs as a consequence.
Increasingly, people in tech, data centre and fossil fuel industries are saying out loud what was once whispered in secret: there is no point trying to limit emissions or fix climate change, when you can promise a future super intelligent robot will pick up the pieces.
The shallow promise of a sci-fi future climate-fixing machine intelligence is sometimes believed sincerely, and sometimes presented cynically. Either way, it does the same job: hype-washing the material, real and measurable damage these companies are doing today.
Microsoft’s blend of righteousness and destructiveness is exceedingly dangerous. Explosive energy demand incentivises fossil fuels. It redirects new renewables into servicing the anxieties of tech executives rather than the deep machete cut into fossil fuel combustion we need right away. The glossy, shallow stories of ‘AI for climate‘ instil a wide-eyed overconfidence, as their bloated generative slop rocket fires them in precisely the wrong direction, burning us up in their wake.
The era of the empty climate pledge is over. No logical plea will pierce the buzzing illogical anxiety driving these business decisions. The only way to protect ourselves is to criticise this mess directly, and forcefully. If they aren’t stopped, the damage they’ll do is physically irreversible.
This is not hyperbole.
- You can access the full dataset I’ve created collating tech company emissions data here 🙂 ↩︎
- The whole report is really worth reading if you’re into a more technical analysis of the company’s renewable claims – it also goes into detail on the whole concept of time-matching renewable energy, which very interesting. ↩︎
- Joe Romm has been writing in plenty of detail about many of the problems with Microsoft’s carbon removal deals, including in this seminal paper from a few years ago. ↩︎
- May I strongly suggest using Webplot Digitizer if you want to extract data from charts! ↩︎
Voluntary carbon offsets are headed for a crash
Carbon offsets — whereby one party pays another party to reduce carbon emissions — have always been dodgy, but now they are threatening to undermine the Paris climate agreement.David Roberts (Volts)