The data centres powering cloud services today are massive consumers of electricity – and demand is soaring. Data centres worldwide used only ~1% of global electricity and are responsible for~0.5% of CO₂ emissions in 2024. To put that in context, data-center power was on the order of a large country's output. But that share is growing fast. The IEA projects data-center electricity demand will more than double by 2030, to around 945 TWh – roughly the size of Japan's entire demand. By 2035, BloombergNEF projects data centers could consume >4% of worldwide electricity. This means, to me, that if data centers were a nation, they would rank fourth in electricity usage. AI and cloud services power much of this growth (analysis estimates data-center demand as ~8–10% of global electricity growth by 2030).
Major cloud providers – AWS, Google, Microsoft Azure – all boast efficiency and renewable-energy use. Amazon Web Services has an industry-leading Power Usage Effectiveness (PUE) of 1.15 (vs ~1.25 industry average). AWS's new silicon ("Graviton" chips) are reportedly running workloads on up to 60% less power than comparable servers. AWS also cites a study that moving from an on-premise data center to AWS can lower a workload's carbon footprint by up to 94%. Amazon has pledged to "water positive" data centers by 2030 (returning more water to communities than they consume) and is using recycled and low-carbon building materials in its buildings.
Google has posted the same types of gains and setbacks. Its 2023 numbers showed 14.3 million tons CO₂e (Scope 1&2) – a 13% year-over-year gain and +48% since 2019 – driven in large part by new AI-hungry facilities. Google's global data-center electricity rose ~17% in 2023 to 24 TWh (roughly 7–10% of overall data-center energy). The company is 100% matching its electricity usage with renewable purchases (seven years in a row) and achieved a 64% carbon-free energy mix in 2023. Its engineers report they are delivering "more than six times as much computing power per unit of electricity" as five years ago. In 2024 Google cut its on-site (non-transport) carbon emissions by 12% while consuming 27% more electricity – a measure of efficiency gains in cooling, chip design, and purchasing clean energy.
Microsoft's Azure cloud operates a network of 300+ data centers. It also has high efficiency: Azure is up to 93% more energy efficient and 98% more carbon efficient than a typical corporate data center. Microsoft has purchased about 7.8 GW of renewable energy globally (under wind/solar PPAs), with the goal of matching 100% of its energy to carbon-free resources by 2025. In practice, Azure's fleet averages PUE ~1.16. The company has unveiled custom designs (open-air cooling, rainwater harvesting, mass timber construction) to cut energy and water use. Importantly, all three suppliers highlight that large, concentrated "hyperscale" clouds allow for much higher server utilization than thousands of under-utilized in-house racks: one AWS analysis claimed that cloud customers use ~77% fewer servers and ~84% less power for the same workloads. In short, clouds can work more efficiently and stronger than dispersed enterprise data centers.
Despite these increases, researchers warn that the cumulative impact of large-cloud expansion will be significant. A Guardian analysis found tech giants' official reports often omit scope-3 emissions and offsets; consolidating on- and off-campus power, Google, Microsoft, Meta, and Apple's data-center emissions were estimated ~7.6× more than reported. In practice, new hyperscale facilities still draw heavily from the grid (often hydrocarbon-intensive at night or in some regions) and require massive cooling. Bloomberg's survey referred to increased local power prices in regions building many data halls. Data centers already consume ~4% of local power in some U.S. regions, a fraction that could surpass 10% by 2028. Skyrocketing hunger for AI might outstrip the rate of clean energy: Goldman Sachs and others estimate data-center emissions at 2.5 billion tons CO₂e by 2030 if left unchecked.
The carbon and energy balance is thus not resolved. Cloud has the potential to radically boost utilization (reducing energy per compute), but businesses are adding servers and GPUs faster than they green the grid. In spite of ambitious renewables goals at AWS, Google, and Microsoft, supply and demand balancing remains challenging. And when critics note that most "renewable" claims rest on certificates and offsets, rather than green electrons in real-time to data centers. In short, centralized clouds do indeed bring efficiency, but they still consume vast absolute resources – and their growth is on track to make computing a major electricity sector in its own right.
Decentralized Alternatives: IPFS, Filecoin, and Others
Decentralized storage networks, on the other hand, aim to spread data (and energy use) across many underutilized machines, which might reduce the reliance on any single grid. IPFS (a peer-to-peer file network) and Filecoin (a blockchain-based storage marketplace) are protocols that enable anyone with spare disk and uptime to contribute to the cloud. Nodes could potentially be anywhere – home PCs, office servers, edge devices – and could run on local renewable energy. Decentralized storage does away with the requirement for huge, power-hungry centralized facilities, enabling optimized, demand-driven utilization and locating nodes in regions with ample green energy. As an example, a NAS or laptop left running for other reasons could also host chunks of shared content; its incremental power use could be low compared to running a 24/7 data center.
In a decentralized system, files are sharded and replicated on many machines. An uploaded file is broken up into encrypted pieces; a number of peers each host some of the pieces, with no single point of failure. That renders the network generally resilient to node failures or censorship (if any node fails the data still persists somewhere). The drawback is additional redundancy: each file may occupy, say, 3× or more total space on the network in order to provide availability. That could translate to extra storage (and power) in total over a single optimized copy. But because the copies are stored on otherwise-unused storage, the advocates argue that incremental power usage is small and can be recovered through scheduling or idle-time exploitation.
The actual energy footprint of large decentralized networks is low. Filecoin, an incentivization of storage on a blockchain, uses very little energy compared to proof-of-work coins: on the order of 0.012 TWh per year (12 GWh) network-wide. To put this in perspective, global Bitcoin mining uses ~150 TWh/year. Most of Filecoin's energy is used up-front (the "sealing" or encoding of data, which is compute-intensive); once stored, maintenance (proof-of-storage) takes very little power. In 2024 Filecoin launched a "Green" program to push miners towards renewables and allow customers to audit carbon sources. IPFS itself does not enforce incentives, so its energy use is simply whatever is needed to keep nodes online. If those nodes are existing always-on devices (say, a home server or campus computer), one could argue the extra consumption is marginal.
Other projects (e.g. Storj, Sia, Arweave) use similar storage proofs. None require heavy, continuous mining: rather they rely on the cost of storing and serving files. This makes their gigabyte-per-energy less than that of a traditional data center but their performance and reliability broad-ranging. Decentralized networks generally trade off speed and guaranteed uptime for strength: access may be slower if peers are far away or off-line, and consistency is more networking-intensive. There are also scaling issues: coordinating millions of nodes adds complexity, and network traffic patterns (gossip protocols, redundant downloads) will waste some bandwidth and energy.
On the plus side, decentralized models automatically include renewables and distributed energy. If a cluster of home nodes is powered by rooftop solar or wind, their storage services run on green electrons. Some visionaries predict "peer-to-peer sharing economies" in which underused laptops donate capacity, the way SETI@home harnesses cycles. In practice this will take a while to flesh out. Thus far these networks are small-scale relative to hyperscale clouds, and typically tied to tokens or specific use-cases. But the technical compromises are clear: decentralized clouds sacrifice some efficiency and speed for autonomy, redundancy and potential sustainability benefits.
Balancing Act: Efficiency vs. Resilience
The net climatic impact of any cloud design depends on many factors. Centralized providers value efficiency and economies of scale, but must scale huge facilities (drawing on community grids) to meet demand. Decentralized networks reduce trust and central risk, and can run on distributed green power, but pay overhead in data sharding and don't have the advanced cooling and server efficiency of purpose-built data centers. Theoretically, a widely-deployed decentralized cloud would absorb computing workloads in spare capacity, flattening peaks and devouring local clean power. Practically, today's internet traffic still is topped by large data-center clusters: video, enterprise, AI, etc. A realistic future could combine both: hyperscale clouds pushing hard on efficiency and renewables, and niche decentralized systems handling archiving or peer-focused apps with lower overhead.
What is certain is that compute and storage of data have a climate price. Data centers use similar energy to whole industries (and are among the only fast-rising emitters). Each search, stream or "ask" of ChatGPT depends on this huge infrastructure.The challenge for society is how to provide it. More efficient cooling, on-site renewables, power-shifting and carbon capture can rein in large clouds' impact. Meanwhile, emerging architectures offer an interesting way of sidestepping some garbage. The future will tell whether we can balance our digital convenience with the planet's constraints – or whether the carbon cloud of the cloud will further dim our climate goals.
Spacecoin: A Cloud Above the Clouds
Spacecoin, a Decentralized Physical Infrastructure Network (DePIN) on the Creditcoin blockchain, turns the problem on its head. Instead of more ground-based server farms, it relies on low-earth orbit satellites that function as decentralized internet nodes.
Why is this relevant to the climate?
- No mega data halls. Satellites lack the large, water-hungry cooling centers of Earth-based hyperscale facilities.
- Energy flexibility. Space-based solar energy powers orbital nodes, bypassing the fossil-fueled grids terrestrial data centers depend on.
- Distributed architecture. Like Filecoin, Spacecoin taps into spare or low-usage resources—on a global scale in this case, enabling users and communities to contribute nodes without building new server warehouses.
- Resilience. A satellite constellation sidesteps the central hub bottleneck. If one node fails, the rest of the mesh keeps on serving, reducing wasteful redundancy.
Briefly, Spacecoin looks not just at decentralization of power, but decarbonization through space-based dissemination.
Greener Internet, Freer Internet
The ecological page is intimately linked to the project's broader goal: breaking up central control. Centralized clouds are not just power-hungry, they're also guardians of the digital economy. Whoever owns the data centers controls content, access, and ultimately, speech. A decentralized orbital mesh can potentially cut the carbon cost of connectivity as well as restore the web's original ethos of openness.
The climate case makes things even clearer: if hyperscale clouds become the fourth-largest "country" of electricity consumption, the internet itself may prove unsustainable. A decentralized, satellite-propelled network, by contrast, would be able to align connectivity with the planet's limits.
The Choice Ahead
Do we continue down the path of ever-larger server farms, hoping efficiency and renewability will keep pace with demand? Or do we toy with radically new concepts—such as distributed storage, peer-to-peer architectures, and orbital DePINs such as Spacecoin?
The answer may not be either/or. Centralized titans will always be needed to do some things. But if projects like Spacecoin work out, they could show that decentralization is not merely a way to avoid censorship. It could be the key to building an internet that doesn't burn the world in order to connect it.