Power Archives - Taking Stock

September 30, 2025September 30, 2025

With all the new AI tech arriving in the new AI data centres – what is happening to the old tech it is presumably replacing?

🧠 What’s Happening to the Old Tech?

Shadow in the cloud

🔄 Repurposing and Retrofitting

Many traditional CPU-centric server farms are being retrofitted to support GPU-heavy or heterogeneous architectures.
Some legacy racks are adapted for edge computing, non-AI workloads, or low-latency services that don’t require massive AI computing power.

🧹 Decommissioning and Disposal

Obsolete hardware—especially older CPUs and low-density racks—is being decommissioned.
Disposal is a growing concern: e-waste regulations are tightening, and sustainability targets mean companies must recycle or repurpose responsibly.

🏭 Secondary Markets and Resale

Some older servers are sold into secondary markets—used by smaller firms, educational institutions, or regions with less AI demand.
There’s also a niche for refurbished hardware, especially in countries where AI infrastructure is still nascent.

🧊 Cold Storage and Archival Use

Legacy systems are sometimes shifted to cold storage roles—archiving data that doesn’t require real-time access.

These setups are less power-intensive and can extend the life of older tech without compromising performance.

⚠️ Obsolescence Risk

The pace of AI innovation is so fast that even new data centres risk early obsolescence if they’re not designed with future workloads in mind.
Rack densities are climbing—from 36kW to 80kW+—and cooling systems are shifting from air to liquid, meaning older infrastructure simply can’t keep up.

🧭 A Symbolic Shift

This isn’t just about servers—it’s about sovereignty, sustainability, and the philosophy of obsolescence. The old tech isn’t just being replaced; it’s being relegated, repurposed, or ritually retired.

There’s a tech history lesson unfolding about digital mortality, and how each new AI cluster buries a generation of silicon ancestors.

Infographic: ‘New’ AI tech replacing ‘Old’ tech in data centres

🌍 The Green Cost of the AI Boom

⚡ Energy Consumption

AI data centres are power-hungry beasts. In 2023, they consumed around 2% of global electricity—a figure expected to rise by 80% by 2026.
Nvidia’s H100 GPUs, widely used for AI workloads, draw 700 watts each. With millions deployed, the cumulative demand is staggering.

💧 Water Usage

Cooling these high-density clusters often requires millions of litres of water annually. In drought-prone regions, this is sparking local backlash.

🧱 Material Extraction

AI infrastructure depends on critical minerals—lithium, cobalt, rare earths—often mined in ecologically fragile zones.
These supply chains are tied to geopolitical tensions and labour exploitation, especially in the Global South.

🗑️ E-Waste and Obsolescence

As new AI chips replace older hardware, legacy servers are decommissioned—but not always responsibly.
Without strict recycling protocols, this leads to mountains of e-waste, much of which ends up in landfills or exported to countries with lax regulations.

The Cloud Has a Shadow

This isn’t just about silicon—it’s about digital colonialism, resource extraction, and the invisible costs of intelligence. AI may promise smarter sustainability, but its infrastructure is anything but green unless radically reimagined.

⚡ The Energy Cost of Intelligence

🔋 Surging Power Demand

AI data centres are projected to drive a 165% increase in global electricity consumption by 2030, compared to 2023 levels.
In the U.S. alone, data centres could account for 11–12% of total power demand by 2030—up from 3–4% today.
A single hyperscale facility can draw 100 megawatts or more, equivalent to powering 350,000–400,000 electric vehicles annually.

🧠 Why AI Is So Power-Hungry

Training large models like OpenAI Chat GPT or DeepSeek requires massive parallel processing, often using thousands of GPUs.
Each AI query can consume 10× the energy of a Google search, according to the International Energy Agency.
Power density is rising—from 162 kW per square foot today to 176 kW by 2027, meaning more heat, more cooling, and more infrastructure.

🌍 Environmental Fallout

Cooling systems often rely on millions of litres of water annually. For example, in Wisconsin, two AI data centres will consume 3.9 gigawatts of power, more than the state’s nuclear plant.
Without renewable energy sources, this surge risks locking regions into fossil fuel dependency, raising emissions and household energy costs. We are not ready for this massive increase in AI energy production.

Just how clean is green?

The Intelligence Tax

This isn’t just about tech—it’s about who pays for progress. AI promises smarter cities, medicine, and governance, but its infrastructure demands a hidden tax: on grids, ecosystems, and communities.

AI is a hungry beast, and it needs feeding. The genie is out of the bottle!

September 23, 2025September 24, 2025

AI power – the energy hunger game!

As artificial intelligence surges into every corner of modern life—from predictive finance to generative art—the question isn’t just what AI can do, but what it consumes to do it.

The energy appetite of large-scale AI models is no longer a footnote; it’s the headline.

Training a single frontier model can devour as much electricity as hundreds of UK homes use in a year. And once deployed, these systems don’t slim down—they scale up.

Every query, every image generation, every chatbot exchange draws from vast data centres, many powered by fossil fuels or water-intensive cooling systems.

The irony? AI is often pitched as a tool for climate modelling, yet its own carbon footprint is ballooning.

This isn’t just a technical dilemma—it’s a moral one. The race to build smarter, faster, more responsive AI has become a kind of energy arms race. Tech giants tout efficiency gains, but the underlying logic remains extractive: more data, more compute, more power.

Meanwhile, communities near data centres face water shortages, grid strain, and rising costs—all for services they may never use.

Future direction

Where is this heading? On one side, we’ll see ‘greenwashed’ AI—models marketed as sustainable thanks to token offsets or renewable pledges. On the other, a growing movement for ‘degrowth AI’: systems designed to be lean, local, and ethically constrained. Think smaller models trained on curated datasets, prioritising transparency over scale.

AI power – the energy hunger game! NASA’s ambition is to place nuclear power on the moon

Governments are waking up, too. The EU and UK are exploring energy disclosure mandates for AI firms, while some U.S. states are scrutinising water usage and land rights around data infrastructure. But regulation lags behind innovation—and behind marketing.

Ultimately, the energy hunger game isn’t just about watts and emissions. It’s about values. Do we want AI that mirrors our extractive habits, or one that challenges them? Can intelligence be decoupled from excess?

The next frontier isn’t smarter models—it’s wiser ones. And wisdom, unlike raw compute, doesn’t need a megawatt to shine.

Why Nuclear Is Back on the Table

Global Momentum: Thirty-one countries have pledged to triple nuclear capacity by 2050, framing it as a cornerstone of clean energy strategy.
AI’s Power Problem: With data centres projected to consume more energy than Japan by 2026, nuclear is being pitched as the only scalable, low-carbon solution that can deliver round-the-clock power.
Baseload Reliability: Unlike solar and wind, nuclear doesn’t flinch at nightfall or cloudy skies. That makes it ideal for powering critical infrastructure—especially AI, which can’t afford downtime.

🧪 Next-Gen Tech on the Horizon

Small Modular Reactors (SMRs): These compact units promise faster deployment, lower costs, and safer operation. China and Russia already have some online.
Fusion Dreams: Still experimental, but if cracked, fusion could offer near-limitless clean energy. It’s the holy grail—though still more sci-fi than supply chain.

⚖️ The Catch? Cost, Waste, and Public Trust

Nuclear remains expensive to build and politically fraught. Waste disposal and safety concerns haven’t vanished, and public opinion is split—especially in the UK.
Even with advanced designs, the spectres of Chernobyl and Fukushima linger in the cultural memory. That’s a narrative hurdle as much as a technical one.

🛰️ Moonshots and Geopolitics

NASA’s push to deploy a nuclear reactor on the moon by 2029 underscores how strategic this tech has become—not just for Earth, but for space dominance.
The U.S.–China race isn’t just about chips anymore. It’s about who controls the energy to power them.

Nuclear is staging a comeback—not as a relic of the past, but as a potential backbone of the future.

Whether it becomes the dominant force or a transitional ally depends on how fast we can build, how safely we can operate, and how wisely we choose to deploy.

🌍 How ‘clean’ is green?

According to MIT’s Climate Portal, no energy source is perfectly clean. Even solar panels, wind turbines, and nuclear plants come with embedded emissions—from mining rare metals to manufacturing components and transporting them.

So, while they don’t emit greenhouse gases during operation, their setup and maintenance do leave a footprint.

How CLEAN is GREEN? Explainers | MIT Climate Portal

⚖️ Lifecycle Emissions Comparison

Here’s how different sources stack up in terms of CO₂ emissions per kilowatt hour:

Energy Source	CO₂ Emissions (g/kWh)	Notes
Coal	~1,000	Highest emissions, plus toxic byproducts
Natural Gas	~500	Cleaner than coal, but still fossil-based
Solar	<50	Mostly from manufacturing panels
Wind	~10	Lowest emissions, mostly from materials
Nuclear (SMR/SNR)	~12–20	Low emissions, but waste and safety debates linger

Source: MIT Climate Portal

October 16, 2024September 12, 2025

Big tech companies are increasingly adopting nuclear power to meet the high energy demands of their AI data centres

Why?

Elevated Energy Needs

AI systems, particularly generative AI, necessitate substantial computational power, leading to significant energy use. Conventional energy sources might not meet these growing demands.

Environmental Commitments

Numerous tech firms have pledged to lower their carbon emissions. Nuclear power, a low-emission energy source, supports these environmental commitments.

Dependability

Nuclear energy offers a consistent and uninterrupted power supply, essential for data centres that operate around the clock.

Technological Advancements

Progress in nuclear technologies, such as small modular reactors (SMRs), has enhanced the feasibility and appeal of nuclear power for extensive use.

For example, Google has entered into an agreement with Kairos Power for electricity from small modular reactors to bolster its AI operations. In a similar vein, Microsoft has collaborated with Constellation to refurbish an inactive reactor at the Three Mile Island nuclear facility.

These collaborations mark a notable transition in the energy strategies of the tech sector, as they pursue dependable, eco-friendly, and robust power solutions to support their AI initiatives.

August 29, 2024September 12, 2025

Company says it can cut data centre energy use by 50% as AI boom places increased strain on power grids

Major technology corporations such as Microsoft, Alphabet, and Meta are channelling billions into data centre infrastructures to bolster generative AI, which is causing a spike in energy demand.

Sustainable Metal Cloud has announced that its immersion cooling technology is 28% less expensive to install compared to other liquid-based cooling methods and can cut energy use by up to 50%.

The surge in artificial intelligence has increased the need for more robust processors and the energy to cool data centres.

This presents an opportunity for Sustainable Metal Cloud, which runs ‘sustainable AI factories’ consisting of HyperCubes located in Singapore and Australia.

These HyperCubes house servers equipped with Nvidia processors immersed in a synthetic oil known as polyalphaolefin, which is more effective at dissipating heat than air. The company claims this technology can reduce energy consumption by as much as 50% when compared to the conventional air-cooling systems found in most data centres.

Additionally, the Singapore-based company states that its immersion cooling technology is more cost-effective to install by 28% than other liquid cooling options. The HyperCubes are modular and can be integrated into any data centre, utilising spaces that are currently unoccupied within existing facilities.

What is a Hypercube?

Structure: A hypercube topology connects nodes in a way that each node is connected to others in a manner similar to the geometric hypercube. For example, in a 3-dimensional hypercube (a cube), each node is connected to three other nodes.
Scalability: This structure allows for efficient scaling. As the number of dimensions increases, the number of nodes that can be connected grows exponentially.
Fault Tolerance: Hypercube networks are known for their robustness. If one connection fails, there are multiple alternative paths for data to travel, ensuring reliability.

Benefits in data centres

High Performance: The multiple pathways in a hypercube network reduce latency and increase data transfer speeds, which is crucial for big tech companies handling vast amounts of data.
Efficient Resource Utilisation: The topology allows for better load balancing and resource allocation, optimising the performance of data centres.
Flexibility: Hypercube networks can easily adapt to changes in the network, such as adding or removing nodes, without significant reconfiguration.
Big Tech Companies: Companies like Google, Amazon, and Microsoft likely use hypercube topologies in their data centres to ensure high performance and reliability.
High-Performance Computing (HPC): Hypercube networks are also used in supercomputers and other HPC environments where efficient data transfer is critical.

May 17, 2024September 12, 2025

“All the batteries currently on Earth store just 10 minutes of the world’s electricity needs.”

It’s electric!

April 24, 2024September 12, 2025

AI energy consumption is shocking!

Powering artificial intelligence (AI) models takes a substantial toll on our planet’s energy resources.

Delving deeper into AI, it becomes crucial to comprehend the environmental impact of this technological revolution.

Current trends

A new peer-reviewed study featured in ‘Joule‘ highlights the significant energy requirements of AI. The research, carried out by Alex de Vries, a data scientist at the Dutch central bank, provides a quantification of the energy usage linked to the trends in AI capacity and adoption.

The energy appetite of AI

The AI industry is experiencing rapid growth as major technology companies incorporate AI-driven services into their platforms. These applications require significantly more power than traditional ones, resulting in online interactions that are more energy-intensive.

Projected impact

Continuing on the present course, NVIDIA could be dispatching 1.5 million AI server units each year by 2027. If these servers were to run at maximum capacity, they would consume a minimum of 85.4 terawatt-hours of electricity annually. For comparison, this amount of energy surpasses the yearly consumption of numerous small nations.

Comparisons

By 2027, it is projected that global AI-related electricity consumption may rise by 85 to 134 terawatt-hours (TWh) annually. This estimate is on par with the yearly electricity requirements of nations such as the Netherlands, Argentina, and Sweden.

Why sustainability matters

While AI heralds significant breakthroughs, its sustainability is a crucial risk factor to consider. Picture Google’s search engine evolving into a ChatGPT-style chatbot, managing nine billion interactions daily. This would cause energy demands to soar, matching the consumption of a nation like Ireland. Although this scenario isn’t immediately likely due to logistical limitations, it highlights the resource-intensive nature of generative AI applications.

As we explore the AI domain, sustainability should not be neglected. Discussing AI’s risks, such as errors and biases, should also include its environmental impact. Innovation must be balanced with responsible energy use for a sustainable future.

Conclusion

In essence, AI’s demand for power is substantial, and the challenge is to leverage its capabilities while reducing its environmental impact. We must proceed with caution to ensure our technological advances do not compromise the health of our planet.

November 23, 2023September 12, 2025

‘With great power there must also come great responsibility!’

Stan Lee 1922 – 2018

Stan Lee was an American comic book writer, editor, publisher, and producer. He rose through the ranks of a family-run business called Timely Comics which would later become Marvel Comics.

**Art illustration of a comic book page.** **‘With great power there must also come great responsibility!’**

October 16, 2023September 12, 2025

Global electricity energy demanded by BIG tech

Many large tech companies are planning to create their own energy supply or source power from 100% renewable generators.

This is mainly because they have high electricity consumption, especially for their data centres, and they want to reduce their carbon footprint and achieve net-zero emissions targets.

BIG tech companies that are generating their own energy or investing in renewable energy projects

Estimated current electricity demand

The global electricity energy demand is the amount of electricity that the world needs in a given day. It can be calculated by multiplying the average global electricity demand in GW by 24 hours. According to the International Energy Agency (IEA), the average global electricity demand in 2020 was about 3 TW or 3 000 GW. This means that the global electricity energy demand in 2020 was about 72000 GWh or 72 TWh per day.

BIG tech companies are generating their own energy or investing in renewable energy projects – how green is it really?

However, this is an average value, and the actual demand may vary depending on the season, time of day, weather, and other factors.

Energy requirement

The global electricity energy demand is expected to increase in the future, as population grows and living standards improve. The IEA projects that the average global electricity demand will reach 3.8 TW or 3 800 GW by 2030 and 5.2 TW or 5 200 GW by 2050 in the Announced Pledges Scenario, which reflects the full implementation of net-zero emissions targets by some countries and regions. This implies that the global electricity energy demand will reach 91 200 GWh or 91.2 TWh per day by 2030 and 124 800 GWh or 124.8 TWh per day by 2050.

Energy sources to change

The sources of electricity generation will also change in the future, as renewable technologies such as solar PV and wind become more dominant and coal use declines. The IEA reports that the main sources of electricity generation in 2020 were coal (34%), natural gas (23%), hydropower (16%), nuclear (10%), wind (8%), solar PV (4%), biofuels and waste (3%), and other renewables (2%). In the Announced Pledges Scenario, renewables in electricity generation rise from 28% in 2021 to about 50% by 2030 and 80% by 2050.

The world counts.

October 2, 2023September 12, 2025

Powering the UK from energy generated in Morocco

The Xlinks Morocco-UK Power Project is a proposal to create a large-scale renewable energy complex in Morocco and feed the electricity to the UK via a long underwater cable.

Key facts

12 million solar panels, 530 wind turbines over 62 square miles.

The project aims to produce 10.5 GW of clean power from solar and wind facilities in Morocco’s Guelmim Oued Noun region. This is equivalent to about 10% of the UK’s electricity demand.
The project also plans to build a 20 GWh/5 GW battery storage facility to ensure a stable and reliable supply of electricity.

The project will use proven high-voltage direct current (HVDC) interconnector technology to transmit the electricity to the UK via a 3,800 km route under the seabed. The cable will connect to two locations in Devon and Wales, each with a capacity of 1.8 GW.
The project will create over 11,000 new green jobs in the UK and Morocco, and contribute to their renewable industrial ambitions. It will also diversify the UK’s energy sources and reduce its dependence on EU interconnectors, LNG imports, and biomass from North America.

The project is seeking a 25-year contract with the UK government to guarantee a fixed electricity price and secure financing for the £20 billion investment.
It hopes to start construction in 2024 and deliver power to the UK by 2028.

Entirely powered by sun and wind

The Xlinks Morocco-UK Power Project will be a new electricity generation facility entirely powered by solar and wind energy combined with a battery storage facility. Located in Morocco’s renewable energy rich region of Guelmim Oued Noun, it will be connected exclusively to Great Britain via 3,800km HVDC sub-sea cables.

Zero carbon power generation

When domestic renewable energy generation in the United Kingdom drops due to low winds and short periods of sun, the project will harvest the benefits of long hours of sun in Morocco alongside the consistency of its convection Trade Winds, to provide a firm but flexible source of zero-carbon electricity.

July 28, 2023September 12, 2025