As artificial intelligence (AI) continues to transform industries and unlock new opportunities, its environmental impact is also a matter of concern. While AI holds immense potential to combat climate change, it paradoxically contributes to the problem it aims to solve. The computational intensity of AI training and deployment leaves a significant carbon footprint. So, what’s the responsible way to savor the benefits of AI without worsening the climate crisis? The answer is Green AI.
So, What Is Green AI?
Green AI is a movement and an innovation that seeks to balance technological advancement with environmental sustainability. Green AI, also referred to as Sustainable AI or Net Zero AI, encompasses practices to reduce the carbon footprint of artificial intelligence technologies. Unlike traditional approaches, Green AI integrates sustainability into every stage of the AI lifecycle, from research and development to deployment and maintenance.
Furthermore, understanding the differences between conventional AI and Green AI is key to addressing this growing challenge.
Traditional AI vs. Green AI: A World of Difference
Traditional AI focuses on achieving unmatched accuracy in tasks like language translation, image recognition, and autonomous driving. While its applications are groundbreaking, this accuracy comes at a cost. Training large-scale AI models often require enormous computational resources, consuming vast amounts of energy.
For example, a nature.com study revealed the carbon footprint of training a single big language model is equal to around 300,000 kg of carbon dioxide emissions. This could be quantified as equivalent to 125 round-trip flights between New York and Beijing, a quantification that laypersons can visualize.
Thus, conventional AI overlooks energy efficiency. It also increases costs for businesses and excludes smaller players from entering the AI landscape. The worst outcome is the damage done to the environment from its carbon footprint, suppressing its potential to mitigate climate change.
In contrast, Green AI prioritizes energy-efficient practices. By focusing on sustainable development and deployment of AI systems, it seeks to minimize environmental harm without compromising innovation. Green AI introduces efficiency as a key metric alongside accuracy. It also advocates solutions that deliver high performance while conserving resources.
AI Powering Innovation but at What Cost?
We projected this study from ScienceDirect to understand the energy appetite of AI solutions. AI is growing rapidly, with bigger data needs and more complex models. However, this doesn’t always lead to equally big improvements in accuracy. While large language models (LLMs) like ChatGPT drive innovation, they come with significant environmental costs. Let’s dig deeper…
AI’s Growing Energy Appetite
The same report explains training GPT-3, for instance, consumed 1287 MWh of electricity and emitted 550 tons of carbon dioxide—comparable to flying 33 times between Australia and the UK.
The energy required for AI isn’t just during training. Using systems like GPT-3 also carries a hefty price. In January 2023 alone, GPT-3 processed 590 million queries, consuming energy equivalent to that of 175,000 people. On a smaller scale, each ChatGPT query uses as much power as running a 5W LED bulb for over an hour.
Fig: CO2 equivalent emissions for training ML models (blue) and of real-life cases (violet). In brackets, the billions of parameters adjusted for each model.
Source: ScienceDirect
Deloitte’s recent report, “Powering Artificial Intelligence: A study of AI’s environmental footprint”, revealed the following findings:
- Between 2021 and 2022, data centers accounted for 98% of Meta’s additional electricity use and 72% of Apple’s between 2022 and 2023.
- AI adoption will fuel data center power demand, likely reaching 1,000 terawatt-hours (TWh) by 2030, and potentially climbing to 2,000 TWh by 2050.
- This will account for 3% of global electricity consumption, indicating faster growth than in other uses like electric cars and green hydrogen production.
AI Data Centers: Energy Efficient or Energy Waste?
Data centers are the backbone of AI training and deployment, often referred to as the “cloud.” However, they rely on physical infrastructure for computing, processing, storing, and exchanging data. They require massive power and contribute heavily to the energy consumption of tech companies.
Different types of data centers have unique energy demands. Basic computer rooms handle simple tasks, while mid-size and large-scale enterprise data centers manage more complex operations. Hyperscale data centers, owned by tech giants have maximum hardware density and handle massive computational workloads, consuming the most energy.
Within this category, AI hyperscale data centers are emerging as a distinct segment. These centers are specifically built for generative AI and machine learning tasks, requiring high-performance GPUs for model training and inference.
This results in higher server power usage and the need for advanced cooling systems, further increasing energy consumption. Smaller data centers often lack the capacity for these high-demand workloads, driving the growth of AI-focused hyperscale facilities.
Fig: Data centers’ electricity consumption by server type and scenarios
But as they expand, a critical question remains: How sustainable are AI hyperscale data centers in the fight against climate change?
Well, this is where the demand for Green AI garners importance.
Why Green AI Matters?
The environmental cost of AI is no longer a hypothesis, it is palpable all around. Even blockchain technologies like cryptocurrency mining have demonstrated how unchecked digital innovation can lead to unsustainable energy consumption.
Coming straight to the topic, Green AI holds the promise of reversing this trend. For example, AI-powered tools can optimize supply chains, reduce waste, and improve energy grid efficiency. If developed responsibly, AI could become the key driving force behind the global effort to achieve carbon neutrality.
Thus, by combining innovation with sustainability, Green AI can meet the growing demand for computational power while reducing its impact on the environment.
Core Principles of Green AI
This means leveraging AI solutions that are not only effective in optimizing energy use in applications but are also inherently low-energy consumers. It’s crucial to balance AI’s benefits with its environmental impact. It means AI should support sustainability goals and not worsen the problems that it aims to solve.
Energy Efficiency
Green AI encourages the design of algorithms and models that consume less energy. Researchers can achieve this by developing lightweight models or installing techniques like pruning, quantization, and model distillation, which reduce computational requirements.
Hardware Optimization
Using energy-efficient hardware, such as GPUs with higher FLOPS per watt or specialized Tensor Processing Units (TPUs), can significantly cut AI’s energy consumption. Parallelizing tasks across multiple cores also helps reduce training times and emissions, though excessive cores may increase energy use disproportionately.
Another technique is edge computing which means processing data locally to avoid energy-intensive transmissions to cloud or data centers and optimizing resources for IoT (The Internet of Things) devices. Together, these strategies enable powerful AI performance with a smaller environmental footprint.
Data Center Optimization
Adopting renewable energy sources for powering data centers and AI operations is a significant milestone of Green AI. Companies like Google and Microsoft are already leading the charge by transitioning their cloud services to run on clean energy.
To make data centers more energy-efficient, researchers have created algorithms and frameworks that balance server loads, optimize cooling systems, and allocate resources more effectively. All these processes are included in data center optimization that cuts down energy use and emissions.
Transparency and Accessibility
Green AI promotes transparency in reporting the environmental costs of AI projects. Standardized metrics for energy consumption and emissions can help developers and organizations make informed decisions about their AI strategies.
Some of the tools that are used to estimate the carbon footprint of AI technologies are CarbonTracker, CodeCarbon, Green algorithms, and PowerTop.
Additionally, by lowering computational barriers, Green AI fosters inclusivity. Smaller organizations and researchers gain access to advanced tools without burdening themselves with high environmental and financial costs.
Fig: Achievable electricity demand reduction through energy savings, “High adoption” scenario
Policies Driving Green AI
The United Nations’ Sustainable Development Goals (SDGs) highlight the need for a sustainable future. Goals like Affordable and Clean Energy and Industry, Innovation, and Infrastructure are driving the rise of Green AI. Industry leaders are rethinking data center designs and operations to lower energy consumption and environmental impacts. This shows their eagerness to demonstrate proactive efforts toward sustainability.
While Green AI initiatives are mostly industry-led, some regions are implementing supportive policies. These range from monitoring low-impact data centers to stricter regulations for areas where grid stability is at risk. Thus, balancing these policies can encourage sustainable practices without moving operations to less regulated regions.
Notable policies include:
- European Code of Conduct for Data Centers (EU DC CoC)
- Energy Efficiency Directive (EED)
- Singapore Green Data Centre Roadmap
China has also introduced measures like the Three-Year Action Plan on New Data Centres, while the U.S. lacks federal-level regulations specific to data centers.
Policymakers can amplify these efforts by co-developing standards with industry leaders. Collaborative strategies ensure data centers meet climate goals without compromising growth or grid stability.
Green AI demonstrates that with the right policies and innovations, the tech industry can lead the way to a more sustainable future.
Green AI Takes the Spotlight at COP29
As world leaders convened in Baku, Azerbaijan, for COP29, discussions pointed to the role of AI in promoting environmental sustainability. A Deloitte-hosted panel brought together experts from NVIDIA, Crusoe Energy Systems, EON, and the International Energy Agency (IEA) to explore strategies for reducing AI’s environmental footprint.
Josh Parker, senior director of legal–corporate sustainability at NVIDIA, said,
“We see a very rapid trend toward direct-to-chip liquid cooling, which means water demands in data centers are dropping dramatically right now.”
According to NVIDIA, designing data centers while keeping energy efficiency at the highest priority right from the beginning is very much essential. As AI demands grow, sustainable infrastructure will be critical. Parker highlighted that current data centers are becoming outdated and inefficient.
He added, accelerated computing platforms are 10X more efficient than traditional systems for running workloads. This creates a significant opportunity to cut energy consumption in existing infrastructures.
Accelerated Computing: A Path to Green AI
Parker once again emphasized that accelerated computing represents the most energy-efficient platform for AI and many other applications. Over the past few years, energy efficiency for accelerated computing has improved dramatically, with a 100,000x reduction in energy consumption.
- In just the last two years, energy use for AI inference tasks dropped by 96%, with systems becoming 25x more efficient for the same workload.
Accelerated computing uses GPUs to process tasks faster and more efficiently than traditional CPUs. By handling multiple tasks simultaneously, GPUs reduce the energy required for AI workloads. It’s one of the techniques that come under hardware efficiency and data center optimization.
Furthermore, NVIDIA emphasized the need for energy-efficient infrastructure in data centers. Innovations like liquid-cooled GPUs are transforming cooling methods. Unlike traditional air conditioning, direct-to-chip liquid cooling consumes less power and water while maintaining effective temperature control.
The Bottom Line
Deloitte’s findings have adeptly showcased AI’s potential in driving climate-neutral economies. Green AI strategies focus on minimizing environmental impact by improving hardware design and increasing the use of renewable energy.
Industry leaders are spearheading these efforts, highlighting the effectiveness of sustainable computing practices. The shift toward accelerated computing and energy-efficient design is paving the way for AI to support global climate goals.
As we face a climate crisis, the integration of Green AI principles is no longer optional—it is essential. By redefining how AI solutions are developed, we can harness their power for good while minimizing their environmental toll. The road ahead demands collective effort, innovation, and accountability. Last but not least, Green AI is not just a technological imperative but a moral responsibility to ensure a greener future.
Key Sources:
- A review of green artificial intelligence: Towards a more sustainable future – ScienceDirect
- AI at COP29: Balancing Innovation and Sustainability | NVIDIA Blog
The post Green AI Explained: Fueling Innovation with a Smaller Carbon Footprint appeared first on Carbon Credits.
The earnings calls that quietly reframed climate from sustainability question to operating risk.
Three earnings calls in the last 18 months tell the story without any help from a press release.
Hershey, May 2024: cocoa price exposure compresses margin, and the company attributes part of the cost shock to West African weather. Olam, July 2024: coffee climate exposure quantified in the annual report. JBS, January 2025: supply chain climate disclosures expanded materially in response to investor pressure and regulatory expectation. None of these companies issued the announcement as climate news. They issued it as financial news. The climate-linked supply chain risk did not arrive with a sustainability framing; it arrived as a P&L line.
You are probably reading this article because you suspect the same thing is happening to your business. This piece walks through what is showing up on which earnings calls, how procurement and finance leaders are quantifying the exposure, and what serious corporates are doing about it before the regulator asks.
Where climate risk has already appeared in earnings
The pattern is consistent across resource-intensive sectors. A weather event compresses supply, the price spikes, the cost flows through the income statement, and the analyst on the call asks whether the event is anomalous or structural. Increasingly, the honest answer is the second one.
Cocoa is the cleanest example. The 2023 to 2024 West African harvest fell sharply on the back of erratic rainfall and disease. Cocoa futures more than tripled. Companies with concentrated West African sourcing absorbed the cost; companies with diversified sourcing absorbed less. The exposure was not climate as ESG topic. It was climate as cost of goods.
Coffee follows the same pattern. Brazilian and Vietnamese harvests have moved on weather more sharply across the last several seasons. Roasters with long-tenor supplier relationships and origin diversification have managed the volatility; roasters with spot-market exposure have not. Wheat, sugar, palm oil, beef: the same dynamic in different commodities, a pattern the IPCC AR6 Working Group II report projects will intensify across agricultural systems through mid-century.
What this means: climate risk is no longer a footnote in the 10-K. It is a line item the CFO has to explain on the call.
The three commodity exposures that hit margin first
For most companies with material Scope 3 exposure, three exposures dominate the near-term P&L risk.
- Concentrated single-origin sourcing in a climate-vulnerable region. If your tier-one supply for any material commodity sits in one geography, you have a concentration risk that climate amplifies. Diversification across origins is the obvious hedge, but it takes years to build and requires relationships you cannot acquire by tender.
- Supplier financial fragility under climate stress. Smallholder farmers, who supply a large share of the global cocoa, coffee, and palm oil market, do not carry the balance sheets to absorb yield shocks. When yields collapse, they exit. When they exit, your supply base shrinks, and the surviving suppliers raise prices. The risk is structural, not cyclical.
- Logistics and storage exposure to extreme weather. Hurricane disruptions to Gulf shipping, drought-driven Panama Canal restrictions, flooding in European inland waterways: each of these has moved input costs in the last three years, a pattern documented in Munich Re’s natural catastrophe data. The exposure shows up as a one-quarter event in the financial press but accumulates over time on the cost line.
TCFD and ISSB disclosure changes
The disclosure architecture has now caught up with the risk. The Task Force on Climate-related Financial Disclosures, whose recommendations are now embedded in the ISSB’s IFRS S2 climate standard, requires companies to disclose climate-related risks across physical and transition categories, with quantification where possible.
For physical risk specifically (the climate-linked supply chain risk you are reading about), the disclosure must address both acute exposures (extreme weather events) and chronic exposures (gradual changes in temperature, precipitation, and growing seasons). The disclosure must address the time horizon over which the risk is material, the parts of the value chain exposed, and the financial impact under different scenarios.
The CSRD imposes similar requirements under European law, with double materiality (both financial and impact materiality) embedded in the assessment. The practical effect: your auditors and your investor relations team now need a defensible answer to the climate-linked supply chain risk question, and the answer needs to be quantified.
What procurement and finance can do now
Three actions matter near-term.
Map your exposure. Most companies do not have a clear view of which tier-one and tier-two suppliers sit in which climate-vulnerable geographies. Without the map, you cannot quantify the risk, and without the quantification, you cannot disclose it credibly. The map is the foundation, and World Resources Institute climate risk research provides useful public tooling to start.
Diversify and deepen, in that order. Diversification across origins reduces concentration risk, but the deeper move is to invest in the resilience of the suppliers you already have. Regenerative practices, agroforestry, soil health interventions: these reduce yield volatility under climate stress and protect your input cost trajectory.
Embed the climate spend inside procurement, not outside it. Treating climate risk as a sustainability cost line subordinates it to the ESG budget. Treating it as a procurement and resilience investment puts it in the budget that matters, which is the cost-of-goods budget that the CFO defends quarterly.
Nature-based supply chain investments are the asset class designed for exactly this purpose. They sit inside the value chain, they reduce climate-linked supply risk, they generate verifiable Scope 3 reductions, and they produce the documentation an auditor and a regulator can both test.
If you are quantifying climate-linked supply chain risk in advance of the next earnings cycle or the next disclosure period, the carbon and sustainability experts at Carbon Credit Capital can help you map your exposure and structure a Dual-Value Model response that addresses reduction, resilience, and disclosure-readiness in a single program. Schedule a consultation.
More and more small and medium-sized businesses are hearing the same question from their larger customers: What is your carbon footprint? That question now travels down entire supply chains, and it arrives next to tender requirements, certification criteria, and rising customer expectations.
The accounting differences that decide whether your nature investment shows up in inventory, in BVCM, or nowhere at all.
The question reaches a procurement team about three weeks before the next sustainability committee meeting. Someone has read about insetting. Someone else has just signed off on an offset purchase. The CSO wants to know if the two are interchangeable. The answer is no, and the GHG Protocol Land Sector and Removals Standard is the reason why.
This article walks through what each term means at audit-grade specificity, what the standards actually say about how each gets counted, and how to decide which tool fits which target. The insetting vs offsetting question is one of the most-searched in corporate climate strategy, and one of the most poorly answered. By the end of this piece, you should be able to brief a committee on the difference without notes.
The two definitions, in plain English
Offsetting means buying carbon credits generated outside your value chain and retiring them against your residual emissions. The reduction happens somewhere else, financed by you, and the credit is the receipt.
Insetting means investing in emission reductions or removals inside your own value chain, typically with suppliers, where the reduction is directly linked to the products and services you buy. The reduction happens inside the boundary of your Scope 3 inventory, and the accounting treatment is fundamentally different.
The shorthand from the University of Oxford’s Nature-based Insetting Initiative is useful: insetting is what you do with the supply chain you have; offsetting is what you do with the supply chain you do not have.
What the GHG Protocol Land Sector Standard actually says
The GHG Protocol Land Sector and Removals Standard, finalised in 2024 after a multi-year pilot, sets the rules for how land-based emission reductions and removals enter corporate inventories. The Standard distinguishes between inventory accounting (Scope 1, 2, and 3) and project or intervention accounting (a separate methodology for crediting).
For insetting, the practical implication is that supplier-level interventions, when properly measured and attributed, can reduce your Scope 3 category 1 (purchased goods and services) emissions in your inventory. The reduction is not a credit retired against the inventory; it is a lower inventory number, period.
For offsetting, the credit is retired separately. It can be reported as a contribution toward a net-zero claim under the SBTi Beyond Value Chain Mitigation framework or as part of a VCMI Carbon Integrity claim, but it does not lower the inventory number.
A practical consequence: if your Science Based Target requires a 50% absolute reduction in Scope 3 emissions by 2030, insetting moves you toward the target. Offsetting does not. This single point of difference reshapes the procurement decision.
When insetting counts toward Scope 3 (and when it does not)
Insetting counts toward Scope 3 only when several conditions are met:
- The intervention must occur with an entity in your value chain.
- The emissions reduction or removal must be measured against a defensible baseline.
- The reduction must be attributed to your share of that supplier’s output, not double-counted with other buyers.
- It must follow the inventory accounting rules in the GHG Protocol Land Sector Standard, not the project accounting rules used to generate credits.
The most common failure mode is double counting. If your supplier sells the same reduction as a credit on the voluntary market and also reports it to you as a Scope 3 reduction, the math breaks. The Standard requires you to address this risk, typically by purchasing and retiring the supplier-issued credit as part of your inventory or by contractual provisions that prevent the supplier from selling the reduction twice.
When insetting does not count toward Scope 3: when the intervention sits with a supplier you do not buy from, when the baseline is not defensible, when the attribution is unclear, or when the documentation does not survive audit. Those cases default to Beyond Value Chain Mitigation, which is still useful but operates on a different ledger.
The procurement and supplier engagement question
Insetting is harder than offsetting. That is the unfashionable truth most buyers eventually confront. Offsetting is a transaction; insetting is a relationship.
To run an insetting program, you need supplier mapping precise enough to know which farms or facilities sit at which Scope 3 boundary. You need an engagement model that gets suppliers to participate, which usually requires multi-year commitments and shared economics. You need an MRV architecture that measures the right things and produces audit-ready documentation. And you need a contractual structure that prevents double counting and protects both sides.
The trade-off you receive in return is significant. Reductions count against your inventory rather than your residual. Supplier relationships deepen, which protects sourcing continuity. Yield and quality improvements often follow regenerative interventions, which reduces your input cost over time. And the regulatory file, under CSRD, CSDDD, EUDR, and the SBTi FLAG Guidance, is materially stronger.
Choosing the right tool for the right target
A practical decision rule. If your target is a science-based Scope 3 reduction and you operate in a FLAG sector or source FLAG commodities, insetting is the structurally correct tool. If your target is a net-zero claim that includes neutralising hard-to-abate residual emissions outside your value chain, BVCM via high-integrity offsets is the structurally correct tool. Most companies with material Scope 3 exposure need both, in different proportions, sequenced over time.
The sequencing matters. Insetting takes longer to stand up but produces a permanent reduction in the inventory. Offsetting can be transacted faster but does not change the inventory and now sits under tighter claim restrictions. Treat them as complementary tools with different jobs, not as substitutes. The Accountability Framework Initiative and the IUCN Global Standard for Nature-based Solutions both provide useful guardrails for the insetting side, with biodiversity, human rights, and benefit-sharing requirements that go beyond carbon math.
If you are mapping a Scope 3 reduction roadmap and need to scope which interventions count toward your inventory versus which sit in Beyond Value Chain Mitigation, the carbon and sustainability experts at Carbon Credit Capital can help you structure a nature-based supply chain investment program that fits your FLAG exposure, your target architecture, and your audit horizon. Schedule a consultation.
-
Climate Change1 year ago
Guest post: Why China is still building new coal – and when it might stop
-
Greenhouse Gases1 year ago
Guest post: Why China is still building new coal – and when it might stop
-
Greenhouse Gases2 years ago嘉宾来稿：满足中国增长的用电需求 光伏加储能“比新建煤电更实惠”
-
Climate Change2 years ago嘉宾来稿：满足中国增长的用电需求 光伏加储能“比新建煤电更实惠”
-
Climate Change2 years ago
Bill Discounting Climate Change in Florida’s Energy Policy Awaits DeSantis’ Approval
-
Renewable Energy10 months agoSending Progressive Philanthropist George Soros to Prison?
-
Greenhouse Gases1 year ago
嘉宾来稿：探究火山喷发如何影响气候预测
-
Carbon Footprint2 years agoUS SEC’s Climate Disclosure Rules Spur Renewed Interest in Carbon Credits