## COMMODITy SPECIAL FEATURE MARKET DEVELOPMENTS AND ThE IMPACT OF AI ON ENERGy DEMAND

## Source details

**Canonical URL:** [COMMODITy SPECIAL FEATURE MARKET DEVELOPMENTS AND ThE IMPACT OF AI ON ENERGy DEMAND](https://www.imf.org/-/media/files/publications/weo/2025/april/english/commodityspecialfeature.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/weo/2025/april/english/commodityspecialfeature.pdf.md)
- [Structured JSON version](/-/media/files/publications/weo/2025/april/english/commodityspecialfeature.pdf.json)

---

### Commodity market developments — summary of recent trends and risks
- Primary commodity prices increased 1.9 percent between August 2024 and March 2025, driven by natural gas, precious metals, and beverage prices.
- Oil
  - Oil prices declined 9.7 percent between August 2024 and March 2025.
  - Futures markets indicate oil prices will average $66.9 per barrel in 2025, a 15.5 percent decline, before falling to $62.4 in 2026.
  - Drivers of decline: trade war fears, strong non-OPEC+ supply growth, unwinding of OPEC+ cuts, sluggish Chinese demand, rising EV penetration.
  - Russia’s oil traded at a $5–$15 discount to Brent after sanctions (harshest sanctions imposed on January 10, 2025); sanctions have not materially disrupted oil flows.
  - Outlook risks: upside from supply disruptions or trade de-escalation; downside from trade war escalation and additional OPEC+ production increases.
- Natural gas
  - TTF prices in Europe rose 7.7 percent between August 2024 and March 2025 to $13.1 a MMBtu.
  - Henry Hub prices doubled over the same period due to harsh weather and higher exports demand.
  - Futures as of April 4: TTF prices average $12.5 a MMBtu in 2025, decreasing to $7.8 a MMBtu in 2030; Henry Hub prices expected to decline from $4.0 a MMBtu in 2025 to $3.3 a MMBtu in 2030.
  - Risks to outlook described as balanced.
- Metals
  - IMF metals price index increased 11.2 percent between August 2024 and March 2025, driven mainly by gold, aluminum, and copper.
  - Aluminum rose 12.7 percent; copper rose 8.4 percent over the same period.
  - Futures predict price declines by end-2026: aluminum −5.7 percent, copper −4.5 percent, iron ore −14.3 percent.
  - Gold set new records, recently surpassing $3,000 per ounce.
- Agricultural commodities
  - IMF food and beverages price index increased 3.6 percent between August 2024 and March 2025, driven by beverages.
  - Cereal prices increased 0.6 percent; coffee prices jumped 33.8 percent (IMF coffee index reached historic highs in February); rice prices fell 26.0 percent.
  - New trade barriers announced April 2 had heterogeneous effects: income-elastic and trade-sensitive crops (coffee, soybeans) declined sharply; staples like corn and wheat less affected.
  - Upside risks: trade disruptions and adverse weather. Downside risks: larger-than-expected harvests, trade war intensification, broader uncertainty.

### Power Hungry: framing, questions, and modeling approach
- Framing and key questions
  - Rapid development/adoption of generative AI (including large language models) requires more data centers with high electricity consumption.
  - Cost structure of large language models: large fixed cost for training + variable costs for operation and responses; electricity is a critical input.
  - Example scale: northern Virginia server warehouse square footage roughly equivalent to the floor space of eight Empire State Buildings (Cushman & Wakefield 2024).
  - Research objectives using IMF-ENV (Chateau and others 2025) CGE model:
    - How fast have AI-related sectors grown and how has their electricity consumption changed?
    - How does projected AI electricity demand by 2030 compare with other drivers (e.g., EVs)?
    - What are impacts on energy prices and electricity mix under alternative policy scenarios?
    - What is the impact of data center growth on carbon emissions?
- Modeling parameterization
  - IMF-ENV captures AI impact via an IT-sector TFP increase in China, the United States, and Europe to match expected data center power demand growth between 2025 and 2030.
  - Projected TFP growth rates: China 22 percent (annual), United States 13 percent (annual), Europe 10 percent (annual) (JP Morgan 2024; McKinsey & Company 2024a, 2024b).
  - Three scenarios simulated:
    1. Baseline: excludes AI-related TFP shock; reflects energy and emissions policies through 2024.
    2. AI under current energy policies: includes AI-related TFP shock; assumes electricity generation composition remains as in baseline.
    3. AI under alternative energy policies: includes AI-related TFP shock; increases share of renewables via feed-in tariffs aligned with regions’ long-term strategies.
  - Results for AI scenarios are reported as deviations from the baseline unless stated otherwise.

### The growing macroeconomic relevance of AI-producing sectors
- US AI-producing sectors’ value added quadrupled from $278 billion (constant 2017 dollars) to $1.13 trillion between 2010 and 2023.
- Share of these sectors in US GDP rose from 2.4 percent in 2013 to 3.5 percent in 2023; data-processing sector nearly doubled its share over the same period.
- Manufacturing share declined by 1.5 percentage points (2013–2023).
- Productivity drivers: value added per employee in data-processing grew about four times faster than that in the whole economy over the past 10 years; growth driven largely by elevated investment in physical capital and complementarities of intermediate inputs rather than labor or TFP alone.

### AI’s demand for electricity — current estimates and projections
- Electricity costs as share of total costs:
  - Data center companies: 13–15 percent.
  - Semiconductor firms and AI service companies: 0.8–1.5 percent (but nearly doubled in less than five years).
- Global electricity consumption from data centers and AI estimated at 400–500 terawatt-hours (TWh) in 2023 (more than double 2015 level).
- US projected electricity demand from data centers: 178 TWh in 2024 to 606 TWh in 2030 under a medium-demand scenario.
- By 2030, AI-driven global electricity consumption could reach 1,500 TWh — comparable to India’s current total electricity consumption.
- Comparative magnitude: projected AI electricity demand by 2030 is about 1.5 times higher than expected demand from EVs.

### Effects of increased electricity demand from AI — supply, prices, GDP, and emissions
- Electricity supply increases (AI scenario under current energy policies, relative to baseline, by 2030):
  - United States: 8 percent (525 TWh).
  - Europe: 3 percent (145 TWh).
  - China: 2 percent (237 TWh).
- AI scenario under alternative energy policies keeps total supply increases identical but shifts composition toward renewables:
  - Solar and wind generation offsets about 166 TWh in China, 58 TWh in the United States, and 35 TWh in Europe, largely replacing coal in China and natural gas in the US.
- Electricity price impacts (AI scenario under current energy policies, point estimates):
  - Price increases of 0.9 percent in the United States, 0.45 percent in Europe, and 0.35 percent in China.
- Potential higher price pressures if renewables scale-up slows or transmission and distribution investments lag:
  - Price increases could escalate up to 5.3 percent in China, 8.6 percent in the United States, and 3.6 percent in Europe by 2030 in the AI scenario under current energy policies.
- Macroeconomic spillovers:
  - Without further transmission and distribution investments, electricity reallocation toward AI could reduce annual value-added growth in energy-intensive manufacturing in the United States by an average of 0.3 percentage point relative to the baseline, lowering annual GDP growth by 0.1 percentage point.
  - AI scenario under alternative energy policies yields more muted electricity price increases due to feed-in tariffs lowering generation prices for solar and wind.
- Greenhouse gas emissions:
  - AI scenario under current energy policies increases 2030 GHG emissions by 5.5 percent in the US, 3.7 percent in Europe, and 1.2 percent in China; global average increase of 1.2 percent.

### Emission impacts, social costs, and macroeconomic outcomes (Section 2)
- Emissions totals and social costs
  - Cumulative global GHG emissions increase of 1.7 gigatons (Gt) between 2025 and 2030 under the AI scenario with current energy policies.
  - In the AI scenario under alternative energy policies, cumulative global GHG emissions increase is limited to 1.3 Gt by 2030, which is 24 percent less than in the AI scenario under current energy policies.
  - The additional social cost of 1.3 to 1.7 Gt of carbon-dioxide-equivalent emissions is about $50.7 billion to $66.3 billion, using a median social cost of carbon estimate of $39 per ton based on 147 published studies with more than 1,800 estimates (Moore and others 2024).
  - The additional social cost represents about 1.3 percent to 1.7 percent of the AI-driven increase in real world GDP between 2025 and 2030.
- Macroeconomic impacts (GDP and sectoral growth)
  - In the AI scenario under current energy policies, the AI shock raises the average annual growth rate of global GDP by 0.5 percentage point between 2025 and 2030.
  - This estimate is in line with previous IMF estimates ranging between 0.1 percentage point and 0.8 percentage point (April 2024 World Economic Outlook).
  - Gains are greater in countries where the projected growth rate of the IT sector and its relative importance in the economy are higher.
  - In the AI scenario under alternative energy policies, growth gains are slightly reduced because of feed-in tariff policies.
  - Total fiscal costs of these tariffs range from 0.3 percent to 0.6 percent of GDP across countries and are financed through increased lump-sum taxes, which slightly reduce household consumption.
  - Despite those fiscal costs, growth benefits from AI expansion far outweigh the costs, resulting in similar average annual GDP growth across both scenarios.

### Distributional, investment, and policy implications
- Distributional and investment implications
  - Economic benefits from AI are likely to be unevenly distributed across countries and among different groups within societies, potentially exacerbating existing inequalities.
  - Demand for computing and electricity from AI service producers is subject to wide uncertainty, which may delay energy investments, causing underinvestment and higher prices.
- Policy conclusions and recommendations
  - Despite challenges related to higher electricity prices and GHG emissions, gains to global GDP from AI are likely to outweigh the costs of the additional emissions.
  - Policymakers and businesses must work together to ensure AI achieves its full potential while minimizing societal costs, including addressing electricity supply responsiveness, investment timing, and distributional impacts.
  - Alternative energy policies (feed-in tariffs) can shift generation toward solar and wind, reduce emissions (24 percent lower cumulative GHG increase to 2030 compared with current policies), and mute electricity price increases, though they entail fiscal costs ranging from 0.3 percent to 0.6 percent of GDP financed via lump-sum taxes.

*International Monetary Fund, Commodity Special Feature — Market Developments and the Impact of AI on Energy Demand (Sections 1–2).*

### Section 1

### COMMODITy SPECIAL FEATURE MARKET DEVELOPMENTS AND ThE IMPACT OF AI ON ENERGy DEMAND

### Commodity market developments — summary of recent trends and risks
- Primary commodity prices increased 1.9 percent between August 2024 and March 2025, driven by natural gas, precious metals, and beverage prices.
- Oil
  - Oil prices declined 9.7 percent between August 2024 and March 2025.
  - Futures markets indicate oil prices will average $66.9 per barrel in 2025, a 15.5 percent decline, before falling to $62.4 in 2026.
  - Drivers of decline: trade war fears, strong non-OPEC+ supply growth, unwinding of OPEC+ cuts, sluggish Chinese demand, rising EV penetration.
  - Russia’s oil traded at a $5–$15 discount to Brent after sanctions (harshest sanctions imposed on January 10, 2025); sanctions have not materially disrupted oil flows.
  - Outlook risks: upside from supply disruptions or trade de-escalation; downside from trade war escalation and additional OPEC+ production increases.
- Natural gas
  - TTF prices in Europe rose 7.7 percent between August 2024 and March 2025 to $13.1 a MMBtu.
  - Henry Hub prices doubled over the same period due to harsh weather and higher exports demand.
  - Futures as of April 4: TTF prices average $12.5 a MMBtu in 2025, decreasing to $7.8 a MMBtu in 2030; Henry Hub prices expected to decline from $4.0 a MMBtu in 2025 to $3.3 a MMBtu in 2030.
  - Risks to outlook described as balanced.
- Metals
  - IMF metals price index increased 11.2 percent between August 2024 and March 2025, driven mainly by gold, aluminum, and copper.
  - Aluminum rose 12.7 percent; copper rose 8.4 percent over the same period.
  - Futures predict price declines by end-2026: aluminum −5.7 percent, copper −4.5 percent, iron ore −14.3 percent.
  - Gold set new records, recently surpassing $3,000 per ounce.
- Agricultural commodities
  - IMF food and beverages price index increased 3.6 percent between August 2024 and March 2025, driven by beverages.
  - Cereal prices increased 0.6 percent; coffee prices jumped 33.8 percent (IMF coffee index reached historic highs in February); rice prices fell 26.0 percent.
  - New trade barriers announced April 2 had heterogeneous effects: income-elastic and trade-sensitive crops (coffee, soybeans) declined sharply; staples like corn and wheat less affected.
  - Upside risks: trade disruptions and adverse weather. Downside risks: larger-than-expected harvests, trade war intensification, broader uncertainty.

### Power Hungry: How AI will drive energy demand — framing and key questions
- Rapid development/adoption of generative AI (including large language models) requires more data centers with high electricity consumption.
- Cost structure of large language models: large fixed cost for training + variable costs for operation and responses; electricity is a critical input.
- Example scale: northern Virginia server warehouse square footage roughly equivalent to the floor space of eight Empire State Buildings (Cushman & Wakefield 2024).
- Research objectives using IMF-ENV (Chateau and others 2025) CGE model:
  - How fast have AI-related sectors grown and how has their electricity consumption changed?
  - How does projected AI electricity demand by 2030 compare with other drivers (e.g., EVs)?
  - What are impacts on energy prices and electricity mix under alternative policy scenarios?
  - What is the impact of data center growth on carbon emissions?

### The growing macroeconomic relevance of AI-producing sectors — facts and drivers
- US AI-producing sectors’ value added quadrupled from $278 billion (constant 2017 dollars) to $1.13 trillion between 2010 and 2023.
- Share of these sectors in US GDP rose from 2.4 percent in 2013 to 3.5 percent in 2023; data-processing sector nearly doubled its share over the same period.
- Manufacturing share declined by 1.5 percentage points (2013–2023).
- Productivity drivers: value added per employee in data-processing grew about four times faster than that in the whole economy over the past 10 years; growth driven largely by elevated investment in physical capital and complementarities of intermediate inputs rather than labor or TFP alone.

### AI’s demand for electricity — current estimates and projections
- Electricity costs as share of total costs:
  - Data center companies: 13–15 percent.
  - Semiconductor firms and AI service companies: 0.8–1.5 percent (but nearly doubled in less than five years).
- Global electricity consumption from data centers and AI estimated at 400–500 terawatt-hours (TWh) in 2023 (more than double 2015 level).
- US projected electricity demand from data centers: 178 TWh in 2024 to 606 TWh in 2030 under a medium-demand scenario.
- By 2030, AI-driven global electricity consumption could reach 1,500 TWh — comparable to India’s current total electricity consumption.
- Comparative magnitude: projected AI electricity demand by 2030 is about 1.5 times higher than expected demand from EVs.

### Modeling scenarios and parameterization
- IMF-ENV captures AI impact via an IT-sector TFP increase in China, the United States, and Europe to match expected data center power demand growth between 2025 and 2030.
  - Projected TFP growth rates: China 22 percent (annual), United States 13 percent (annual), Europe 10 percent (annual) (JP Morgan 2024; McKinsey & Company 2024a, 2024b).
- Three scenarios simulated:
  1. Baseline: excludes AI-related TFP shock; reflects energy and emissions policies through 2024.
  2. AI under current energy policies: includes AI-related TFP shock; assumes electricity generation composition remains as in baseline.
  3. AI under alternative energy policies: includes AI-related TFP shock; increases share of renewables via feed-in tariffs aligned with regions’ long-term strategies.
- Results for AI scenarios are reported as deviations from the baseline unless stated otherwise.

### Effects of increased electricity demand from AI — supply, prices, GDP, and emissions
- Electricity supply increases (AI scenario under current energy policies, relative to baseline, by 2030):
  - United States: 8 percent (525 TWh).
  - Europe: 3 percent (145 TWh).
  - China: 2 percent (237 TWh).
- AI scenario under alternative energy policies keeps total supply increases identical but shifts composition toward renewables:
  - Solar and wind generation offsets about 166 TWh in China, 58 TWh in the United States, and 35 TWh in Europe, largely replacing coal in China and natural gas in the US.
- Electricity price impacts (AI scenario under current energy policies, point estimates):
  - Price increases of 0.9 percent in the United States, 0.45 percent in Europe, and 0.35 percent in China.
- Potential higher price pressures if renewables scale-up slows or transmission and distribution investments lag:
  - Price increases could escalate up to 5.3 percent in China, 8.6 percent in the United States, and 3.6 percent in Europe by 2030 in the AI scenario under current energy policies.
- Macroeconomic spillovers:
  - Without further transmission and distribution investments, electricity reallocation toward AI could reduce annual value-added growth in energy-intensive manufacturing in the United States by an average of 0.3 percentage point relative to the baseline, lowering annual GDP growth by 0.1 percentage point.
  - AI scenario under alternative energy policies yields more muted electricity price increases due to feed-in tariffs lowering generation prices for solar and wind.
- Greenhouse gas emissions:
  - AI scenario under current energy policies increases 2030 GHG emissions by 5.5 percent in the US, 3.7 percent in Europe, and 1.2 percent in China; global average increase of 1.2 percent.

*International Monetary Fund, Commodity Special Feature — Market Developments and the Impact of AI on Energy Demand (Section 1).*

### Section 2

### commodityspecialfeature - Section 2

### Emission impacts of IT/AI expansion
- Cumulative global GHG emissions increase of 1.7 gigatons (Gt) between 2025 and 2030 under the AI scenario with current energy policies.
- In the AI scenario under alternative energy policies, cumulative global GHG emissions increase is limited to 1.3 Gt by 2030, which is 24 percent less than in the AI scenario under current energy policies.
- The additional social cost of 1.3 to 1.7 Gt of carbon-dioxide-equivalent emissions is about $50.7 billion to $66.3 billion, using a median social cost of carbon estimate of $39 per ton based on 147 published studies with more than 1,800 estimates (Moore and others 2024).
- The additional social cost represents about 1.3 percent to 1.7 percent of the AI-driven increase in real world GDP between 2025 and 2030.

### Macroeconomic impacts (GDP and sectoral growth)
- In the AI scenario under current energy policies, the AI shock raises the average annual growth rate of global GDP by 0.5 percentage point between 2025 and 2030.
- This estimate is in line with previous IMF estimates ranging between 0.1 percentage point and 0.8 percentage point (April 2024 World Economic Outlook).
- Gains are greater in countries where the projected growth rate of the IT sector and its relative importance in the economy are higher.
- In the AI scenario under alternative energy policies, growth gains are slightly reduced because of feed-in tariff policies.
- Total fiscal costs of these tariffs range from 0.3 percent to 0.6 percent of GDP across countries and are financed through increased lump-sum taxes, which slightly reduce household consumption.
- Despite those fiscal costs, growth benefits from AI expansion far outweigh the costs, resulting in similar average annual GDP growth across both scenarios.

### Electricity supply, generation mix, and prices
- Feed-in tariffs increase generation from solar and wind sources under alternative energy policies (panel description from source).
- The total increase in electricity supply relative to the baseline scenario in TWh is identical under both current energy policies and alternative energy policies (panel note from source).
- Increasing electricity demand from the IT sector will stimulate overall supply; if supply is sufficiently responsive, the result is a small increase in electricity prices; more sluggish supply responses lead to much stronger price surges.
- In the United States, AI expansion alone could increase electricity prices by up to 9 percent, according to the source.

### Distributional and investment implications
- Economic benefits from AI are likely to be unevenly distributed across countries and among different groups within societies, potentially exacerbating existing inequalities.
- Demand for computing and electricity from AI service producers is subject to wide uncertainty, which may delay energy investments, causing underinvestment and higher prices.

### Conclusions and policy implications
- Despite challenges related to higher electricity prices and GHG emissions, gains to global GDP from AI are likely to outweigh the costs of the additional emissions.
- Policymakers and businesses must work together to ensure AI achieves its full potential while minimizing societal costs, including addressing electricity supply responsiveness, investment timing, and distributional impacts.

*Source: commodityspecialfeature - Section 2 (IMF PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/weo/2025/april/english/commodityspecialfeature.pdf_