AI economic boost to outstrip climate cost, IMF says

With previously unheard-of increases in productivity, creativity and growth, artificial intelligence is drastically changing the global economy. However, there is a significant climate footprint underlying its digital promise – a paradox highlighted by a recent IMF report, which predicts that AI’s economic benefits will outweigh its environmental costs, despite the fact that its rising energy needs present significant obstacles for international infrastructure and policy.

Artificial Intelligence – the catchphrase of the decade – promising surges in growth, innovation, and productivity that have the potential to drastically alter both our daily lives and the global economy as a whole. Based on lines of code and enormous neural networks, we see a world that is becoming more and more digital — possibly even ethereal. However, the unpleasant reality, the murky little secret concealed beneath the shiny servers and slick user interfaces, is that AI is not virtual. It has a strong physical connection and an insatiable desire for power.

This contradiction is highlighted by recent research from the International Monetary Fund (IMF), presented at its annual Spring Meetings. The topline finding, widely reported, is undeniably optimistic: the economic gains from AI are likely to outweigh the cost of the additional carbon emissions it generates. According to the Fund, AI may significantly accelerate growth by increasing global GDP by around 0.5% a year between 2025 and 2030.  On the surface, the environmental cost of the required electricity consumption appears to be negligible in comparison to this economic plenty.

The Price of Progress

AI technologies around the world are driven by enormous data centers – nameless warehouses filled with servers that run around the clock – containing enormous computing resources and energy to train today’s huge language models. Demand is increased by operating and reacting to user cues. A dramatic illustration of the enormous physical scale involved is that Northern Virginia, a significant center, already houses server space equal to almost eight of New York’s iconic Empire State buildings.

The figures are staggering.  Data centers consumed 400–500 TWh of power globally in 2023, more than double that of 2015. By 2030, the world’s AI-driven power consumption may more than treble to 1,500 TWh. That’s about the same as India’s current electricity use, so by that time, AI will be a major new player on the global energy scene and may require multiple times the power needed by all the electric cars in the world.

The intensity too is rising, not simply the magnitude as a whole. Between 2019 and 2023, power expenditures as a percentage of overall expenses for vertically integrated AI enterprises almost quadrupled. Although this percentage is still rather low when compared to operators of specialized data centers, whose power expenses might account for 13–15% of overall expenses, the trend is evident and increasing.  These businesses’ direct energy impact becomes considerably more significant as they construct more of their own data centers.

Economic Gain, Environmental Headache

In comparison to the anticipated economic benefits, the IMF research indicates that the climate cost of these additional emissions would be negligible. The growth of the IT industry driven by AI may result in a 1.2% rise in global greenhouse gas emissions by 2030 under present energy policy. This might add up to an extra 1.7 Gt between 2025 and 2030, which is almost the same as all of Italy’s energy-related GHG emissions during a five-year period.

The increased social cost of these emissions is estimated to be between $50.7 billion and $66.3 billion based on a median social cost of carbon calculation. Even while this number is significant on its own, it only represents 1.3% to 1.7% of the anticipated rise in global GDP over the same time period due to AI.

‘Minor’ however, does not imply inconsequential. This contributes to an already alarming accumulation of emissions. Furthermore, even while the overall economic advantages could outweigh the overall environmental costs on a global scale, it is unclear that the advantages would be distributed fairly among nations or within communities, which might exacerbate already-existing disparities. Meanwhile, the ensuing pollution will undoubtedly affect the whole world.

The Griddle of the Grid

The difficulty lies not just in the amount of energy but also in how we provide it. The data center and the transmission line intersect here, or more precisely, here is where the rubber meets the road.  It goes without saying that rising demand for power will drive up costs.  The energy supply’s reactivity and, more importantly, infrastructure play a major role in how much pressure there is.

Price increases might be significant if the supply of power is slow, especially if investments in transmission and distribution infrastructure and the expansion of renewable capacity are slow.  According to IMF research, AI expansion alone might result in a considerable increase in power costs by 2030, possibly by as much as 8.6% in the U.S. under existing policies, given the limited infrastructure expansion and restricted growth in renewables compared to the baseline.  One important consideration is the grid’s capacity — or lack thereof.  Energy-intensive companies may face difficulties if electricity must be diverted from other sectors if the system is unable to keep up.

Compared to a scenario with ‘current policies,’ alternative energy policies, such subsidies for renewables, can assist move the generation mix away from fossil fuels, reducing emissions and price increases.  This emphasizes how important legislation is in managing the physical requirements of AI.

—

The future is equally clouded with a great deal of uncertainty. Enhancements in algorithm efficiency may lower computation expenses and energy consumption.  However, demand might easily climb further due to the search for more potent models, the emergence of energy-intensive ‘reasoning models,’ and the straightforward fact that cheaper prices stimulate greater usage of AI. There is also risk associated with this uncertainty itself, which might cause future price increases and postponed energy expenditures.

Even if AI seems to have a bright future for the economy, the route to getting there passes right through the actual realm of energy generation and delivery. Authorities, tech firms, and energy businesses ‘must play an active role in ensuring AI is used intentionally, equitably, and sustainably,’ according to one expert.  Ignoring AI’s ‘power hungry’ nature puts us at risk for rising energy expenses and emissions, as well as perhaps impeding the same development we want to attain in the event that the underlying infrastructure fails.  In order to guarantee that this transformation is sustainable in the widest meaning of the word, energy policy must be in line with AI development not only for environmental reasons but also for economic ones.

Read: International Monetary Fund, 2025. Power Hungry: How AI Will Drive Energy Demand. IMF Working Paper No. 2025/081. Available here.