AI energy demand: Costs & Investments
The AI boom is driving up demand for not only chips, but also electricity and infrastructure. This leads to rising costs and new challenges for energy providers and data center operators.
Quelle: Own Presentation
AI Boom: Costs & Infrastructure
The costs of the AI boom include not only chips but also electricity and concrete. Wholesale electricity prices near large data hub clusters are in some regions up to 267 percent over five years increased. At the same time, Citigroup forecasts AI-related investments of 2.8 trillion US dollars by 2029.
AI Capex includes long-term investments by hyperscalers and operators in data centers, power connections, grid upgrades, cooling, land and hardware such as GPUs/TPUs. Hyperscalers are cloud giants like Microsoft, Amazon and Alphabet, that run huge data centers. On the electricity side it is about load (capacity in gigawatts) and energy (terawatt-hours per year).
The International Energy Agency (IEA) expects data center electricity consumption to rise to around 945 TWh, , which is roughly equal to Japan's current electricity consumption. AI is the strongest driver here. BloombergNEF estimates that data center power consumption in the US will by 2035 rise from about 35 GW in 2024 to 78 GW higher, more than double will be.
Current status & implications
Bloomberg reports that wholesale electricity in US regions with concentrated data hub projects today is up to 267 percent more expensive than five years ago. The IEA has raised its electricity forecasts for 2025/26, among other reasons due to additional data center load in the US and Europe. For the overall market the IEA expects US electricity demand to grow by 2.3 percent and 2.2 percent in 2026, , which is well above the trend of the last decade.
On the investment side, Citi raises its estimate for AI-related spending by Big Tech through 2029 to 2.8 trillion US dollars; ; already by 2026 AI capex is expected to reach around 490 billion US dollars. Individual companies set the pace: Microsoft announced a quarterly capex of 30 billion US dollars in prospect. Utilities respond with multi-year programs, for example CenterPoint with 65 billion US dollars by 2035, , explicitly due to data center load peaks. Parallel, end-customer prices are rising: Bloomberg recently cites an increase in electricity costs in the US of 4.5% year-over-year, , well above overall inflation.
Quelle: forbes.com
Global electricity demand from data centers, AI and cryptocurrencies is expected to rise sharply by 2026.
The sprint is driven by several motives. First, capacity creates market share: whoever secures enough GPU capacity, space and grid connections first, scales AI services faster, from cloud training to generative assistants. Second, site policy is decisive: regions with affordable, reliable energy, fast permitting and cool climates attract more data centers; grid bottlenecks and high peak prices deter. Third, operators must reduce emissions, limit water usage and alleviate local grids due to regulation and ESG requirements to avoid delays, restrictions or charges.
The media narrative amplifies the topic: individual megaprojects, price spikes and trillion-dollar figures attract a lot of attention, while grid and permitting details are often only visible in hindsight.
Quelle: YouTube
It is documented that wholesale electricity prices in some data hub regions have risen sharply; Bloomberg cites up to 267 percent in five years. . It is also evidenced by the clear upward revision of AI capex forecasts to 2.8 trillion US dollars by 2029. . The IEA expects a doubling of data center electricity consumption by 2030 to around 945 TWh.
Quelle: amritaroy.substack.com
The capital expenditures (Capex) of hyperscalers for AI show a significant rise through 2025.
It is unclear how much of the extra demand is actually covered by additional generation from renewables and how strongly demand-side management (load shifting) can smooth peaks; estimates vary by region and grid expansion. The claim that AI is solely responsible for general electricity price increases is misleading, as other drivers such as fuel prices, grid bottlenecks, weather extremes and general electrification in industry and buildings also play a role.
Investors tolerate the high capital intensity so far, as long as revenue and cloud margins grow; analysts point to higher utilization and pricing power in the AI stack. Critics warn of circular financing and grid risks when connection capacities grow faster than local generation. Some utilities are therefore pushing large grid investments and tying data centers to flex and storage obligations.
For operators this means thinking about site selection holistically: available capacity (MW), permitting duration, electricity price structure (Peak/Off-Peak), grid expansion plans, water and PPA availability. For energy providers and municipalities, early cluster planning, grid expansion and storage are crucial to avoid price spikes and acceptance issues. Companies without their own data centers must note that cloud prices and latency will increasingly depend on regional energy and grid conditions; multi-region architectures can provide a hedge. For households, regional price premiums are possible, but not monolithic; local grid reports and regulator documents should be consulted for context.
Quelle: YouTube
Future & Challenges
Open questions remain: How quickly can grid and generation capacity really be expanded, including permitting, line construction and transformers? What role will flexible loads, storage and waste heat utilization play in system stability? Concrete, comparable PUE and WUE data at site level would be needed to transparently assess burdens and efficiency gains. How resilient are the 2.8-trillion-US-dollar capex trajectories, , if financing costs rise or regulators impose stricter site requirements? And to what extent do hyperscalers shift workloads to regions with cooler climates and higher renewable generation to cut costs and CO2?
Quelle: thenewstack.io
Despite efficiency gains, the electricity demand from data centers is expected to continue rising.
The AI boom is not only a chip but an energy and infrastructure issue. Evidence shows rising regional electricity prices at data hubs, a capex wave in the trillions, and drastically growing loads. For companies this means: decisions on AI scaling should take the local energy reality as seriously as model benchmarking – with clear grid commitments, PPAs, efficiency plans and a realistic timeline. Those who think about energy, location and financing together will secure speed and keep costs under control.