Electricity is quietly becoming the limiting factor for the digital economy. For years, the conversation around data centers has focused on compute capacity, cooling efficiency, and the pace at which cloud providers can build new facilities. But a new wave of reporting is shifting attention to a more fundamental constraint: power availability. According to the latest analysis highlighted in TechCrunch’s coverage, data centers are expected to use roughly four times more electricity by 2035 than they do today. Even more striking, the report warns that new data centers coming online through 2033 could consume as much electricity as India uses today.
Those numbers are big enough to feel abstract—until you translate them into what they mean for grids, permitting timelines, fuel supply, and the everyday reality of keeping servers running without interruption. The core issue isn’t simply that data centers will grow. It’s that their growth is happening at the same time as other electricity-intensive trends—electrification of transport and heating, industrial decarbonization, and the broader expansion of renewable generation that still requires transmission and storage to match demand. When multiple systems compete for the same infrastructure, the bottleneck moves from “can we build?” to “can we reliably supply power?”
What the projection is really measuring
The headline figure—4x electricity use by 2035—should be understood as a scenario based on expected build rates, utilization patterns, and the energy intensity of modern facilities. Data centers don’t just consume electricity when they’re actively processing workloads; they also draw power for cooling, networking, storage, and redundancy systems designed to keep operations stable. As AI workloads become more common, the power profile changes again. Training and inference can increase demand dramatically, but the relationship between compute and electricity isn’t linear in a simple way. Efficiency improvements can reduce energy per unit of work, yet overall demand can still rise faster than efficiency gains.
That’s why the report’s second point matters: if new data centers built through 2033 could reach electricity consumption comparable to an entire country’s current usage, then the question becomes less about incremental growth and more about whether the power system can scale quickly enough. In other words, the projection is not merely a statement about technology—it’s a statement about timing. Electricity infrastructure takes years to plan and build, and grid upgrades often face regulatory, land, and interconnection hurdles.
Why “as much as India uses today” is a wake-up call
Comparing data center demand to a national electricity consumption level is a rhetorical device, but it’s also a useful one. It forces readers to confront the magnitude of the load. India’s electricity use represents a massive, distributed demand across millions of homes, businesses, and industries. If data centers approach that scale within a relatively short window, then the impact won’t be confined to a few utility territories or a handful of large projects. It would reshape planning priorities: where new generation is built, how transmission lines are routed, and how quickly utilities can connect new loads.
There’s also a second implication: even if data centers are located in regions with strong generation capacity, the grid must still move that power to where it’s needed. A data center can’t “choose” electricity the way a consumer chooses a product. It depends on the physical network. That means the bottleneck can appear in unexpected places—at substations, along transmission corridors, or in the interconnection queue where utilities evaluate whether they can safely add new load without destabilizing voltage and frequency.
The hidden energy costs inside a data center
When people think about data center electricity, they often picture servers. But the electricity story is broader. A typical facility includes:
1) IT load: servers, GPUs/accelerators, storage systems, and networking equipment.
2) Cooling and air movement: chillers, pumps, fans, and air handling units.
3) Power conversion and distribution: transformers, switchgear, UPS systems, and power supplies.
4) Redundancy: N+1 or 2N designs that ensure uptime during failures, which can increase baseline consumption.
5) Controls and monitoring: building management systems, security systems, and operational overhead.
Modern designs aim to reduce waste. Liquid cooling, improved airflow management, hot/cold aisle containment, and better power supply efficiency all help. Yet the report’s warning suggests that even with these improvements, the sheer volume of new capacity could overwhelm efficiency gains. This is a recurring pattern in energy transitions: efficiency reduces energy per unit of service, but total service demand grows so fast that absolute energy use still rises.
AI changes the shape of demand
AI workloads are often described as “more compute,” but from a grid perspective, the key is that AI can increase both peak and sustained power draw. Some AI training runs may be scheduled, but many inference workloads are continuous and latency-sensitive. That means operators may need to keep capacity ready rather than cycling it down frequently. Additionally, AI clusters can be deployed in ways that concentrate power density—more compute per square meter—which can raise cooling requirements and increase the facility’s maximum draw.
This is where the unique take on the story matters: the grid doesn’t just need more electricity; it needs electricity that can arrive reliably at the right time and in the right form. High-density loads can stress local infrastructure even if the region’s total generation capacity looks sufficient on paper. Utilities and grid operators care about localized constraints—transformer loading, feeder capacity, and voltage regulation—because those determine whether power can be delivered safely.
The interconnection bottleneck: why “capacity” isn’t the same as “availability”
One of the most misunderstood parts of the electricity debate is the difference between generation capacity and deliverable power. A region might have enough planned generation to meet long-term demand, but if new data centers cannot get connected quickly, the load can’t materialize. Interconnection queues can become a bottleneck, and the process can be slow because utilities must study impacts on reliability and system stability.
In practice, this means data center operators may face delays, higher costs, or forced compromises. Some may pursue behind-the-meter solutions such as backup generators, batteries, or dedicated power purchase agreements. Others may negotiate for power from specific substations or contract for capacity in ways that shift risk onto utilities or developers. These strategies can help individual projects, but they don’t automatically solve the systemic problem of scaling transmission and distribution.
Cooling innovation is real—but it’s not a magic wand
It’s tempting to treat cooling improvements as the solution. And there are genuine advances: direct-to-chip liquid cooling, immersion cooling, heat reuse for district heating in some locations, and more sophisticated thermal management. These can reduce energy used by fans and chillers, and they can allow facilities to operate closer to ambient temperatures.
However, the report’s projections imply that even substantial efficiency gains may not keep pace with demand growth. There’s also a practical limit: cooling systems can only reduce energy so far before they hit thermodynamic constraints and design tradeoffs. If the facility’s IT load increases by a factor of several, cooling improvements may reduce the percentage of energy spent on cooling, but the absolute cooling energy can still rise.
The climate angle: emissions depend on the grid mix
The climate implications of data center growth are often discussed in terms of carbon emissions, but the emissions outcome depends heavily on the electricity source. If new demand is met by fossil generation, emissions rise sharply. If it’s met by renewables and low-carbon resources, emissions can be lower—though not necessarily zero, because building new infrastructure and manufacturing equipment also have embodied emissions.
There’s another nuance: even if a data center signs contracts for renewable energy, the timing and location of generation matter. Grid constraints can lead to situations where additional demand causes marginal generation to come from higher-emitting sources until the grid fully adapts. That’s why the conversation increasingly includes not just “renewable procurement” but also grid buildout, transmission expansion, and the pace at which low-carbon generation can be delivered to the specific nodes where data centers connect.
A unique perspective: the data center boom is also a grid modernization test
Seen through a different lens, the data center surge is less a standalone story and more a stress test for the electricity system’s ability to modernize. Grids are evolving—smart meters, demand response, better forecasting, and more flexible generation—but they are still constrained by physical infrastructure and regulatory processes.
Data centers are unusual customers. They can be extremely large, predictable in some cases, and highly sensitive to downtime. That combination makes them valuable partners for grid planners when managed well. For example, some operators can participate in demand response programs, shift workloads, or use thermal storage to reduce peak draw. But the scale implied by the projections suggests that even with flexibility, the system must expand.
This is where policy and market design become central. If electricity markets reward capacity and reliability appropriately, utilities and developers have incentives to build the necessary generation and transmission. If not, the system can end up with a mismatch: data centers want power now, while the grid can only deliver it later.
What happens if the timeline slips?
If the projected electricity demand arrives faster than the grid can supply it, several outcomes are possible:
1) Higher electricity prices in constrained regions, which can change the economics of hosting and cloud services.
2) Delays in data center construction or commissioning, potentially slowing some AI deployments.
3) Increased reliance on backup generation, which can raise emissions and operating costs.
4) More aggressive contracting for dedicated capacity, shifting risk and cost to consumers or taxpayers depending on local structures.
5) Political pressure to accelerate permitting for transmission lines and generation, which can be contentious.
None of these outcomes are guaranteed, but the risk is real because electricity infrastructure is slow compared to software deployment. A model can be trained and rolled out quickly; a new transmission corridor cannot.
The “through 2033” detail: why the early years matter most
The report’s warning about new data centers built through 2033 consuming electricity comparable to India’s current usage highlights that the near-term buildout is critical. The early years determine whether the grid can adapt smoothly or