The world’s data centers will consume nearly 1,000 terawatt-hours (TWh) of electricity by 2030—more than the current annual demand of Germany. Much of that surge is driven by artificial intelligence, whose insatiable appetite for computing power is reshaping global energy systems. But the grid cannot keep up. Transmission lines take years to build, while AI demand spikes in months. This mismatch isn’t just a logistical headache—it risks blackouts, wasted renewable energy, and higher emissions.
Now, a new study shows that data centers themselves could help solve the problem. By shifting computing workloads across geography—migrating tasks from one server farm to another—companies can effectively create what researchers call a virtual transmission line (VTL). This isn’t about moving electrons. It’s about moving demand.
In a coordinated planning model, geographic workload migration reduces total system costs by up to 12%, cuts unserved data-center load by 80%, and slashes renewable curtailment by nearly half—all while requiring one fewer physical transmission line over an eight-year horizon. The virtual line doesn’t replace infrastructure. But it buys time, reduces congestion, and makes better use of existing capacity. For the first time, researchers have quantified how much flexibility data centers can offer—and under what conditions it matters most.
The Science
The study, led by Mengxin Zhu and colleagues at Shanghai Jiao Tong University (Zhu et al., 2026), models data-center workload migration as a virtual transmission line—a digital proxy for physical power flow. Unlike traditional demand response, which shifts load in time, VTL shifts it in space. When a data center in Texas migrates compute jobs to one in New Mexico, it reduces local electricity demand and increases it elsewhere—without any physical transfer of energy. The grid sees only the net change in nodal load.
To test this idea, the team built a mixed-integer linear programming (MILP) model that co-optimizes two decisions over an eight-year planning horizon: where and when to build new transmission lines, and where and when to migrate data-center workloads. The model includes real-world constraints: construction delays (some lines take years to complete), annual growth in demand and renewables, communication bandwidth limits, and latency thresholds (you can’t migrate workloads to a data center if the network delay exceeds acceptable levels).
The framework is evaluated on a modified IEEE 14-bus system—a standard test network in power systems research—with three data centers, two wind farms, and five candidate transmission lines
. Four planning cases are compared:
- C1 (No expansion, no VTL): Status quo—no new lines, no migration.
- C2 (Build-only): Physical expansion only.
- C3 (VTL-aware build): Coordinated expansion and migration.
- C4 (VTL-only): Migration without new lines.
A counterfactual scenario also tests what happens when VTL is used after transmission decisions are locked in—separating the operating value of migration (flexibility in real time) from its planning value (influencing long-term investment).
The model minimizes total discounted cost, including transmission investment, generation, carbon emissions, renewable curtailment, unserved data-center load, and migration costs. Crucially, it accounts for the fact that not all workloads can be moved: only a fraction (denoted ) of each data center’s demand is considered migratable, based on task dependencies and service-level agreements.
What They Found
The results reveal that coordination is everything. When VTL is integrated into planning (C3), the system performs dramatically better than when either resource acts alone.
First, costs drop significantly. Over eight years, the coordinated plan (C3) reduces total system cost by 12% compared to the build-only scenario (C2). This savings comes from avoiding one entire transmission line—saving $187 million in capital investment—while still meeting demand and reducing congestion.
Second, reliability improves. In the build-only case, rapid AI load growth leads to unserved data-center load in later years—up to 4.2 TWh annually by year 8. But with VTL-aware planning, unserved load drops to just 0.8 TWh—a reduction of 81%. That’s equivalent to keeping 500,000 additional AI training jobs running each year.
Third, renewables are used more efficiently. Without coordination, wind power is curtailed (wasted) due to transmission bottlenecks. In C2, curtailment peaks at 3.1 TWh/year. In C3, it falls to 1.7 TWh—a 45% reduction. This means more clean energy is used, fewer fossil plants are dispatched, and carbon emissions drop.
Annual System Performance: C2 vs C3
Comparison of key metrics between build-only (C2) and VTL-aware (C3) planning over eight years.
| Label | Value |
|---|---|
| Year 1 | 0.5 |
| Year 2 | 0.7 |
| Year 3 | 1.1 |
| Year 4 | 1.8 |
| Year 5 | 2.4 |
| Year 6 | 3 |
| Year 7 | 3.6 |
| Year 8 | 4.2 |
Figure 3 shows the construction schedule: C3 builds four lines, while C2 builds five. The avoided line (between buses 6 and 13) is replaced not by physical infrastructure, but by strategic workload migration.
The VTL-only case (C4) fails to meet demand, confirming that migration alone cannot replace transmission. But when combined, the two resources are synergistic. The planning value of VTL—its ability to shape investment decisions—is nearly twice as large as its operating value. In other words, knowing that workloads can be moved changes how we build the grid.
Sensitivity analysis reveals when VTL delivers the most value:
- When transmission capacity is tight, VTL reduces congestion by shifting load away from bottlenecks.
- When construction costs are high, avoiding even one line yields big savings.
- When data-center flexibility is greater (higher ), more migration is possible.
- When load growth is rapid, VTL acts as a shock absorber.
Renewable Curtailment and Migrated Workload Over Time
Annual renewable curtailment and migrated workload under VTL-aware planning (C3).
| Label | Value |
|---|---|
| Year 1 | 2.1 |
| Year 2 | 2.3 |
| Year 3 | 2.6 |
| Year 4 | 2.8 |
| Year 5 | 3 |
| Year 6 | 2.9 |
| Year 7 | 2.5 |
| Year 8 | 1.7 |
Why This Changes Things
For decades, grid planners have treated demand as fixed and location-bound. You build power plants and transmission lines to serve where people live and work. But data centers are different. They are large, flexible, and digitally connected. A single hyperscale facility can draw as much power as a small city—but unlike a city, it can, in principle, move its load.
This study reframes data centers not just as energy consumers, but as spatial demand resources. The virtual transmission line is a new kind of infrastructure—one that exists in software, not steel.
The implications are profound. In the U.S., the average major transmission project takes 10 years to permit and build (DOE, 2023). Meanwhile, AI data-center demand is growing at 15–20% annually. By the time a new line is energized, the load profile may have shifted entirely. VTL offers a way to bridge that gap.
Consider Texas, where ERCOT has faced repeated grid stress due to data-center growth. In 2023, the grid operator warned of potential blackouts if demand outpaced supply. A VTL-aware approach could allow Google or Microsoft to temporarily shift AI training jobs from Dallas to Amarillo when wind generation is high in the Panhandle—relieving congestion on overloaded lines without building new ones.
Globally, this could accelerate the clean energy transition. In China, wind-rich Inner Mongolia is often disconnected from high-demand coastal cities. Workload migration could allow Beijing-based AI firms to “import” clean power not by building ultra-high-voltage lines, but by running computations in Inner Mongolia instead. The electrons don’t move—the work does.
The model also challenges the assumption that more transmission is always the answer. While physical lines remain essential, this study shows they can be used more efficiently. In the 118-bus validation test, coordinated planning reduced physical expansion needs by 20% while maintaining reliability. That’s not just cheaper—it’s faster. And in the race to decarbonize, speed matters.
What’s Next
The study has limitations. It assumes perfect foresight—planners know future load and renewable growth exactly. In reality, uncertainty abounds. Future work could incorporate stochastic optimization to account for variability in AI adoption, policy shifts, or extreme weather.
It also treats data-center flexibility as a given. But in practice, migration depends on corporate incentives, contractual obligations, and cybersecurity concerns. A bank may not want its AI fraud detection models running in a foreign data center. The model sets (60% migratable), but real-world values could be lower.
Still, the core insight stands: spatial demand flexibility has quantifiable value. The next step is policy. Regulators could treat VTL like other grid resources—allowing data centers to bid migration capacity into transmission planning processes, just as batteries bid into energy markets.
Some companies are already moving in this direction. Google’s “carbon-intelligent computing” platform shifts non-urgent tasks to times and places with cleaner grids (Google, 2022). Microsoft has experimented with underwater data centers to leverage ocean cooling. But these are operational tweaks. This study shows that migration can—and should—be part of infrastructure planning.
The biggest barrier may be mindset. Grid operators are trained to think in megawatts and miles. Data-center operators think in petaflops and milliseconds. Bridging that gap requires new institutions, new markets, and new metrics. One candidate: a “virtual transfer capacity” rating, analogous to physical line ratings, that quantifies how much demand a data-center network can flex across geography.
The energy system of the future won’t be built only with wires and turbines. It will also run on algorithms and fiber optics. The virtual transmission line is not a replacement for the physical grid. But it may be the most cost-effective upgrade we have.
As Zhu et al. write: “VTL-aware planning… requires one fewer candidate line than build-only planning.” That single line—avoided through smarter coordination—could save hundreds of millions of dollars, prevent years of delays, and unlock more renewable energy. In a world racing to meet AI’s energy demands without frying the planet, that’s not just an engineering insight. It’s a lifeline.