Network Arbitrage

A mental model for understanding whats happening in energy

Last year when I was at the Department of Energy, I helped draft our data center load growth strategy. I had the opportunity to talk with hyperscalers, data center colocation providers and private equity firms, gridtech vendors, utilities, and even electrician unions about what they were seeing on the ground in terms of load growth and data center demand. “Network Arbitrage” increasingly became my mental model for how to understand and organize a number of seemingly different technology and grid trends.

---

A useful model is to imagine the world as many overlaid networks which occasionally intersect. Historically, networks like pipelines, railroads, trucking routes, and shipping lanes moved commodities—such as crude oil, gasoline, and natural gas—between endpoints. Conversion points, like refineries or gas stations, allowed commodities to transfer seamlessly from one network to another. Understanding these networks and their evolution can provide financial reward: high-frequency traders famously optimized latency performance on fiber optic networks. Today, networks like power grid and fiber optic play an increasingly vital role alongside traditional networks such as railroads, trucking routes, and pipelines.

One way to understand many of the dynamics today in the energy world is that the power grid is uniquely bottlenecked by constraints on building new transmission lines i.e. edges in a network, at speed and scale. This is a unique constraint compared to historical growth in other networks, like the transcontinental railroad and the rural electrification campaign. In this case, network arbitrage is primarily arbitraging against the electric network to expanding with new edges, enabled by cheaper distributed energy resources like solar and storage.

For instance, Casey Handmer’s Terraform Industries uses solar to create hydrogen and capture carbon dioxide, and then synthesize natural gas. This is not only a bet on the continued end use of natural gas (including in extraterrestrial applications) but more locally arbitraging the inability to deliver electrons from solar through the electric transmission network for a less constrained natural gas network (and betting that developing an extensive hydrogen transportation network will remain challenging). X.AI’s Tennessee data center is similarly taking advantage facility that is colocated with transmission and a natural gas pipeline that runs adjacent to the facility to run gas generators in the short-term while the data center waits for power grid interconnection.

Data centers themselves are factories that convert electrons moving along power lines into bits that move along broadband, as are crypto mining facilities. Because the growth of power lines are constrained, AI hyperscalers are building broadband edges between data centers instead of centralizing along the power network. Power grid network constraints today are already causing curtailment of renewable energy, sometimes up to 25% of energy production. Companies like Crusoe Energy are taking advantage of the fact that renewable installations also have fiber connections, to colocate data centers with renewable energy sites, absorbing excess energy that would otherwise go to waste.¹

Batteries are also a form of network arbitrage. As an energy storage medium analogous to oil, they can also move along other networks before plugging back into power networks e.g. transporting batteries via trains to transfer electrons between regional grids lacking sufficient transmission capacity. One can imagine a startup that floats barges of batteries down the Mississippi to arbitrage between MISO (grid operator in midwest) and SPP (grid operator for Nebraska, Oklahoma).

Base Power takes advantage of the overbuild capacity in residential networks to install distributed batteries which can aggregate together across a network as a “virtual power plant”, rather than waiting for a single, large-scale grid interconnection. Batteries can also buffer intermittent loads, smoothing out demand spikes, and reducing the size of interconnection needed to the power grid e.g. Electric Era’s battery-buffered EV chargers.

Indeed, distributed energy resources create an increasingly fractal power grid. Instead of utility-scale energy storage projects, Base Power installs behind-the-meter batteries in homes. Companies like Impulse Labs, which pair batteries with electric induction stove tops to avoid needing higher voltage outlet upgrades, move batteries from behind-the-meter to behind-the-stove. One can imagine soon heat pumps and water heaters might also come with their own batteries. In this case, the bottleneck is the electrician needed to upgrade the outlet (or electric panel) to manage higher voltage appliances, but this “interconnection for home appliances” is a variation on a common theme of bigger wires as the bottleneck.

Finally, solar combined with storage enables the ultimate arbitrage: exiting the grid entirely. Pakistan has seen ~10% load decrease from its grid operators, alongside an unaccounted for surge in solar imports from China, which suggests that the CapEx of going off-grid is already more appetizing than dealing with high socialized network costs and poor performance for certain power grids. And notably in the US, a titanium plant in West Virginia is going off-grid and using solar + storage financed by Berkshire Hathway. Recent analysis has also demonstrated how this could be feasible even for large scale data centers, with some natural gas backup.

In short, everyone is looking for new network arbitrage opportunities and betting against new transmission. There are a number of complex issues for why building transmission is hard, ranging from securing right-of-way and permits, to cost allocation, to constrained energy market sizes which disincentivize regional transmission.

In the short-term, grid-enhancing technologies (GETs) are a class of transmission technologies that can increase utilization and capacity of existing grid infrastructure. Dominion is deploying the world’s largest dynamic line rating project, which models transmission line capacity dynamically based on weather conditions, rather than as a static constant. Reconductoring replaces older wires on existing transmission with newer wires with higher capacity i.e. the graph topology doesn’t change but the edge capacity increases. Last year, the Department of Energy (DOE) reduced the required permitting for reconductoring and funded VEIR, a superconducting power line company, to reconductor existing power lines in Michigan.

All of these different forms of network arbitrage are to varying degrees exposed to simply building more transmission faster. DOE has taken some steps in that direction, including designating national priority transmission corridors with expedited permitting timelines, and improving federal permitting coordination. But until we figure out how to build transmission, network arbitrage will continue to reshape the way transact energy.

1. Amazon’s Talen acquisition deal and Microsoft and Constellation’s Three Mile Island restart deal are similar styles of colocation, though it is worth noting they will still draw substantial load from the grid during refueling and maintenance operations.