For the past year, Evolved Energy Research has partnered with ClearPath to build something the energy-transition planning toolkit has been missing: a capacity-expansion model that can reason about transmission the way planners actually do — line by line, upgrade by upgrade, on the real network.
The result is a new version of our RIO capacity-expansion model that fuses long-range, economy-wide resource planning with nodal power-flow physics. To prove it out, we used it to answer a question the industry is wrestling with right now and that conventional models simply cannot express: when should you reconductor an existing line, and when do you have to build greenfield?
Applied to ERCOT, the model returns a two-sided answer. Reconductoring is genuinely powerful: re-stringing existing corridors meets high-growth load quickly and cheaply, avoiding roughly 6,000 miles of new lines and $20 billion by 2040 and getting power to data centers years sooner than new construction can. But reconductoring alone is inoperable; a higher-ampacity conductor relieves a thermal limit, yet it cannot solve a network problem. Restrict ERCOT to reconductoring only and within the next few years, the system cannot operate. The two are complements, not substitutes: greenfield is necessary, and treating reconductoring as a reason to ease off on siting and permitting reform is short-sighted. The detailed findings are in the attached study deck. Our partners at ClearPath have also released a companion report for a policy focused audience.
The problem: Zones are an abstraction that hides the question
Capacity-expansion models are the workhorse of long-range electricity planning. But almost all of them represent transmission as pipe flow: a corridor between two zones where power moves freely up to a capacity limit, with no regard for the physics that actually govern how electricity distributes across a network. That's precisely the oversimplification that hides the reconductor-or-greenfield question: a single capacity number can't tell you what kind of physical upgrade a corridor actually needs.
ERCOT's full nodal network has roughly 5,000 buses and 7,000 lines, compressed by modelers into a handful of zones and a few inter-zonal "pipes" apiece. That is a steep reduction: thousands of lines with different physical characteristics, all of it lost when consolidated into a handful of zone-to-zone numbers.
Therein lies the central insight that drove this work: the moment you merge physical lines into a zone, you lose the very information needed to choose between reconductoring and building new. Once a corridor is a pipe with a single number on it, you can no longer ask:
- Is this a line I can reconductor (i.e., re-string with a higher-capacity conductor on the same towers) or do I need a new line on new right-of-way?
- What voltage does the fix require?
- Will the upgrade survive an N-1 contingency (i.e., the loss of any single line elsewhere on the system)?
- What does the next increment of transfer capacity actually cost, once the cheap fixes are used up?
A pipe can't answer any of that, because a pipe has no conductor, no voltage, no impedance, and no neighbors. The question "should we reconductor this line or build a new one?" applies to a real physical line, but if your model has already collapsed all those lines into a pipe, that line no longer exists in the model. You've thrown away exactly the information needed to answer the question.
The conventional workflow is a one-directional chain: capacity expansion feeds a power-flow model. In principle, the modeler iterates back to reconcile them, but that loop is costly enough that it rarely closes. Instead, transmission planning ends up being purely reactive; it works out the details for the devised generation plan, but it rarely informs such a plan at a deeper level. Most of the time each planning stage simply hands its answer to the next, so the coupled generation-and-transmission problem is never optimized as a whole.
The breakthrough: Keep the nodal network intact, and break in and out of the optimization
Our approach refuses the trade-off. Instead of replacing the nodal network with zones, we keep the full nodal network alive the entire time and let the optimization and the network talk to each other.
The optimization itself stays small and fast: a zonal capacity-expansion linear program (LP; 20 zones for ERCOT) that co-optimizes generation, storage, fuels, load, and transmission together. This LP no longer treats its corridors as free pipes; rather, it is continuously grounded in the real ~5,000-bus, ~7,000-line topology through a set of side calculations that runs between the optimization years. We solve the economic problem, we step out to the physical network to recompute what is actually feasible, and we step back in. The network is never abstracted away; it remains the substrate the solution sits on.
Figure 1. Nodal inside zonal. A single zonal transfer, here moving power from Lubbock to Permian, rendered on the real network. Left: how the model maps zonal generation and load down onto thousands of individual buses (the GSK). Right: how that transfer actually distributes across the ~7,000-line network according to physics (the PTDF). The optimization operates in zones, but the flows, limits, and costs are all measured on the lines.
This is what lets us do something no other capacity-expansion model does: select the upgrade type (reconductor, parallel greenfield, or voltage upgrade) endogenously inside the optimization, price each one from the real network rather than a single average $/MW-mile assumption, and screen every choice against N-1 reliability physics rather than handling reliability in a disconnected post-process.
What's more, it runs on public data. The entire ERCOT network was assembled from public sources — HIFLD and OpenStreetMap for topology, voltage class, and routing, with electrical characteristics inferred from voltage and length. No CEII (Critical Energy/Electric Infrastructure Information) was required. When we benchmarked the results against NLR's ReEDS data, which is built with CEII access, we found that the results tracked. Parallel efforts to produce this type of transmission dataset have been underway at Microsoft: see their GridSFM dataset.
Four pieces make it work. The boxes below walk through each one for readers who want the mechanics; skip ahead to the results if you don't.
How it works
Physics in the LP, without going nodal
Inside the optimization, transmission obeys a B-θ DC power-flow formulation at zonal granularity. Each zone carries a voltage angle (θ), each inter-zonal corridor an equivalent reactance (x_path), and a constraint of the form x_path × flow = θ_from − θ_to (with one reference zone pinned to θ = 0) enforces Kirchhoff's voltage law: power divides across parallel corridors by physics, not by whichever route looks cheapest. Hard thermal limits cap each corridor's flow. The LP stays small: ~20 zonal angles and corridor flows, not 5,000 nodal variables.
Those coefficients (each corridor's equivalent reactance and thermal limit) aren't assumed; they're measured on the real network between optimization steps. We map zonal generation and load onto the thousands of individual buses with Generation Shift Keys (GSKs), built from plant locations and least-cost dispatch, then solve a full nodal DC power flow across the ~7,000-line network and use Power Transfer Distribution Factors (PTDFs) to fit each corridor's x_path and capacity to what the network actually does.
The proof: Both reconductoring and greenfield are essential
To stress-test the method we ran 16 ERCOT cases (three load-growth levels crossed with restrictions on the transmission toolkit) all driven by a wave of new data-center demand. Bulk transmission is only 7–8% of total system investment, yet it decides whether the rest of the plan can be delivered. Watch the network build-out in the high-growth scenario: reconductoring (orange) lights up the dense eastern cities first, while greenfield (teal) and the planned 765 kV backbone (red) arrive later to carry the long-haul, west-to-east transfer:
Figure 3. ERCOT transmission buildout, 2020→2040 (high growth, all techs). ~1,300 line upgrades and ~1,180 GW of new line rating by 2040 — the buildout starts immediately and never pauses. Orange = reconductor, teal = greenfield, red = planned 765 kV.
The two tools do different jobs. Reconductoring is the speed-to-power play: existing right-of-way, roughly half the cost, fast enough to meet the near-term load wave (about three-quarters of the high-growth build is energized by 2030). Greenfield is the structural layer: it carries the long-haul bulk transfer and is the only way to reach a higher voltage class. Which tool the model picks is set by the network itself. Meshed corridors with parallel paths absorb growth through reconductoring, while radial corridors must build new:
Figure 4. Redundancy predicts the tool. Build choice as a function of how many parallel paths already exist at the corridor. Meshed corridors reconductor; radial corridors build new.
The detailed findings are in the attached study deck.
The takeaway: Why an integrated model matters
The precise Texas magnitudes will shift as data and scenarios sharpen, but the conceptual findings are durable, each one rooted in power-flow physics rather than any single cost assumption:
- Reconductoring is powerful, but it cannot stand alone.
- Greenfield is necessary.
- The physical network itself informs which upgrade a corridor needs.
The method behind those findings generalizes for the same reason: it captures how the network actually behaves, not just how ERCOT happens to be built. Put simply:
Grid expansion is dynamic and discontinuous. Like a road network, adding capacity in one place creates new constraints somewhere else. Those impacts are non-local, and only a physical power-flow model captures them. The exact mix of upgrades matters less than where and when new constraints emerge as the build-out evolves.
That is the whole reason to keep the nodal network intact rather than abstract the network away up front. It is also why this generalizes: the framework applies to any region with public network topology, and ERCOT — unusually well-suited to long-distance buildout — is in many ways a conservative case. Where greenfield is harder and slower to permit, reconductoring matters even more.
This is a proof of concept. It demonstrates that integrated capacity-expansion and power-flow modeling is not only possible but practical: fast enough for full scenario matrices, grounded enough to answer the technology-level questions planners and policymakers are actually asking.
If your team is facing transmission-upgrade, interconnection, or resource-adequacy questions that conventional capacity-expansion models can't express, we'd like to hear about them. Learn more about RIO and our modeling work at evolved.energy/models.
This work was conducted in partnership with ClearPath. Study team: Ryan Jones, Ben Preneta, Jeremy Hargreaves, Ben Haley (Evolved Energy Research); Alexandra von Meier (independent consultant); Jim Williams (University of San Francisco).
Study DeckHigh Ampacity Conductors: Opportunities to Expand Transmission — ERCOT Case Study (PDF)