America's Energy Problem: We Need A New Grid
Summary
America’s grid problem is institutional atrophy as much as physical capacity. Ryan McEntush says the system “effectively froze” in the early 2000s as manufacturing shifted to Asia, leaving operators without the workforce or skill to plan and execute large projects quickly or cheaply. David Ulevitch adds the stark comparison: U.S. per-capita energy use peaked in 1973, while China’s energy use increased ninefold over the same period.
Distributed generation and storage could leapfrog a grid where interconnection can take a decade and transformer backlogs reportedly exceed 20 years. David describes the grid as roughly century-old technology operating near capacity; Erin Price-Wright argues for placing solar, batteries and load together, bypassing some interconnection and delivery infrastructure whose costs keep rising even as generation gets cheaper. Data centers already embody the urgency: “I can’t afford to wait 10 years… I need this power now, today.”
Texas is the panel’s proof point for deploying cheap solar and batteries at speed. Erin says the state roughly doubled solar capacity in three years and added thousands of batteries, improving its ability to absorb sharp demand changes; Ryan argues every state should study ERCOT’s distributed model, while acknowledging that regulated markets make replication harder. Erin describes the manufacturing dependence on China, while David warns that losing Chinese battery supply “could be catastrophic… in a very, very short period of time.”
The winning energy mix is “yes, and,” not a single-resource bet. Erin’s personal bet is that solar and batteries will remain cheap and fast, but she says gas, nuclear, geothermal and hydro will remain necessary because the long-tail cost of intermittency becomes severe once variable resources reach roughly 50%-75% of the grid. Her broader point is about load design: expanding data centers, EVs, heat pumps and autonomy will enlarge both baseload and daily peaks, so the system must not oversolve for either.
Flexible compute is a more plausible demand-response asset than centrally controlling Americans’ thermostats. David says consumers will “flatly reject” dictated indoor temperatures, while Erik points to crypto mining and David suggests shifting noncritical data-center jobs to cheaper hours. David also pushes back on Erin’s view that data-center demand may be overstated, arguing society may instead be underestimating AI electricity consumption across the next 10-50 years.
Grid software is the clearest venture-scale layer in an otherwise capital-heavy rebuild. Erin explains that the grid lacks the internet’s bidirectional communication, data layer and control plane; David argues that software should enter from the edges rather than through a slow top-down utility rollout. The grid lacks “no Splunk,” “no Palo Alto Networks” and “no Looker,” while bottom-up telemetry from batteries, chargers and distributed generation could improve forecasting that still relies heavily on weather. Erin says AI could also turn permitting work performed by armies of consultants over months or years into “minutes or hours.”
Nuclear combines baseload, resilience and defense value, but scaling it requires rebuilding America’s megaproject muscle. David highlights Radiant Nuclear’s proposed 1-megawatt, truck-transportable microreactor and separately imagines transportable reactors being flown on C-130s. He contrasts that flexibility with military fuel that can cost more than $200—and sometimes $400—per gallon to deliver. The closing priorities are categorical: Ryan says there is no safety, national defense or national security without a reliable electrical grid; Erik says energy policy should prioritize power that is “cheap, reliable and clean—in that order.”
Deep dive
1. America forgot how to expand its grid
Ryan’s history begins with a simple model: build a large power plant, connect transmission to substations, and distribute electricity to factories and homes. Growth slowed in the 1980s and 1990s, then the grid “effectively froze” as heavy industry and manufacturing moved to Asia.
Erik Torenberg’s pushback—worth keeping: was America prohibited from building, or did it genuinely lose the capability? Ryan’s answer is the latter; operators forgot “how to plan, how to move quickly, how to do it cheaply,” with nuclear providing the most extreme example.
Now manufacturing, data centers and other concentrated loads are returning with demand that is “now, now, now” and almost price-insensitive. The system must relearn large-project execution while redesigning itself around technologies that do not require thermal-plant scale.
2. Delivery bottlenecks make colocated power economically compelling
David’s diagnosis: the grid is century-old technology operating near capacity. A new project’s interconnection may take a decade, while he cites a 20-plus-year transformer backlog, effectively one company making the transformers and one U.S. plant producing the required electrical steel.
That leads to the episode’s central question: “Should we just leapfrog the grid?” Erin argues that solar and batteries can be installed beside demand, allowing data centers to build power on-site and bypass interconnection rather than waiting for centralized infrastructure to catch up.
Colocating generation, storage and consumption also creates a tractable optimization problem. David argues reinforcement learning could coordinate these local systems far more efficiently than is possible across a grid whose physical state is poorly observed.
Ryan contrasts Texas’s “connect and manage” approach—build where desired, accepting that the grid may curtail output—with states that model numerous peak scenarios before approving a continuously connected project. Sensors and other grid-enhancing technologies could expose unused capacity on lines that may average roughly 50% utilization but must accommodate summer peaks.
3. Texas makes the decentralized-grid case—and exposes battery risk
David’s resilience framing borrows from distributed computing: local solar, batteries, microgrids or small reactors let users survive failures in an interdependent grid. For military bases and data centers, independence from brittle transmission is becoming a requirement rather than a luxury.
Erin points to Texas, which she says doubled solar capacity in approximately three years and deployed thousands of batteries after major grid failures. During the heat wave discussed, that combination reportedly gave ERCOT more elasticity than New York and neighboring systems without requiring an overnight expansion of gas or nuclear baseload.
Erin says the United States invented lithium-ion batteries but now depends on batteries made in Chinese factories or by Chinese companies in Vietnam. David adds that startups cut off by Chinese suppliers have found alternative procurement “very, very difficult.”
Ryan wants other states to adopt the lesson without pretending implementation is uniform. Utilities resist colocation, and residential solar remains so burdened by permitting and installation costs that he believes it is cheaper to install in Germany than in the United States.
4. The grid needs every resource—and more responsive demand
Erin’s energy doctrine is “yes, and.” Despite her enthusiasm for solar and storage, she stresses her Palantir oil-and-gas background and argues that America needs every available tool deployed where it makes technical and economic sense.
Erin expects solar and batteries to keep gaining because they are cheap and fast, but not to displace everything. Beyond roughly 50%-75% of the grid, rare periods without sufficient output can make intermittency very expensive; gas, nuclear, geothermal and hydro therefore remain necessary forms of baseload or dispatchable power.
Erin sees load becoming more complicated: data centers are generally baseload, while EVs, heat pumps, air conditioners and industrial autonomy widen peaks and troughs. Erik’s example is a gas peaker operating perhaps one week annually at around $10,000 per megawatt-hour when every cheaper resource is exhausted.
Erik proposes demand response, including thermostats moving a few degrees; David says consumers will reject that level of control. David prefers shifting noncritical compute, while Erik cites crypto mining as already price-responsive. David nevertheless disputes Erin’s view that data-center growth may be overstated: “We actually probably are underestimating” long-run compute demand.
The discussion is notably skeptical of wind: it can be cheap when operating but is difficult to plan around, may fail to produce for extended periods, and is hard and dangerous to service. David says he read that one-third of wind turbines are out of service globally at any given time.
5. Grid software must enter from the edges
Erin notes that the electrical grid lacks the internet’s bidirectional messaging, data layer and control plane. Monitoring and messaging are often handled out-of-band over the internet because native signaling is so limited.
David doubts a startup can sell a top-down platform to a utility such as PG&E and achieve rapid deployment. His preferred strategy is to place software “almost insidiously on the grid,” gathering telemetry near individual loads and generators before connecting those observations into a broader picture.
Erin says day-ahead dispatch still depends heavily on weather, household locations and population. With solar systems, batteries and EV chargers producing real-time data, operators could forecast loads, locate capacity and price electricity more efficiently; today, the weather specialist may be the highest-paid person on an energy desk besides the portfolio manager.
David’s request for startups is explicit: build the grid equivalents of Splunk, Palo Alto Networks and Looker, covering logging, cyber defense, analytics, orchestration and demand-response markets. Erin and David also identify site selection, permitting, supply-chain coordination and project management for developments involving thousands of interdependent workers and suppliers.
6. Nuclear’s opportunity is shifting from bespoke plants to manufactured reactors
David says public sentiment changed over the past three or four years as nuclear became widely recognized as clean energy, though political resistance remains. He says Taiwan’s decision to close its final reactor is especially risky because, in his estimate, an oil-and-gas blockade could leave the island roughly seven days from blackout. The discussion attributes the closure to political pressure, and a speaker calls it “colossally stupid.”
The regulatory burden spans plant design, fuel production, transport and storage, requiring tens of millions of dollars, thousands of pages and “an army of consultants.” One speaker also argues that spent fuel should be discussed less reflexively as “waste,” since almost all of it can be recycled and reused.
Large AP1000 reactors remain bespoke megaprojects; David says the Georgia units were about 10 years late and multiple billions of dollars over budget. Erik’s broader point is unforgiving: if factory-built modules still require major on-site assembly, “you’re still a construction company.”
Radiant Nuclear represents the alternative: David describes its proposed 1-megawatt microreactor as factory-produced and transportable on an 18-wheeler. Separately, he imagines a transportable reactor being flown on a C-130, dropped where needed and providing power for five years. He contrasts that flexibility with military diesel deliveries costing over $200 and sometimes $400 per gallon.
7. Rebuilding the grid is an industrial-policy project
Ryan says true grid-scale capital needs are not merely billions but “tens of billions, hundreds of billions.” China is the uncomfortable benchmark: after common blackouts in the early 2000s, it roughly quadrupled its grid through hydro, storage, battery manufacturing and high-voltage direct-current transmission.
Erin argues that venture capital cannot finance the webbing of multibillion-dollar megaprojects alone, but technology can improve every phase: site selection, permitting, construction logistics and coordination across a project with 4,000 workers. She says AI could make a regulatory application perhaps 85% reusable, then flag project-specific differences for both applicant and regulator.
Labor is already binding. David laments that specialized Vogtle crews returned to highways and bridges instead of building reactors five through 10; Erik says Microsoft’s Georgia data-center project at one point employed or contracted more than one-third of the state’s electricians.
Ryan’s final warning is that solving one bottleneck merely reveals another: transmission requires transformers, transformers require electrical steel, and battery cells require active materials and mining. He makes the national-security case that a reliable, dispatchable grid is foundational, not ancillary, to defense.