Facing America’s rapidly growing AI-driven electricity demand, U.S. capacity expansion is still too slow.
In 2025, U.S. investment in clean energy, clean transportation, building electrification, and carbon management reached $278 billion, a record high. The problem is that, compared with China, this number still looks small. In the same year, China’s investment in clean-tech manufacturing and deployment reached $849 billion, about 3.1 times the U.S. level.
The structural difference is clear: the United States faces a demand curve growing faster than its institutional deployment capacity, while China faces an installation curve growing faster than its system absorption capacity.
The U.S. added 43.4 GW of solar, 24 GW of battery storage, and 11.8 GW of wind in 2025. But China is expanding at a completely different speed. In 2025, China added roughly 315 GW of solar, about 8 times the U.S. level, and around 119 GW of wind, about 10 times the U.S. level.
The United States is clearly seeing a boom in solar and storage construction, but it is still expanding within a relatively constrained institutional speed limit. China, by contrast, has entered an industrial-scale deployment phase in which several hundred gigawatts of new renewable capacity can be added in a single year.
This is the deeper difference. U.S. clean energy expansion is driven mainly by capital markets, state-level incentives, corporate PPAs, and private-sector demand. China’s expansion reflects a much more systemic mobilization: manufacturing capacity, grid investment, local governments, state-owned enterprises, private firms, supply-chain cost reduction, and national energy-security strategy all moving in the same direction.
America’s electricity gap is therefore not just a power-supply gap. It is also a national system-organization gap. The United States has technology, capital, corporate demand, and innovation capacity. But when it tries to convert these advantages into large-scale, low-cost, rapidly deployed infrastructure, it runs into permitting delays, interconnection bottlenecks, interstate coordination problems, transmission constraints, equipment supply-chain shortages, and political-cycle friction.
These are the problems the U.S. needs to solve quickly. Otherwise, if residential electricity prices rise sharply again, data centers and hyperscalers will almost certainly face large-scale political resistance from local communities. That is the biggest risk.
This is an interesting point and I’d add one additional note: you mention corporate PPAs. Tech companies have historically been among the biggest contractors of renewable energy PPAs, resulting in a strange situation where hyperscale data center operators are, on the one hand, driving demand for new gas generation that probably wouldn’t have been built otherwise and contributing to overall upward price pressure, but also providing one of the biggest sustained sources of capex for more renewable generation (particularly as federal support pulls back).
Yes, I think this is the right complication to add.
Tech companies are indeed major buyers of renewable PPAs. Their credit quality, long-term power demand, and capital capacity make it easier for many wind, solar, and storage projects to secure financing. This matters a lot, especially when federal subsidies, tax support, or policy certainty are weakening. Corporate PPAs can become a private-sector anchor for clean energy investment.
But the root tension is that AI data centers need 24/7, low-latency, highly reliable power that can be connected quickly. Wind, solar, transmission, storage, and nuclear all move on much longer construction timelines, and they are constrained by permitting, interconnection queues, transmission bottlenecks, and local politics. As a result, hyperscalers can sign renewable PPAs on one side while also pushing the system to add gas capacity in the short term.
This is not because they are “hypocritical.” It is because the time scale of AI power demand does not match the time scale of power-system expansion.
So this is fundamentally an infrastructure governance problem. The issue is not whether hyperscalers have green commitments. The issue is whether the U.S. can turn the huge new load from AI into a coordinated expansion of clean firm power, transmission, storage, demand response, and grid flexibility.
Otherwise, the result will be fragmented: tech companies support renewables through PPAs, the grid still relies on gas to fill the near-term gap, cost pressure is shifted onto other ratepayers, and system expansion remains poorly coordinated.
Right. Agreed with all this. A couple questions that I’d be curious for your thoughts on.
Do you think the turbine backlog impacts the timeline calculations as far as go-to source substantially, either on the utility or data center side? In the latter case what we’re seeing is repurposing of other equipment (e.g., jet engines) into turbines and then combining that with plans for overcapacity of onsite storage (sometimes 2-3x the generating capacity of the gas) to maintain reliability/uptime requirements when using the Jerry-rigged gas builds.
On the utility side it seems to me like trying to speed interconnection of the large amounts of solar in the queue and then firming with storage would make more sense, speed-wise, than new turbines but the degree to which that’s being pursued seems to vary. You see a lot more of it in ERCOT vs PJM, and I suspect this has to do with the respective regulations/cost recovery models in the different regions.
Beyond PPAs it seems like the hyperscalers are each kind of pursuing their own model for speed to power right now (with Google out on one pole as far as leaning way in on renewables + long-duration storage or geothermal wells etc, xAI on the other pole with whatever gas they can get their hands on, and the others somewhere in the middle). Do you think this will last or do you think it’s likelier they converge on the same model before long?
And the current dispersion in hyperscaler strategies is unlikely to remain as fragmented as it is today over the long run, but neither are they likely to converge on a single uniform model. What seems more likely is a kind of limited convergence. In the near term, each company will continue to pursue a different mix depending on project location, state-level regulation, interconnection timing, corporate ESG constraints, and tolerance for higher capital or operating costs.
Over the medium term, they will probably begin to converge around the same broad framework: first secure speed to power, then gradually fold temporary gas, backup generation, and onsite storage into a more standardized hybrid architecture.
In other words, the end state is unlikely to be a purely Google-style clean firming model, but it is also unlikely to become an xAI-style “whatever gas you can get” approach across the board. The more plausible outcome is some combination of grid access, renewables PPAs, storage, and a certain amount of dispatchable thermal or firm clean resource, with the weighting varying by company.
Turbine backlogs will materially weaken gas as a universal fast-answer solution, but they will not eliminate gas’s appeal in specific local use cases. My sense is that for data center projects that genuinely require 24/7 uptime and are willing to pay a high premium for speed to power — especially those able to pursue onsite generation and tolerate higher costs — aeroderivatives, repurposed equipment, and temporary gas builds will continue to play a role. Cases like xAI may be extreme, but they show that under severe time constraints, the market will accept a system that is inelegant but functional.
At the same time, the longer turbine backlogs persist, the more utility-scale projects — and even some data-center-side projects — will be pushed toward a hybrid model built around solar, storage, grid access, and selective firming. Once new gas turbine delivery is no longer fast enough to dominate every other consideration, the speed advantage of solar and storage becomes much more attractive, especially in market structures like ERCOT. In other words, turbine scarcity itself may end up strengthening the relative competitiveness of renewables plus storage in certain regions.
Thanks—that’s basically my sense too, fwiw. I think the BTM builds will trend toward some mix of gas/storage/solar (or other fast-to-build onsite source) and will also be geared toward eventual interconnection in a majority of gases vs permanent islanding.
Yes, this is an important distinction. What Germany has legalized is not exactly the same as conventional residential rooftop solar. It is closer to plug-in balcony solar: small, standardized systems, often around the sub-kilowatt scale, that can be installed by renters or apartment dwellers and connected through certified microinverters.
China has a huge residential and distributed solar market, but it is mostly rooftop-based and grid-connected, not Germany-style plug-in balcony solar. In 2024, China added about 118 GW of distributed PV, including nearly 30 GW of residential PV, and rooftop installations surged again in early 2025. That is already an enormous deployment machine. But China’s model is still closer to “distributed power-station deployment” than “household appliance deployment.”
The U.S. also has residential rooftop solar, but the market is much smaller and more expensive. In 2025, U.S. residential solar additions were only around 4.6 GW, while total U.S. solar additions were about 43 GW. More importantly, plug-in balcony solar remains trapped in regulatory ambiguity in much of the U.S. Utilities often treat even very small plug-in systems like full interconnection projects.
So I think your point is exactly right: the U.S. does not only need more utility-scale solar and transmission. It also needs a new regulatory category for small, certified, zero-export plug-in solar systems. That would not solve the entire electricity gap, but it would create a fast, low-friction household layer of clean power, especially for renters and apartment dwellers. This is precisely the kind of deployment speed the U.S. system currently lacks.
1.) Wjat is capacity? Real capacity should be thought of as the amount of power that can be generated over an extended period, a year, for example, with the plant running at the maximum rate consistent with long asset life.
Yes, many other things matter in generation, but, in discussion of supply meeting rapidly growing demand, what matters is the capacity discussed in the first paragraph.
2.) The chart in this piece shows the US adding no capacity, none, from coal or nuclear this year.
For context, the PRChina added >60% of total US coal capacity in just 14 months from Jan 2025 to Feb 2026. They added 1 Australia's coal generation capacity in just two months at the beginning of this year.
3.) Nameplate capacity, as is commonly used, is terribly misleading for energy sources with inherently low capacity factors. Solar in less favorable locations exemplifies this.
When one looks at real capacity, rather than illusory nameplate capacity, we see that coal was the number one source of new generation capacity in the PRChina in 2025. Yes. More real capacity was added than real capacity from solar.
The US hasn't built a new coal-fired power plant in ~15 years despite possessing enormous quantities of cheap coal.
4.) Batteries are not generation. They are fuel tanks. They add zero real capacity.
5.) When thinking about supply and demand over an extended period, it's useful to think about watt-hours, not watts.
Capacity is independent of how it is used. That a simple cycle peaker sits idle 90% of the time tells you about its utilization, not about its capacity. To calculate its utilization, you must know its real capacity (point 1 above).
Batteries play the same role in electricity that tanks play for liquid fuels. Neither tanks nor batteries have capacity to produce. When the tank runs dry, it must be filled. The same is true for batteries. They are eminently useful, but they are not production capacity.
What gets built is driven by the rules of taxation and subsidy and permitting.
Electricity consumption should be growing a high single digit rates — and be priced to do so.
Capacity should be built ahead of that demand, resulting in substantial available excess production capacity at all times.
The value of storage tanks in such a system is much lower than in a system with little available excess production capacity.
I enjoyed this comment, some interesting takes. Looking at capacity as sustained annual output though I believe is misguided. A simple-cycle peaker sits idle 90% of the year, yet we all appreciate its value to capacity. What makes a resource count is showing up in the scarcity window. For an intermittent or duration-limited resource, the real capacity value is a number the market has already calculated. I agree nameplate inflates it, but the zero-capacity claim erases it. PJM’s 59% ELCC for a four-hour battery quantifies that number directly: it sits near the average winter loss-of-load event but runs short of the six-hour-plus tail, and every four-hour battery hits empty at the same moment. The number begins to capture these resources’ real contribution.The real question underneath that curve is whether PJM’s duration premium — the jump from 59% to 78% as you go from four to ten hour resources— is steep enough to actually pull developers toward building longer-duration storage, or whether four-hour economics still win on installed cost even at the lower accreditation.
The explosion in data center investment and the resulting energy gap highlights the exact structural blind spot in our current macroeconomic design. We are watching a $1T AI and infrastructure boom scale at 21st-century speeds, yet we are trying to manage and absorb its massive resource demands using a legacy, debt-backed baseline.
When a nation's infrastructure and technology undergo a paradigm shift, the economic plumbing must shift with it. Tinkering with standard corporate tax rates or relying on debt-financed federal subsidies to bridge the electricity gap creates a compounding fiscal drag.
The structural solution requires transitioning to an asset-backed architecture. By implementing a native equity layer—specifically a Sovereign Investment Fund fueled by a Universal Revenue Tax—the immense productivity and capital gains from the AI boom can be captured directly to anchor the national balance sheet. This turns a massive infrastructure strain into a compounding public asset.
Version 1.9 of the National Growth Compact maps out the exact mathematics for this institutional transition. It’s entirely live and open-source for a data-driven economic audit. Available here on Substack.
Great piece Joey. Something related I’m watching is whether there will be any impact to utility or grid operator planning from the deployment of substantial 4-8 hour battery capacity data centers becoming nearly universal, which is something that both energy analysts and DC operators told me they are seeing in new projects this year (for reasons including management of facility level power fluctuations and providing excess backup for BTM power, including gas turbines). Combine this with the proliferation of demand response programs, some of them mandatory, and potential forthcoming NERC guidance that DCs should “ride through” grid disturbances rather than islanding and you soon have a large source of storage available to the grid beyond what utilities are building. I have no idea whether this will impact utility planning or, if so, how, but it seems worth watching. I’d be curious for your thoughts.
Good comment. The data center angle is really interesting. Their load is essentially price-inelastic — they’ll curtail only when forced, so the demand response they provide is the mandatory kind, not an economic one. That makes the on-site storage the real question. Does BTM data center storage ever become a capacity resource PJM can accredit and count, or does it stay invisible to the capacity market? If it gets counted, it compresses the ELCC value for everyone else’s batteries. If it doesn’t, it’s reliability the market can’t monetize. Either way, it changes the merchant storage investment case.
Agreed. I’m not well versed in PJM capacity accreditation rules—do you have a sense of what sort of rule change would be needed to allow them to accredit and count BTM data center storage?
Accreditation in PJM really just boils down to the resource's ability to deliver when the system is at highest risk, and a binding commitment to show up when called (penalties if you don't). PJM weights resource classes on top of that differently based on reliability (ie. Nuclear would have a higher factor than intermittent solar) but the obligation is the part that matters here.
For BTM storage to get counted there either has to be a capacity price incentive or a rule mandate. The price signal would have to clear a very high bar and that is actually currently being suppressed due to the FERC price collar that was put in place, or the capacity market here would be clearing much higher than it did recently. If PJM chooses mandate and stops chasing price signal, they could just make capacity contribution a condition of interconnection.
There's already some traction for the mandate route under Bring Your Own Generation out of PJM's Critical Issue Fast Path (CIFP) process. This is more of a carrot than a stick as they are expediting projects that are committing to helping to solve this capacity strain. If data centers are driving the strain, the case for requiring them to bring relief is hard to argue with. It can be chalked up to being a cost of doing business.
I'd be interested in hearing about legalized plug-in solar. Germany has it and it's helped reduce stress on that nation's power grid. It lowers the knowledge barrier to solar access: if you can plug an appliance into an electrical outlet, you can use plug-in solar.
The "commercial power growth in 4 years > prior two decades" line is the one I think most aren't pricing yet. A capacity gap of that kind is supply-side, and monetary policy has no tool that closes it - every additional GW of hyperscaler load drops straight into the electricity component of CPI. Curious whether the EIA's 4.6% generation projection already includes the queue past 2024 (xAI Memphis, Anthropic Ohio, etc.) or just announced load through last year.
The “data centers are commercial load growth” part of your electricity story is the same binding constraint I keep bumping into when people hand-wave “agents will pay in stablecoins” without pricing power, wires, and tariffs. I connect that physical-world constraint back to the money-stack thesis (GENIUS + agents + BTC reserve thinking) from my side of the desk here: https://thiagopedicosaragiotto.substack.com/p/the-genius-revolution-dollar-stablecoins
And the Wall Street contribution is the $67 billion deal where Florida-based NextEra Energy acquires Dominion Energy. Creates the largest regulated electric utility by market value. And adds 0 KWH to supply. And invests $0 in the grid.
Ooops—- THE US MILITARY has been using SOLAR POWERED WEAPONS for years…
U.S. police and military forces have developed and deployed advanced technology to detect shallow-draft and stealthy drug vessels, often referred to as "go-fast" boats, low-profile vessels (LPVs), and narco-submarines.
The U.S. uses the following specific technologies to combat this maritime threat:
Unmanned Surface Vehicles (USVs): The U.S. Navy and the Defense Innovation Unit deploy long-range, solar-powered drones like the Saildrone Voyager, which scan thousands of square miles to autonomously track dark, low-signature targets and transmit real-time intelligence to multi-agency task forces
Demand growth keeps front running supply. The interesting question is which utility names are quietly compounding rate base while everyone chases AI hyperscalers.
This is cool. I would be great to also look into transmission infrastructure. In NY a total electricity bill has >50% of the cost covering for transmission rather than electricity production. The production cost is a fraction of the total bill and this aspect is one of the key driver of the cost.
Shaving $0.02 of the electricity production cost while the transmission increases by 10% isn't going to reduce the total cost.
NY isn't that much of an expensive state compare to HW, RI, CA, MA.
Commercial load growth, the EIA category that captures data centers, is now expected to exceed industrial and residential growth combined. That reframes the AI capex story: GPUs are not the only constraint. Grid interconnection and transformer supply are the bottlenecks most balance sheets are not pricing in. Texas alone capturing 55% of new US battery capacity is part of why ERCOT power generation is on track to be 57% above pre-COVID levels by 2027.
Yes, power is the part of the AI story that refuses to stay abstract. Capital can move faster than interconnection queues, and that mismatch changes which founders are actually building with time on their side.
Also tough to build more and to create a nationwide project of electrifying the country when the President is obsessed with oil and spends his time paying off Total Energies to stop building their windmills off the coast.
Facing America’s rapidly growing AI-driven electricity demand, U.S. capacity expansion is still too slow.
In 2025, U.S. investment in clean energy, clean transportation, building electrification, and carbon management reached $278 billion, a record high. The problem is that, compared with China, this number still looks small. In the same year, China’s investment in clean-tech manufacturing and deployment reached $849 billion, about 3.1 times the U.S. level.
The structural difference is clear: the United States faces a demand curve growing faster than its institutional deployment capacity, while China faces an installation curve growing faster than its system absorption capacity.
The U.S. added 43.4 GW of solar, 24 GW of battery storage, and 11.8 GW of wind in 2025. But China is expanding at a completely different speed. In 2025, China added roughly 315 GW of solar, about 8 times the U.S. level, and around 119 GW of wind, about 10 times the U.S. level.
The United States is clearly seeing a boom in solar and storage construction, but it is still expanding within a relatively constrained institutional speed limit. China, by contrast, has entered an industrial-scale deployment phase in which several hundred gigawatts of new renewable capacity can be added in a single year.
This is the deeper difference. U.S. clean energy expansion is driven mainly by capital markets, state-level incentives, corporate PPAs, and private-sector demand. China’s expansion reflects a much more systemic mobilization: manufacturing capacity, grid investment, local governments, state-owned enterprises, private firms, supply-chain cost reduction, and national energy-security strategy all moving in the same direction.
America’s electricity gap is therefore not just a power-supply gap. It is also a national system-organization gap. The United States has technology, capital, corporate demand, and innovation capacity. But when it tries to convert these advantages into large-scale, low-cost, rapidly deployed infrastructure, it runs into permitting delays, interconnection bottlenecks, interstate coordination problems, transmission constraints, equipment supply-chain shortages, and political-cycle friction.
These are the problems the U.S. needs to solve quickly. Otherwise, if residential electricity prices rise sharply again, data centers and hyperscalers will almost certainly face large-scale political resistance from local communities. That is the biggest risk.
This is an interesting point and I’d add one additional note: you mention corporate PPAs. Tech companies have historically been among the biggest contractors of renewable energy PPAs, resulting in a strange situation where hyperscale data center operators are, on the one hand, driving demand for new gas generation that probably wouldn’t have been built otherwise and contributing to overall upward price pressure, but also providing one of the biggest sustained sources of capex for more renewable generation (particularly as federal support pulls back).
Yes, I think this is the right complication to add.
Tech companies are indeed major buyers of renewable PPAs. Their credit quality, long-term power demand, and capital capacity make it easier for many wind, solar, and storage projects to secure financing. This matters a lot, especially when federal subsidies, tax support, or policy certainty are weakening. Corporate PPAs can become a private-sector anchor for clean energy investment.
But the root tension is that AI data centers need 24/7, low-latency, highly reliable power that can be connected quickly. Wind, solar, transmission, storage, and nuclear all move on much longer construction timelines, and they are constrained by permitting, interconnection queues, transmission bottlenecks, and local politics. As a result, hyperscalers can sign renewable PPAs on one side while also pushing the system to add gas capacity in the short term.
This is not because they are “hypocritical.” It is because the time scale of AI power demand does not match the time scale of power-system expansion.
So this is fundamentally an infrastructure governance problem. The issue is not whether hyperscalers have green commitments. The issue is whether the U.S. can turn the huge new load from AI into a coordinated expansion of clean firm power, transmission, storage, demand response, and grid flexibility.
Otherwise, the result will be fragmented: tech companies support renewables through PPAs, the grid still relies on gas to fill the near-term gap, cost pressure is shifted onto other ratepayers, and system expansion remains poorly coordinated.
Right. Agreed with all this. A couple questions that I’d be curious for your thoughts on.
Do you think the turbine backlog impacts the timeline calculations as far as go-to source substantially, either on the utility or data center side? In the latter case what we’re seeing is repurposing of other equipment (e.g., jet engines) into turbines and then combining that with plans for overcapacity of onsite storage (sometimes 2-3x the generating capacity of the gas) to maintain reliability/uptime requirements when using the Jerry-rigged gas builds.
On the utility side it seems to me like trying to speed interconnection of the large amounts of solar in the queue and then firming with storage would make more sense, speed-wise, than new turbines but the degree to which that’s being pursued seems to vary. You see a lot more of it in ERCOT vs PJM, and I suspect this has to do with the respective regulations/cost recovery models in the different regions.
Beyond PPAs it seems like the hyperscalers are each kind of pursuing their own model for speed to power right now (with Google out on one pole as far as leaning way in on renewables + long-duration storage or geothermal wells etc, xAI on the other pole with whatever gas they can get their hands on, and the others somewhere in the middle). Do you think this will last or do you think it’s likelier they converge on the same model before long?
And the current dispersion in hyperscaler strategies is unlikely to remain as fragmented as it is today over the long run, but neither are they likely to converge on a single uniform model. What seems more likely is a kind of limited convergence. In the near term, each company will continue to pursue a different mix depending on project location, state-level regulation, interconnection timing, corporate ESG constraints, and tolerance for higher capital or operating costs.
Over the medium term, they will probably begin to converge around the same broad framework: first secure speed to power, then gradually fold temporary gas, backup generation, and onsite storage into a more standardized hybrid architecture.
In other words, the end state is unlikely to be a purely Google-style clean firming model, but it is also unlikely to become an xAI-style “whatever gas you can get” approach across the board. The more plausible outcome is some combination of grid access, renewables PPAs, storage, and a certain amount of dispatchable thermal or firm clean resource, with the weighting varying by company.
Turbine backlogs will materially weaken gas as a universal fast-answer solution, but they will not eliminate gas’s appeal in specific local use cases. My sense is that for data center projects that genuinely require 24/7 uptime and are willing to pay a high premium for speed to power — especially those able to pursue onsite generation and tolerate higher costs — aeroderivatives, repurposed equipment, and temporary gas builds will continue to play a role. Cases like xAI may be extreme, but they show that under severe time constraints, the market will accept a system that is inelegant but functional.
At the same time, the longer turbine backlogs persist, the more utility-scale projects — and even some data-center-side projects — will be pushed toward a hybrid model built around solar, storage, grid access, and selective firming. Once new gas turbine delivery is no longer fast enough to dominate every other consideration, the speed advantage of solar and storage becomes much more attractive, especially in market structures like ERCOT. In other words, turbine scarcity itself may end up strengthening the relative competitiveness of renewables plus storage in certain regions.
Thanks—that’s basically my sense too, fwiw. I think the BTM builds will trend toward some mix of gas/storage/solar (or other fast-to-build onsite source) and will also be geared toward eventual interconnection in a majority of gases vs permanent islanding.
Legalize plug-in solar as Germany has done. Ease the grid stress!
Yes, this is an important distinction. What Germany has legalized is not exactly the same as conventional residential rooftop solar. It is closer to plug-in balcony solar: small, standardized systems, often around the sub-kilowatt scale, that can be installed by renters or apartment dwellers and connected through certified microinverters.
China has a huge residential and distributed solar market, but it is mostly rooftop-based and grid-connected, not Germany-style plug-in balcony solar. In 2024, China added about 118 GW of distributed PV, including nearly 30 GW of residential PV, and rooftop installations surged again in early 2025. That is already an enormous deployment machine. But China’s model is still closer to “distributed power-station deployment” than “household appliance deployment.”
The U.S. also has residential rooftop solar, but the market is much smaller and more expensive. In 2025, U.S. residential solar additions were only around 4.6 GW, while total U.S. solar additions were about 43 GW. More importantly, plug-in balcony solar remains trapped in regulatory ambiguity in much of the U.S. Utilities often treat even very small plug-in systems like full interconnection projects.
So I think your point is exactly right: the U.S. does not only need more utility-scale solar and transmission. It also needs a new regulatory category for small, certified, zero-export plug-in solar systems. That would not solve the entire electricity gap, but it would create a fast, low-friction household layer of clean power, especially for renters and apartment dwellers. This is precisely the kind of deployment speed the U.S. system currently lacks.
It would make it so much easier for apartment and condominium dwellers to charge their EVs at home.
Exactly! You know America's EV public charge is so poor. 1 charger serves 35 cars while in China this number is 9 cars.
1.) Wjat is capacity? Real capacity should be thought of as the amount of power that can be generated over an extended period, a year, for example, with the plant running at the maximum rate consistent with long asset life.
Yes, many other things matter in generation, but, in discussion of supply meeting rapidly growing demand, what matters is the capacity discussed in the first paragraph.
2.) The chart in this piece shows the US adding no capacity, none, from coal or nuclear this year.
For context, the PRChina added >60% of total US coal capacity in just 14 months from Jan 2025 to Feb 2026. They added 1 Australia's coal generation capacity in just two months at the beginning of this year.
3.) Nameplate capacity, as is commonly used, is terribly misleading for energy sources with inherently low capacity factors. Solar in less favorable locations exemplifies this.
When one looks at real capacity, rather than illusory nameplate capacity, we see that coal was the number one source of new generation capacity in the PRChina in 2025. Yes. More real capacity was added than real capacity from solar.
The US hasn't built a new coal-fired power plant in ~15 years despite possessing enormous quantities of cheap coal.
4.) Batteries are not generation. They are fuel tanks. They add zero real capacity.
5.) When thinking about supply and demand over an extended period, it's useful to think about watt-hours, not watts.
Capacity is independent of how it is used. That a simple cycle peaker sits idle 90% of the time tells you about its utilization, not about its capacity. To calculate its utilization, you must know its real capacity (point 1 above).
Batteries play the same role in electricity that tanks play for liquid fuels. Neither tanks nor batteries have capacity to produce. When the tank runs dry, it must be filled. The same is true for batteries. They are eminently useful, but they are not production capacity.
What gets built is driven by the rules of taxation and subsidy and permitting.
Electricity consumption should be growing a high single digit rates — and be priced to do so.
Capacity should be built ahead of that demand, resulting in substantial available excess production capacity at all times.
The value of storage tanks in such a system is much lower than in a system with little available excess production capacity.
I enjoyed this comment, some interesting takes. Looking at capacity as sustained annual output though I believe is misguided. A simple-cycle peaker sits idle 90% of the year, yet we all appreciate its value to capacity. What makes a resource count is showing up in the scarcity window. For an intermittent or duration-limited resource, the real capacity value is a number the market has already calculated. I agree nameplate inflates it, but the zero-capacity claim erases it. PJM’s 59% ELCC for a four-hour battery quantifies that number directly: it sits near the average winter loss-of-load event but runs short of the six-hour-plus tail, and every four-hour battery hits empty at the same moment. The number begins to capture these resources’ real contribution.The real question underneath that curve is whether PJM’s duration premium — the jump from 59% to 78% as you go from four to ten hour resources— is steep enough to actually pull developers toward building longer-duration storage, or whether four-hour economics still win on installed cost even at the lower accreditation.
The explosion in data center investment and the resulting energy gap highlights the exact structural blind spot in our current macroeconomic design. We are watching a $1T AI and infrastructure boom scale at 21st-century speeds, yet we are trying to manage and absorb its massive resource demands using a legacy, debt-backed baseline.
When a nation's infrastructure and technology undergo a paradigm shift, the economic plumbing must shift with it. Tinkering with standard corporate tax rates or relying on debt-financed federal subsidies to bridge the electricity gap creates a compounding fiscal drag.
The structural solution requires transitioning to an asset-backed architecture. By implementing a native equity layer—specifically a Sovereign Investment Fund fueled by a Universal Revenue Tax—the immense productivity and capital gains from the AI boom can be captured directly to anchor the national balance sheet. This turns a massive infrastructure strain into a compounding public asset.
Version 1.9 of the National Growth Compact maps out the exact mathematics for this institutional transition. It’s entirely live and open-source for a data-driven economic audit. Available here on Substack.
Great piece Joey. Something related I’m watching is whether there will be any impact to utility or grid operator planning from the deployment of substantial 4-8 hour battery capacity data centers becoming nearly universal, which is something that both energy analysts and DC operators told me they are seeing in new projects this year (for reasons including management of facility level power fluctuations and providing excess backup for BTM power, including gas turbines). Combine this with the proliferation of demand response programs, some of them mandatory, and potential forthcoming NERC guidance that DCs should “ride through” grid disturbances rather than islanding and you soon have a large source of storage available to the grid beyond what utilities are building. I have no idea whether this will impact utility planning or, if so, how, but it seems worth watching. I’d be curious for your thoughts.
Good comment. The data center angle is really interesting. Their load is essentially price-inelastic — they’ll curtail only when forced, so the demand response they provide is the mandatory kind, not an economic one. That makes the on-site storage the real question. Does BTM data center storage ever become a capacity resource PJM can accredit and count, or does it stay invisible to the capacity market? If it gets counted, it compresses the ELCC value for everyone else’s batteries. If it doesn’t, it’s reliability the market can’t monetize. Either way, it changes the merchant storage investment case.
Agreed. I’m not well versed in PJM capacity accreditation rules—do you have a sense of what sort of rule change would be needed to allow them to accredit and count BTM data center storage?
Accreditation in PJM really just boils down to the resource's ability to deliver when the system is at highest risk, and a binding commitment to show up when called (penalties if you don't). PJM weights resource classes on top of that differently based on reliability (ie. Nuclear would have a higher factor than intermittent solar) but the obligation is the part that matters here.
For BTM storage to get counted there either has to be a capacity price incentive or a rule mandate. The price signal would have to clear a very high bar and that is actually currently being suppressed due to the FERC price collar that was put in place, or the capacity market here would be clearing much higher than it did recently. If PJM chooses mandate and stops chasing price signal, they could just make capacity contribution a condition of interconnection.
There's already some traction for the mandate route under Bring Your Own Generation out of PJM's Critical Issue Fast Path (CIFP) process. This is more of a carrot than a stick as they are expediting projects that are committing to helping to solve this capacity strain. If data centers are driving the strain, the case for requiring them to bring relief is hard to argue with. It can be chalked up to being a cost of doing business.
I'd be interested in hearing about legalized plug-in solar. Germany has it and it's helped reduce stress on that nation's power grid. It lowers the knowledge barrier to solar access: if you can plug an appliance into an electrical outlet, you can use plug-in solar.
The "commercial power growth in 4 years > prior two decades" line is the one I think most aren't pricing yet. A capacity gap of that kind is supply-side, and monetary policy has no tool that closes it - every additional GW of hyperscaler load drops straight into the electricity component of CPI. Curious whether the EIA's 4.6% generation projection already includes the queue past 2024 (xAI Memphis, Anthropic Ohio, etc.) or just announced load through last year.
The “data centers are commercial load growth” part of your electricity story is the same binding constraint I keep bumping into when people hand-wave “agents will pay in stablecoins” without pricing power, wires, and tariffs. I connect that physical-world constraint back to the money-stack thesis (GENIUS + agents + BTC reserve thinking) from my side of the desk here: https://thiagopedicosaragiotto.substack.com/p/the-genius-revolution-dollar-stablecoins
And the Wall Street contribution is the $67 billion deal where Florida-based NextEra Energy acquires Dominion Energy. Creates the largest regulated electric utility by market value. And adds 0 KWH to supply. And invests $0 in the grid.
Ooops—- THE US MILITARY has been using SOLAR POWERED WEAPONS for years…
U.S. police and military forces have developed and deployed advanced technology to detect shallow-draft and stealthy drug vessels, often referred to as "go-fast" boats, low-profile vessels (LPVs), and narco-submarines.
The U.S. uses the following specific technologies to combat this maritime threat:
Unmanned Surface Vehicles (USVs): The U.S. Navy and the Defense Innovation Unit deploy long-range, solar-powered drones like the Saildrone Voyager, which scan thousands of square miles to autonomously track dark, low-signature targets and transmit real-time intelligence to multi-agency task forces
Demand growth keeps front running supply. The interesting question is which utility names are quietly compounding rate base while everyone chases AI hyperscalers.
This is cool. I would be great to also look into transmission infrastructure. In NY a total electricity bill has >50% of the cost covering for transmission rather than electricity production. The production cost is a fraction of the total bill and this aspect is one of the key driver of the cost.
Shaving $0.02 of the electricity production cost while the transmission increases by 10% isn't going to reduce the total cost.
NY isn't that much of an expensive state compare to HW, RI, CA, MA.
Joseph, thanks for this report. It appears you meant to write "production" in the first sentence and not "consumption".
Speaking of energy... https://thetellstoffel.substack.com/p/the-leverage-equation
Tax the billionaires until they're millionaires and pump it into infrastructure in all sectors. If they leave, they leave. Oh no.
Commercial load growth, the EIA category that captures data centers, is now expected to exceed industrial and residential growth combined. That reframes the AI capex story: GPUs are not the only constraint. Grid interconnection and transformer supply are the bottlenecks most balance sheets are not pricing in. Texas alone capturing 55% of new US battery capacity is part of why ERCOT power generation is on track to be 57% above pre-COVID levels by 2027.
Yes, power is the part of the AI story that refuses to stay abstract. Capital can move faster than interconnection queues, and that mismatch changes which founders are actually building with time on their side.
Also tough to build more and to create a nationwide project of electrifying the country when the President is obsessed with oil and spends his time paying off Total Energies to stop building their windmills off the coast.