One hundred and five megawatts

I read interconnection paperwork the way other people read match reports. It is a strange habit and I recommend it, because the files answer a question the announcements never do. Everyone can tell you which company has the best chips. Almost nobody can tell you which of the announced machines will ever be plugged in, and by whom, and in what year. That answer is not in a keynote. It is in a filing cabinet, and this month the cabinet has been unusually loud.

What the local law actually says

The city is Plattsburgh, New York. On 20 August its mayor introduced Local Law P-2 of 2026, and the hearing on it was held at half past five on 3 September in the council chambers at 41 City Hall Place. The title is not shy: a local law establishing a temporary land use moratorium on the establishment and expansion of certain high energy computing facilities pending completion of the city’s comprehensive zoning update.

Read the definition before you form an opinion of it. A High Energy Use Facility is any commercial or industrial facility “primarily devoted to cryptocurrency mining, blockchain validation, artificial intelligence computing, cloud computing, machine learning, digital asset processing, data center operations, server farm operations, colocation facilities, or other substantially similar high-density computing operations” which requires high density load service under the city code, or which has “a connected electrical demand of three hundred (300) kilowatts or greater.” Three hundred kilowatts. That is the line. Below it you are a business; above it you are a category of land use the city has not written rules for yet.

And that is what the law is for. For twelve months, no officer or board of the city may issue a building permit, zoning permit, special use permit, site plan approval or certificate of occupancy for establishing or expanding one of these facilities. But the same section says that nothing in it “shall prohibit the continued lawful operation” of a facility already running lawfully on the day it takes effect. There is a hardship clause: on written application and after a public hearing, the council may grant relief. And the council’s own findings say why the pause exists at all — the city’s existing zoning “does not contain land-use standards specifically designed to address the distinctive characteristics and potential impacts of such facilities, including their unusually high electrical demand, utility infrastructure requirements, cooling requirements, noise, emergency response and fire protection needs.”

This is not a ban on artificial intelligence. It is a city admitting, in public, on its own website, that a use has arrived which its rulebook has no page for, and buying itself a year to write the page. The list of what it intends to study during that year reads like an inventory of everything nobody worked out in advance: zoning districts, electrical demand thresholds, noise standards, cooling systems, fire protection, battery energy storage, decommissioning requirements, infrastructure reimbursement, financial assurance, cumulative impacts.

Now the number in the headline. Plattsburgh owns and operates its own electric utility, the Municipal Lighting Department, and that is the reason any of this is the city’s business at all. The council’s findings put it this way: “Unlike many municipalities, the establishment and operation of exceptionally large electrical users may directly affect the City’s utility planning, electrical infrastructure, reserve capacity, capital improvements, system reliability, municipal finances, and the efficient delivery of electric service.” Mayor Wendell Hughes told the local broadcaster WCAX what he is protecting: “Our total allotment for the City of Plattsburgh is 105MW. So, that’s a half of the usage in the city of Plattsburgh. You know, we have to protect the city of Plattsburgh,” Hughes said. The station reports that a new crypto or data centre facility, in his account, can use 50 megawatts.

Take the two figures at their plainest. The city’s entire allocation of electricity is 105 megawatts. One building would want 50 of them. The mayor’s middle sentence is doing something I cannot resolve from the transcript — a half of which usage, measured when — so I will leave it where he put it rather than tidy it into a cleaner statistic. The proportion survives the ambiguity, and the proportion is the entire argument of this piece. We are six paragraphs in, so let me spend the rest of it on the parts that are less obvious.

The first less obvious part: the mayor does not claim these facilities are raising anyone’s bills. WCAX reports the opposite — he “adds that the high-energy facilities don’t contribute to higher utility rates for residents.” The same thing happened the first time. In March 2018 Plattsburgh became, as WCAX put it then, “the first city in the country to put a moratorium on cryptocurrency”; that one lasted eighteen months, applied only to new operations, and passed unanimously. Residents at the time argued about bills. The city’s electric department told the station that the spikes were “likely due to cold weather.”

So twice now, eight years apart, a small city has paused new computing load while saying out loud that the computers were not the reason your bill went up. The pause is not about price. It is about an allocation of a fixed size and a queue of people who want it.

The same day, the other end of the same problem

On 3 September the US Department of Energy issued an emergency order under Section 202(c) of the Federal Power Act, authorising Duke Energy Carolinas to dispatch specified units and to order their operation as needed to maintain reliability during hot weather, and to direct backup generation as a last resort. It ran from 3 September to 8 September. The department’s own announcement notes that the president “declared a national energy emergency” on day one of the administration.

I want to be careful here, because this is where the argument I was asked to make either earns its keep or falls over. The suggestion I keep meeting — and it is a serious one — is that Western countries need to reach for extraordinary powers, a wartime-scale national programme, to build generation fast enough for what is coming. Look at what that reaching produces in practice. The emergency power exists. It is being used. What it does is keep existing plants running for five days in a heatwave. It is a mechanism for not going dark, and there is nothing in it that builds anything.

Now the measurements, and what they count

Before any of these numbers, the boring paragraph that makes them usable. Two things get counted here and they are not the same. Capacity is how much a fleet of power stations could deliver at once, measured in gigawatts — it is a nameplate, and a nameplate flatters a system with a lot of solar in it, because solar is not delivering at night. Generation, or demand, is how much electricity actually moved over a year, measured in terawatt-hours. One terawatt-hour is a billion kilowatt-hours. When the two ratios disagree, the disagreement is the information.

China’s National Energy Administration published its figures for January to July 2026 on 25 August. At the end of July, total installed generating capacity across the country stood at what the agency writes as 40.8亿千瓦 — 亿 is the Chinese unit for a hundred million and 千瓦 is the kilowatt, so that is 4.08 billion kilowatts, or 4,080 gigawatts — up 11.0% on a year earlier, with solar at 1,290 GW (up 16.1%) and wind at 690 GW (up 19.5%). The United States Energy Information Administration reports that at the end of 2025 the United States had 1,280,799 MW, about 1,281 GW, of total utility-scale generating capacity, plus an estimated 59,506 MW of small-scale solar counted separately. The American figure covers plants of at least one megawatt and is reported, in most EIA products, as net summer capacity. The Chinese page gives a single national total for installed generating capacity and states no seasonal rating convention at all. They are close enough to compare and not close enough to subtract.

On capacity, then, China’s fleet is a bit over three times the American one. On electricity delivered, the gap is smaller: Ember’s Global Electricity Review 2026 puts China’s 2025 demand at 10,573 TWh against 4,536 TWh for the United States — a third of all the electricity used on earth, in one country, for the first time. Three times the metal, twice the electricity. That difference is solar sitting idle at night, and anyone quoting you only the capacity ratio is selling you something.

The measurement that decides this is not a stock but a rate. In 2025, Ember reports, electricity demand grew by 503 TWh in China. In the United States it grew by 131 TWh. In the European Union, by 13.5 TWh. In Japan, by 13.4 TWh. Ember puts China’s share of the world’s extra electricity plainly: it “led global demand growth in 2025, for the tenth year running, adding 503 TWh (+5.0%). This accounted for 59% of the global net increase in electricity demand (+849 TWh).” Tenth year running.

And the supply kept up. China’s solar generation alone rose by 336 TWh in a single year — more, Ember notes, than the entire world added in 2023, and more than the United Kingdom uses in a year. It was four times the next largest increase, which was the American one at 85 TWh; the EU added 60. On new hardware, China installed 58% of the world’s solar in 2025 (378 GW measured DC) and 72% of the world’s wind (119 GW). The EU added over 65 GW of solar and 15 GW of wind. Japan added 570 MW of wind, after 510 MW the year before. Its second-largest annual rise on record was in wind generation, which is a different measurement from the 570 MW and I nearly ran the two together myself. The American entry in that last comparison is generation rather than new hardware, and those two are not interchangeable either.

Set the American demand growth of 131 TWh next to the Chinese solar growth of 336 TWh and you have the shape of the thing. One country’s new panels, in one year, produced more than two and a half times all the extra electricity the United States consumed. Ember writes that data centres are projected to account for around half of total US demand growth through 2030. Half of 131 is not a number that wins a race.

One more figure belongs here, because it cuts against the story I have been telling you and you should have it. The International Energy Agency’s projections, as Ember reports them, have data centres at around 500 TWh in 2025 rising to around 950 TWh in 2030 — growing four times faster than electricity demand overall. And then: “even at this rapid pace, data centres would account for less than 3% of global electricity demand by 2030, though regional impacts will differ significantly.” Ember adds that the estimate for 2025 carries high uncertainty.

Less than three per cent. So the thing being fought over is not the world’s electricity. It is a small, concentrated, extremely fast-arriving slice of it, landing on particular substations in particular counties, and the last four words of that quotation — regional impacts will differ significantly — are where Plattsburgh lives. A city with 105 megawatts does not care what percentage of global demand its problem represents.

The plan says it in writing

The argument that the AI contest will be decided by electricity rather than by silicon is usually delivered as a provocation by someone with shares in it. It does not have to be. One of the two governments involved has written it down as policy, with a document number.

On 21 July 2026 the National Development and Reform Commission and the National Energy Administration jointly issued the Fifteenth Five-Year Plan for the Construction of a New-Type Power System (发改能源〔2026〕942号), published on 3 August. Its headline table sets out where the system is meant to be in 2030. Total installed capacity: 38.9 亿 kilowatts in 2025, rising to 54 亿 kilowatts by 2030. In the units used above, that is a plan to go from 3,890 GW to 5,400 GW — to add about 1,510 GW in five years. The entire American fleet, as counted by the EIA, is 1,281 GW. The plan is to add more than that, on top, by 2030. The same table takes non-fossil generation from 42.3% to 50% of the mix, west-to-east transmission from 340 GW to over 420 GW, and new-type storage from 136 GW to 300 GW.

I have read a lot of energy strategies and most of them are wishes with a font. What makes this one different is not the ambition. It is section thirty-seven, which is titled 促进电力与算力协同发展 and reads, in part: “增强电力供应保障能力，满足各类算力设施高可靠供电需要。统筹能源资源配置与算力设施建设，协同规划布局算力、电力项目，做到以电强算、以算促电。” My own translation, and I am printing the original because you should be able to check me: promoting the coordinated development of electricity and computing power. Strengthen the security of electricity supply and meet the requirement of highly reliable power for computing facilities of every kind. Coordinate the allocation of energy resources with the construction of computing facilities, and plan and lay out computing and electricity projects together, so as to strengthen computing through electricity and promote electricity through computing.

Whatever else you think about that sentence, notice what it is not. It is not a forecast, and it is not an appeal to industry. It is a paragraph in an implementing instruction sent to provincial commissions and named enterprises, telling them that data centres and power stations are one planning problem and are to be laid out on one map. Plattsburgh spent this month discovering that they are one planning problem. The difference between the two documents is not conviction, and it is not engineering. It is that one of them was written before the buildings arrived.

So is it the system?

From this paragraph to the end of the section I am giving you my opinion rather than a document, and I will say where it stops.

I think the framing I was handed — open societies are structurally too slow, one-party states are structurally too fast, and the gap is the future — is half right in a way that makes it more dangerous than if it were wrong. It is right that speed is the variable. It is wrong about which mechanism is producing the speed, and getting that wrong leads to bad prescriptions.

What the documents in front of me show is not a contest between democracy and its absence. It is a contest between a country where the electricity system and the computing system are planned by the same office on the same timetable, and countries where they are planned by different people, at different levels of government, in different decades, and reconciled afterwards by whoever is holding the permit. Plattsburgh’s ordinance is not the voice of a slow society. It is the sound of a decision landing on the only body that was legally obliged to have an answer and had never been asked the question. The council did not obstruct anything. It discovered it was the last stop.

Two things follow that I would want anyone reaching for emergency powers to sit with. The first is that the Western bottleneck is not generation and is not law in general; it is permission and sequence — the queue, the study, the zoning that was written for a different century. You can suspend a permit process by decree. You cannot decree a transformer into existence, and the emergency power that actually exists, as we saw on 3 September, dispatches old plants rather than building new ones. The second is that Plattsburgh is not being slow. It is proposing to write, in twelve months, the rules that every other jurisdiction in this story also does not have. A city that has noticed the problem and given itself a deadline is doing more than the headline about it suggests.

Opinion ends there. Back to what I can hand you.

Is fusion the way out?

Two facts about ITER, from ITER, that between them dispose of most of what is said about it. The first: “seven partners—China, the European Union, India, Japan, Korea, Russia and the United States—have pooled their financial and scientific resources to build the biggest fusion reactor in history.” It is not a Western programme. China is inside it. The second, from the same paragraph: “ITER will not produce electricity, but it will resolve critical scientific and technical issues in order to take fusion to the point where industrial applications can be designed.” The organisation explains elsewhere that a generator was considered and left out, because the experimental pattern of operation “will allow for very limited generation of electricity” and the cost would not have been worth it.

The schedule is public too. Under the 2024 baseline, full magnetic energy comes in 2036, three years later than the 2016 reference, and the deuterium–tritium phase starts in 2039, four years later. That is the machine that does not make electricity reaching its main experiment in the year 2039.

Meanwhile the Institute of Plasma Physics of the Chinese Academy of Sciences describes its own follow-on, CFETR, as aiming “to demonstrate fusion energy production up to 200 MW initially and to eventually reach DEMO relevant power level which is over 1 GW… and net electricity generation by fusion”, supported by a national facility, CRAFT, whose construction the institute’s page says began on 20 September 2019. Those are the objectives an institute has published for itself, not a date on which anything will be delivered.

And then there is the detail that made me want to write this piece at all. On 28 April 2026, Commonwealth Fusion Systems announced that it had become the first fusion company to apply to PJM Interconnection — a market serving about 182,000 MW of capacity across thirteen states and the District of Columbia. Its 400-megawatt ARC plant is to be built in Chesterfield County, Virginia. The company’s explanation of why it filed now, years before there is a plant, is the plainest sentence in this whole article: submitting the request is a long-lead action “because the start of the study process to the generation of electricity can take four to six years.” Its chief executive, Bob Mumgaard, put it as: “when you’re serious about building a power plant in the early 2030s, you act now.”

Four to six years, from beginning the study to sending power. Not from breaking ground — from beginning the study. A fusion company that has solved none of your problems yet has already worked out that the physics is not what will make it late, and got into the line. If fusion is the answer, fusion is standing in the same queue as everyone else, and it knows it.

What nobody has said

• How much is waiting in the queue. Not one of the documents here says how much generating capacity is sitting in American interconnection queues, unbuilt, waiting for a study to finish. That total, rather than any single plant, is the size of the bottleneck, and it is missing from every announcement on both sides of this story.
• Where Plattsburgh’s 105 megawatts comes from, and on what terms. The mayor calls it the city’s allotment from the state. The ordinance never mentions it. Neither document I opened explains how the allocation is set or whether it can be enlarged.
• A build rate measured the same way for everyone. China’s, the EU’s and Japan’s 2025 installations are published in gigawatts in one place; the American figure in that comparison is generation. Until the same measure is published for all four, growth in demand is the fairer column.
• Whether the 2030 targets will be met. They are targets in a government document, published by the government that set them. This piece has not verified a single megawatt of them.
• What any of this does to a household bill. Two mayors of the same small city, eight years apart, have both declined to blame the computers. That is two data points from one city.

Watch it land

Every stream here is the broadcaster’s own official channel. The money side of this story runs on CNBC, Bloomberg Television and Yahoo Finance long before it reaches a bulletin, and the American hearings themselves turn up on C-SPAN. For the other side of the comparison, CGTN is China’s English-language broadcaster; this site marks it as state-affiliated in the channel list, and you should read it as what a government says about itself, which is exactly why it is worth hearing on this subject. Or open the full channel list and choose by region.

Sources

• City of Plattsburgh, New York, “Public Hearing — Local Law P-2 of 2026” — the full text of the proposed local law: the definition, the 300 kW threshold, the twelve-month pause, the exemption for existing operations, the hardship clause and the list of matters to be studied.
• WCAX, “Plattsburgh considers short-term ban on crypto ventures” (Alek LaShomb, 8 September 2026) — Mayor Hughes on the 105 MW allotment, the 50 MW figure, and his statement that the facilities are not raising residents’ rates.
• WCAX, “Plattsburgh becomes first in nation to ban bitcoin mining” (Kelly O’Brien, 16 March 2018) — the 2018 moratorium, its eighteen-month term, and the electric department on cold weather.
• US Department of Energy, “Energy Secretary Secures Carolinas’ Grid Ahead of Holiday Weekend” (3 September 2026) — the Section 202(c) emergency order and its dates.
• Ember, Global Electricity Review 2026 (21 April 2026, PDF) — 2025 demand growth for China, the US, the EU and Japan; solar and wind generation growth; installation shares; total demand by country; China’s 59% share of the global increase; and the IEA data-centre projections quoted, including the “less than 3%” figure.
• US Energy Information Administration, “Electricity generation, capacity, and sales in the United States” — the 1,280,799 MW figure, the small-scale solar estimate, and the definitions of utility-scale and net summer capacity.
• National Energy Administration of China, national electricity statistics for January–July 2026 (25 August 2026) — installed capacity at the end of July and the year-on-year growth rates.
• National Development and Reform Commission and National Energy Administration of China, notice issuing the Fifteenth Five-Year Plan for the Construction of a New-Type Power System (document 942 of 2026, dated 21 July 2026, published 3 August 2026); the plan itself is the attached PDF — the 2030 target table and section thirty-seven, quoted above in the original.
• ITER Organization, frequently asked questions — the seven partners, “ITER will not produce electricity”, the reason no generator was fitted, and the 2036 and 2039 dates in the 2024 baseline.
• Institute of Plasma Physics, Chinese Academy of Sciences, CRAFT — the CFETR objectives and the CRAFT construction date, as the institute states them.
• Commonwealth Fusion Systems, announcement of its PJM interconnection application (28 April 2026) — the four-to-six-year figure, the 400 MW ARC plant in Chesterfield County, the size of PJM, and the Mumgaard quotation.

Every figure and quotation above was taken from the pages linked here on 9 September 2026. Chinese documents are quoted in the original and translated by the writer, who is not a professional translator; the originals are printed so you can check. Capacity figures from China and the United States are compiled on different definitions, described in the text, and should not be subtracted from one another. The International Energy Agency’s own report pages and the Lawrence Berkeley National Laboratory’s figures on interconnection queues could not be opened from here; the IEA projections quoted above therefore reach this piece through Ember, which is said in the text, and no figure for the size of the interconnection queue appears at all. Where this piece states an opinion, it says so and says where it ends.