More Power Will Not Make Us Wealthy: Power Into Protein and the Case for Building What We Own

This piece starts from a chart about electricity and wealth that has been circulating, and uses Iceland to test it. I rebuilt the chart from public data. The numbers and the method are at the end.

A disclosure before I begin. I run Blue Nova Ventures, a fund that invests in the Blue Economy. I have been thinking about this sector a great deal, and that is the reason for this post. Blue Nova holds no stake in any company named here.

Let me start with the chart. a16z published a plot of electricity consumption per person against GDP per person, every country a bubble, with one line written across the empty corner: “There are no low electricity, rich countries.” That caption is true and I have no quarrel with it.

My quarrel is with the conclusion people draw from it, which is that a country gets rich by producing more electricity. Iceland is the best test of that idea anywhere in the world, because nobody uses more electricity per person than we do. On the original chart we are the small unlabelled dot at the very top.

We punch far above our weight in power

World Bank (GDP, population) and Our World in Data (electricity demand). 185 countries, 2021. Bubbles sized by population.

In 2021 Iceland used about 52,000 kWh of electricity per person. That is twice Norway, four times the United States and eight times Denmark, a country with almost exactly our income. A nation of 370,000 people built that system out of rivers and geothermal fields. Credit where it is due. Few countries of any size have done anything comparable.

The power is not what made us rich

Look at where the electricity goes. In 2020, large industrial users bought 78% of everything we generated, and metal production took 94% of that, according to National Energy Authority figures reported by Morgunblaðið. Homes used under 5%. Take the large users out and Iceland’s consumption falls to about 11,400 kWh per person, which is ordinary for a Nordic country. Sweden is at 14,000.

Now look at who works there. The US Geological Survey counts about 1,800 people in metal manufacturing in Iceland, 0.9% of everyone employed. Aluminium was 37% of our goods exports in 2021. So roughly three quarters of the electricity and more than a third of goods exports run through an industry that employs fewer than one worker in a hundred.

I want to be fair to the smelters. The long contracts they signed are what justified building generation on this scale, and power sold cheaply is better than power not sold. But household wealth in Iceland was built where Icelanders work, and that is not in the potrooms. Measured as GDP per kilowatt-hour, Denmark earns about $10.70, the United Kingdom $9.70, the United States $5.80 and Norway $3.70. Iceland earns $1.35.

More electricity does not mean more income

Across 185 countries the link between electricity and income is real, and for poor countries it is steep. Nigeria has 219 million people and 165 kWh each. Nothing about a modern economy is available at that level, and for countries like Nigeria the advice to build generation is correct.

The mistake is to carry that advice up the income scale. When I fit a curve to the data instead of a straight line, a 1% rise in income goes with a 1.4% rise in electricity use at $1,000 a head, 0.9% at $10,000, and somewhere between 0.4% and 0.8% at $70,000, depending on how income is measured. The bend starts at around $5,000. Among countries above $40,000, income explains 4% of the differences in electricity use. Denmark and Iceland sit at the same income and differ by a factor of eight.

What survives is the floor. No country above $40,000 uses less than the United Kingdom’s 4,900 kWh per person. A country needs electricity to become rich. Past that floor, having more of it explains very little.

AI data centres are the next version of the same bet

The argument I now hear most often is that Iceland should sell its power to AI. It is the smelter argument with newer equipment: we have cheap, clean electricity and a cold climate, so bring the load here. I do not think it creates the wealth people expect for the community that supplies the power, and the record so far supports me.

Data centres use about 6% of the electricity we generate. The Federation of Icelandic Industries has put the figure on a par with every home in the country. For years most of that power mined cryptocurrency. A KPMG report in 2018 found that 90% of Iceland’s data centre capacity was doing so. The machines are now being swapped for AI hardware, but the business is unchanged. The customer rents power, cooling and floor space, and the value of what gets computed belongs to the customer, who is somewhere else.

Ownership tells the same story. atNorth, the largest operator here, was bought by the Swiss firm Partners Group in 2021 and sold this year to Equinix and Canada’s public pension fund in a deal worth $4.1 billion. A great deal of value was created. The seller was Swiss and the buyers are American and Canadian.

The global numbers show where the money in computing sits. Industry estimates put a data centre landlord’s revenue at $0.09 to $0.30 per kilowatt-hour, with a median near $0.18. That is about what a bitcoin miner earns and about what our smelters earn. Companies that own the chips and rent them out earn around $0.94. Meta, which owns the product, took in roughly $9 for every kilowatt-hour its data centres used in 2024. Iceland’s operators stand on the first rung.

The places that have hosted the most tell the same story. Virginia is the largest data centre market in the world. Its legislature’s auditors counted about 8,000 full-time jobs on 5,050 MW and found that most of the economic benefit comes while the buildings are going up. Two thirds of the capital goes on equipment bought from outside the state. In Ireland data centres now take 23% of the country’s electricity, and the regulator now attaches strict conditions to new connections.

I want to be fair here too. atNorth’s chief executive says its Icelandic sites employ 120 people on 25 MW, about five jobs per megawatt against roughly one at the smelters, and that 85% of them are specialists. Demand from data centres also strengthens Landsvirkjun’s hand when it negotiates with the smelters, and Landsvirkjun belongs to all of us. Those are real gains.

They fall well short of what is being promised. A hundred and twenty jobs at the largest operator is a small return on a sector that uses as much electricity as all our homes. And hosting someone else’s computers does nothing to build what Iceland is short of, which is companies of our own that turn compute into products. If AI makes us richer, it will be because Icelandic firms use it in fishing, aquaculture, energy and health. That takes very little of our power.

What a kilowatt-hour earns

If the amount of power does not explain income, the useful question is what each unit of it earns. So I compared the industry that takes most of our electricity with one that is being built now.

Conversion	Electricity used	Revenue	Revenue per kWh
Metals: aluminium and ferrosilicon	about 14.0 TWh a year	$2.43 billion of exports	about $0.17
Data centre hosting: atNorth in Iceland	25 MW, about 219 GWh a year	$72.3 million	about $0.33
Land-based salmon	6 to 10 kWh per kg	about $7 per kg	$0.70 to $1.20

A kilowatt-hour turned into land-based salmon earns roughly four to seven times the revenue of one turned into metal. Data centre hosting sits between the two, at about twice the metals figure.

The data centre row rests on one operator. atNorth’s Icelandic company reported revenue of $72.3 million for 2025, and its chief executive puts its load at 25 MW. I have assumed that load ran all year. Revenue nearly doubled in 2025, so the load was probably lower for part of it and the true figure somewhat higher.

I should be clear about how soft the salmon number is. The price is Iceland’s own 2024 average. The energy range comes from Norwegian research, because no Icelandic land-based farm has published its electricity use per kilo. The power contracts give a rough check. Thor Landeldi has secured 5 MW from ON for a 5,000-tonne phase, which works out at just under 9 kWh per kilo if the full capacity were drawn all year. That is a ceiling, and it sits inside the Norwegian range. The comparison is also revenue against revenue, with neither imported alumina nor imported feed taken out. I put the comparison to someone building one of these farms. He agreed with it and, being conservative, would put the multiple at four to five.

Power into protein

This is why land-based aquaculture interests me. Aquaculture exports reached ISK 53.8 billion in 2024, about 6% of our goods exports, on 54,800 tonnes of production, nearly all of it from sea cages. The land-based projects are of a different order. First Water has a 20 MW agreement with Landsvirkjun for its farm at Þorlákshöfn. GeoSalmo has planned for up to 24,000 tonnes. Samherji is building towards 40,000 tonnes beside the Reykjanes power plant, an investment of more than ISK 45 billion.

The Samherji project sits inside the model I think we should copy. HS Orka’s Resource Park at Reykjanes has more than ten companies living off one power plant: its electricity, its waste heat and its warm seawater. By HS Orka’s own account, 84% of the energy the cluster uses is waste heat. Stolt Sea Farm has harvested sole there since 2014, grown in 32°C water from the plant mixed down to 22°C. Each business uses a stream that would otherwise run into the sea.

Land-based salmon will not employ as many people as fishing or tourism. First Water expects about 300 people at Þorlákshöfn at full production of 50,000 tonnes, and Samherji expects 100 direct jobs at Reykjanes. Set against the electricity used, though, that is several times more work per kilowatt-hour than the smelters provide. The rest of the case is what aquaculture can give us more of than the smelters did: value per kilowatt-hour, owners who live here, and a supply chain of feed, equipment, processing and software that Icelandic companies can build and then sell abroad.

The highest-value conversion used almost no power

Kerecis takes cod skin, which used to be thrown away, and turns it into wound-care products in Ísafjörður. In 2023 Coloplast agreed to buy the company for up to $1.3 billion. Its revenue the year before was DKK 510 million, 98% of it from the United States. The electricity involved is trivial.

I have spent seventeen years working with founders in this country, and Kerecis is the pattern I would bet on. The raw material was Icelandic and nearly free. The value came from people who knew what to do with it. Cheap, clean power makes that kind of company easier to build here, but power alone has never started one.

What to ask of the next terawatt-hour

Power in Iceland is now scarce. When Landsvirkjun signed its term sheet with GeoSalmo, its chief executive said demand for the company’s energy “is much more than we can meet” and that it had turned projects down. Every new contract is therefore a choice between uses, and we should make it with open eyes.

I would ask three questions of any project that wants our power. What does it earn per kilowatt-hour? Who owns it, and where do the profits go? What does it leave behind in the community in skills, suppliers and new companies?

The smelters answered the first question poorly and the second from abroad. Data centres do better than the smelters on jobs per megawatt and no better on ownership. Land-based aquaculture, the Resource Park and companies like Kerecis give better answers, and they are being built now. We are moving in the right direction.

We have already done the hard part, which was building the power. What we turn it into is still ours to decide.

Sources and notes

All figures are for 2021 unless I say otherwise.

• Income and population: World Bank GDP per capita, GNI per capita, GDP per capita at purchasing power and population.
• Electricity: Our World in Data per-capita electricity demand, cross-checked against the World Bank’s electric power consumption per capita. The two series correlate at 0.997. The original chart cites the US Energy Information Administration, which I could not access.
• Iceland without large users: the 2020 share (78%) applied to the 2021 total. My estimate.
• Metals row: 78% x 94% of 19,127 GWh generated in 2020, against 2021 exports of $2.24 billion of aluminium and $189 million of ferrosilicon (USGS). The years differ by one.
• Salmon row: ISK 47.7 billion of salmon exports on 49,300 tonnes produced in 2024, at about 138 krónur to the dollar. Energy from an NTNU model of recirculating grow-out (6 to 10 kWh per kg) and a RaboResearch figure for hybrid flow-through (up to 8).

How I tested the bend. Ordinary least squares on logs, each country counted once. A straight line gives an elasticity of 0.98. Adding a squared term gives a curvature coefficient of -0.28 with a t-statistic of -5.0. A two-segment fit puts the break near $5,000, with an elasticity of 1.4 below and 0.7 above.

GDP per person	Countries	Elasticity	95% interval	R²
Under $3,000	57	1.16	0.72 to 1.60	0.33
$3,000 to $12,000	66	0.71	0.28 to 1.14	0.14
$12,000 to $40,000	38	0.74	0.34 to 1.13	0.28
Over $40,000	24	0.34	-0.41 to 1.09	0.04

The top band alone cannot tell a flat line from a proportional one. The evidence for the bend is the curvature across the whole sample, and it holds in every version I ran:

Specification	Countries	Curvature t-stat	Elasticity at $1k	at $10k	at $70k
GDP, Our World in Data electricity	185	-5.0	1.44	0.88	0.41
Same, without Iceland	184	-5.2	1.45	0.87	0.37
Same, populations over 1 million	154	-5.4	1.49	0.85	0.32
GNI, Our World in Data electricity	184	-4.7	1.43	0.88	0.42
GDP at purchasing power, Our World in Data electricity	183	-4.4	1.98	1.34	0.79
GDP, World Bank electricity	141	-3.9	1.39	0.88	0.45
GNI, World Bank electricity	141	-3.9	1.39	0.87	0.44

What I could not test. Rich service economies import electricity-intensive goods, and Iceland exports them. Counting electricity where goods are consumed would raise the United Kingdom and Denmark and lower Iceland, and could steepen the line at the top. This is also a single year, so it shows how countries differ, not what happens to one country as it grows.