Loviisa 3 would already be producing electricity if the Centre Party had not obstructed Fortum’s permit application in 2010. The plant would have been built without subsidy schemes.
The doubling of electricity consumption in the near future is strongly doubted. What is that estimate based on? Data centers will indeed come, but they won’t double electricity consumption. Hydrogen projects are not progressing, and industrial consumption is rather decreasing.
Electricity consumption asymptotically approaches infinity when everything runs on electricity. Trains, cars, motorcycles, mopeds, and bicycles, for a start. Hand tools like saws, screwdrivers, etc., already run on battery power and are charged frequently. Even toothbrushes are already electric, etc…
Now, let’s build that nuclear reactor before the power crisis surprises us.
It won’t double with current projections, but they might have been overly optimistic… However, the consumption of future Microsoft data centers alone is already quite large, over 600 megawatts, so at least one reactor would be needed. Fortum will likely supply the electricity there.
The estimated increase in electricity consumption in future data centers is strongly based on the use of artificial intelligence; for example, it has been estimated in the US that electricity consumption will increase by the amount of Germany’s consumption in the future, of which a relatively large portion is taken by data centers and especially power-hungry AI chips.
History is full of these explosive electricity consumption forecasts. Cryptocurrency mining was supposed to use the entire world’s electricity production already in 2020 - as predicted by Newsweek and the World Economic Forum in 2017. Before cloud services, it was predicted that on-premise data centers would consume all electricity - now Power Usage Effectiveness (PUE) has doubled in data centers in ten years. So, there has been plenty of electricity for other things despite the predictions. In Finland, electricity consumption has barely changed in 10 years, and I would be surprised if billions of capital suddenly flowed here into new electricity consumption. Google and Microsoft are nice exceptions to the rule.
Data centers are a capital-intensive business - many other constraints arise before the power runs out of the socket. Furthermore, AI GPU capability improves faster than power requirements. Blackwell GPUs consume 75% less power for training the same model than the previous Hopper architecture, even though a single chip consumes 70% more watts.
Liebreich describes very different scenarios well in his analysis. From the perspective of Fortum and other energy companies, it is interesting how Exxon and others plan to efficiently implement CO2 capture with gas power plants in their data center concepts.
Regarding Finland, replacements (in one way or another) will eventually be needed for the current reactors when they reach the end of their operational life. This means that at some point, action must be taken anyway, preferably sooner rather than later, of course.
Well, of course not. Electricity consumption can rise steadily as electrification increases, but it won’t become asymptotically infinite in any situation. Maximum consumption is limited, at least for Earth. But that consumption is also practically limited by price. If production cannot be increased, the price will also rise steadily, which continuously reduces consumption to reach some equilibrium – not infinity.
Updated view. 
Updated view.
That new nuclear power plant is being explored for Loviisa.
Finland needs both new steadily operating base load power and regulating power. Time will tell how much.
P.S. The German state is holding sales discussions for Uniper with at least 4 parties. Nothing should be counted on that, but when the sale proceeds, it will be interesting to see what happens to Uniper’s Swedish production and especially to the joint power plants with Fortum.
Nvidia’s impending collapse tonight (now nearly -10% in futures), the double-digit plunge of Swedish AI subcontractors and data center companies (Munters etc.) on the Stockholm stock exchange this morning, etc. And what does this have to do with Fortum?
DeepSeek’s emergence from the bushes to disrupt the AI sector is significant from Fortum’s perspective because the same AI outputs can be achieved with only a fraction of the computing power and electricity consumption. At least, this appears to be the case based on the information at hand.
From the perspective of Fortum or other electricity generation companies, the situation should not necessarily be over-dramatized. Although AI with its data centers etc. is a growing electricity consumer, its share of total consumption is limited. DeepSeek may also expand and generalize the use of AI even further, partly due to its affordability, which in turn increases electricity consumption.
Fortum’s owners should indeed closely monitor the development, and I will be watching the share price of Brookfield Renewable, which has ridden the AI electricity consumption wave, tonight.
I would be more concerned if I had mega-techs, SP500, or world indices (which are largely mega-techs) in my portfolio. We might be on the verge of a major tech disruption - or maybe not 
The biggest impact could be on investor sentiment.
It is critical to note that this DeepSeek disruption applies only to AI model training.
DeepSeek’s solution does not change the situation regarding Inference (the use of AI models). So, when Microsoft runs CoPilot on its servers for its customers, even if the model were Chinese, the consumption of computing power and electricity remains the same as with current OpenAI offerings.
Naturally, as the market is currently in a panic, many players do not yet realize this, and forecasts are being rapidly revised. And naturally, AI model training is also a significant driver, but even on that side, the final effect is still very unclear. If a current-level model can be trained with a fraction of the electricity consumption and hardware because “team Chinese” invented something great, what happens if a model is trained with the same system on a huge hardware stack of a large megacap? Presumably, it should still scale, and the model would be even better. Only then, if there is certain information that the trick does not scale, can AI training hardware and electricity consumption forecasts be cut.
I don’t understand much about IT myself, but I have observed that at least since computers transitioned from vacuum tubes to microprocessors, their computing power and energy consumption have increased.
Whenever computing power has increased, new uses have always been found for it, and/or the scope and accuracy of models/services have been improved.
New individual innovations, whether on the software or hardware side, can certainly at least postpone investments, and perhaps some company will forgo an investment if it can cost-effectively purchase the service it needs, produced in a new way.
It’s unlikely that the dip in AI firms will significantly affect Fortum’s outlook. It’s actually amusing that the share price drops 4% in the wake of AI firms. A buying opportunity if it drops further from here. Not a cent invested in AI firms; I knew to stay away when pump-and-dumps started appearing in almost every Yahoo Finance article, even those dealing with completely different stocks.
Fortum and the Finnish green hydrogen pioneer P2X Solutions Oy have signed a five-year power purchase agreement. The majority of hydrogen in Europe is still produced from fossil natural gas, but the agreement now signed enables the production of green hydrogen in accordance with strict EU requirements at P2X’s Harjavalta plant, which is currently starting its production. Electricity will be produced at Fortum’s renewable energy plants in Finland, and the agreement also covers guarantees of origin related to electricity production. The agreement period lasts until the end of 2029.
According to the EU regulation on renewable hydrogen (RFNBO), the production of green hydrogen and renewable energy must be matched monthly until 2029, after which, according to current regulations, production and consumption must occur simultaneously with hourly accuracy. The agreement now made covers the amount of renewable energy required by P2X Solutions’ green hydrogen production plant in its initial phase. With this agreement, RFNBO-compliant green hydrogen production will begin in Finland.
One could have perhaps also stated in such an announcement how much electricity is being sold. Now it tastes a bit like hype.
Well, it depends a bit on how you compare. The truth is that computational power per energy has only improved over the years. See, for example, Koomey’s law: Koomey's law - Wikipedia
Room-sized vacuum tube computers consumed an incredibly large amount of energy while still being astonishingly less efficient than even the smallest modern computers. A transistor does what a vacuum tube does but with many orders of magnitude less consumption.
Overall consumption has increased because computers are used so much and because of digitalization. Computers are everywhere and used for everything. But in itself, energy efficiency has improved tremendously and continues to improve.
However, the pace has slowed, and Moore’s law no longer works linearly as the laws of physics begin to push back.
Here is a Q4 preview report crafted by a man from Sotkamo 
Fortum will publish its release next Tuesday.
According to our assessment, Fortum’s last quarter remained relatively weak operationally, as electricity prices stayed moderate during a seasonally strong period, and there were challenges in the availability of both nuclear and hydropower. We have revised our Q4 forecasts downwards and simultaneously lowered our forecasts for the coming years due to the decrease in electricity futures since the previous update. We will not make changes to our view before the financial statement release on February 11th, but we see slight downward pressure on our target price along with the forecasts. On the other hand, electricity futures have fluctuated strongly and recently risen from their lows.
There’s a bit of a negative tone in the preview report again. Dividends are also gnawing, even though Inderes itself, as a growth company, proposes a dividend payout of 92% of its earnings. Perhaps even Inderes’ owners need money for kitchen renovations 
It’s true that the operational result will be weak. Still, it’s good to remember that Fortum is now a net debt-free company, and the adventures of world conquest have finally ended. I have already received half of the average share price I paid for my shares back in dividends. If we compare Fortum to Elisa, the dividends are significantly more substantial, and Elisa is up to its ears in debt.
Q4 is an interesting quarter because it includes the latter dividend payment as an expense and divestments as income, the largest of which is that 800 million crap plant. Partially, the losses generated by this plant will be gone, and Fortum’s own cost-saving measures will also start to bite.
We reiterate our dividend target of €1.30, because we, the main owners, need money 
The latest consensus has not yet been published by Vara Research. I will post the figures as soon as they arrive on the computer.
Sweden’s Forsmark 3 nuclear power reactor will shut down on Wednesday due to a problem with a valve, and is expected to be back online on Friday, its main owner Vattenfall said on Tuesday
These appeared yesterday. Our analyst @Juha_Kinnunen brought the figures in such a book-shaped ATK printout. @Sijoittaja-alokas happened to be visiting, so we took the opportunity and asked Juha to explain what P/E means.
We already know other investment-related terms, but this is the most difficult, even though we have studied investment literature extensively. We still didn’t quite understand it in eight hours.
The consensus does not promise any flattering results, but we believe in miracles and fear how many shorts Marshall Wace will push onto the sell side.
Let’s hope for the best and fear the worst 