Podcast · #paasipodi #73 · 2026-05-14 · 58:32 · In Finnish
Warming Up for the Tampere Investment Fair — Sami Miettinen on Paasipodi #73
Published as “Tampereen sijoitusmessujen etkot ft. Sami Miettinen | #paasipodi 73”
Hosted by: Martin Paasi
Both men are on the bill at the Tampere investment fair the following week — Paasi giving a keynote, Miettinen on the panel about AI’s effect on investing — and both admit on air that they have not prepared. So they rehearse live.
Paasi’s keynote. He has, he says, either not prepared at all or prepared for over ten years, depending on how you count: the talk is an evergreen he has given more times than anything else he has, and this outing is its honourable farewell, refreshed with new figures because he noticed the worked example was a couple of notches off. Miettinen’s objection is that an hour of “buy low-cost index funds and don’t sell them” gets repetitive, and pushes him towards the sharper version — the €5,000 newborn. Put five thousand euros into a fund at birth, leave it alone, and it buys that child an extra average Finnish pension.
The arithmetic underneath is the part worth repeating. On a median Finnish salary of about €3,200 a month, total employment cost is around €4,000, of which roughly a thousand euros a month goes into the pension system — twelve thousand a year, a hundred and twenty thousand a decade, for your entire working life. Finland’s pension system is perfectly decent, Miettinen says, but it is extremely expensive relative to what a single €5,000 investment at birth would achieve. Paasi’s slogan for it, workshopped on air: from grandpa to baby, not baby to grandpa. He notes that a Finnish family organisation has floated a comparable baby-fund scheme, and suggests it would work better in low-cost index funds.
The layer model: how to actually use AI for investing
The core of the episode is Miettinen’s architecture, and it is the clearest statement of it he has given on the show. Four layers, and most people only ever touch one.
The foundation model — Opus, GPT, Qwen, Gemma — is the hammer. Miettinen’s insistence is that these are mean-reverting models: fed an enormous volume of human-produced text, they settle on the middle of the village. That is not a defect, it is the point, and it is exactly why they are so effective at coding, at legal interpretation, at slides and at images — all bounded domains where the human consensus is the right answer. What it also means is that they are not good at predicting share prices, which the models themselves told Paasi when he interrogated them, and hopeless at anything genuinely novel.
The harness is the provider’s control layer sitting on top of the base model whenever you reach it over the internet — Anthropic’s, OpenAI’s, Google’s. It is invisible and it shapes every answer. Run a model locally and the harness is gone: the model is alone in the box with you, unable to ask an American or Chinese owner how it ought to answer. Miettinen finds this genuinely liberating rather than sinister, because it puts the control layer under your own hand.
Paasi’s contribution here is his own experiment. He spent a night bombarding the four Hello Humans models about what they know, what they do not, where the fog begins and whether they remember being given their guardrails — the answer to the last being no. He also runs Google’s best private model, Gemma 4 26B, on a 64GB laptop with no internet connection at all, and notes with some delight that Sami Miettinen turns out to be baked into its weights: a certain kind of immortality. Miettinen’s answer on the models’ centrism is that the Chinese model is not reverting to the human mean but to a different one, which is why four models beat one — two American, one European, one Chinese.
Memory and curated context is where Miettinen says the value actually sits, and where almost nobody bothers. Asking a base model “how will Nokia do” is an absurd question, because what it holds is a vague human consensus from six months ago. The work is bringing current facts in, then a hypothesis — the share has moved not because of 9G networks but because of optical data transmission selling hard into data centres — then curating that as a data point in your own store, then expanding it with searches from trusted sources, then extending the thesis outward: do the American hyperscalers use Nokia components in their data centre buildouts? Layer that over agents like Samantha and Stöbä that retain it across long horizons and you get something that is not remotely the same thing as a one-shot question to a chat box.
His planned five-minute demo at Tampere follows that structure: his AI workers debating a listed stock the audience cares about — why Wärtsilä has run so hard, with diesel generators as backup power for AI complexes as the partial answer, extended to what that implies for Alphabet, whose cloud demand exploded and whose search business conspicuously failed to die. Then Claude in Excel analysing his own portfolio and arguing the case for selling each holding. Then simulations of his limited partnership commitments — he is unusually a direct private investor in American private equity funds, with the ten-plus-two year capital call and distribution cycle to model. His aside on that is a good one: if you want to listen to a finance professional, make sure they have stood on every corner — client, seller, service provider and investor — because only then do they see how the thing actually works.
Miettinen’s larger irritation is with the columnists. People who do not pay for the models, prompt them badly, and then generalise their own incompetence into a property of the system — who are, he suspects, equally bad at giving instructions to humans. The counterpart of his delete, delegate, do framework is that AI collapses all three into one interactive structure: you delete the bad ideas intelligently, you delegate to agents, and you still do — and the people who operate this way see their productivity compound while everyone else’s pyramid falls over. Finland’s active wage bill is over a hundred billion euros, and his blunt framing is that this is the market you are competing in.
Small caps, big caps, and the Finnish market
The exchange rose around 35–40 per cent last year, and Miettinen’s own equity savings account — which forces him into direct holdings, since index funds are not permitted in the wrapper — did well on Alphabet, Shopify, Nokia, Neste, Wärtsilä and Nordea. He is candid about the scale: the thousand-euro dividend punts that doubled are pleasant but immaterial, and he declines the stock influencer’s trick of showcasing whichever of thirty tickers happens to be up. One of them, a healthcare holding, is down.
The genuine puzzle is that small caps did not follow, and small software companies in particular went backwards — so investors who backed small Finnish firms out of conviction, active ownership or patriotism are looking at minus five while everyone else has plus forty. Miettinen’s frame is Fama and French: small companies cheap on price-to-book have returned close to 10 per cent real over 88 years. That has failed recently, and his reading is that Fama and French were locally right but not globally right — strip the Magnificent Seven out of the US market and it went nowhere either. The usual pattern is that the large caps run first and money rotates down; he is not certain it will this time, and notes drily that Aalto teaches every anomaly to be permanent.
The data centre argument
The episode’s strongest passage is Miettinen against the sceptics — Risto Siilasmaa among them — who argue that ten billion euros of data centre investment is the wrong kind of investment and will consume Finland’s electricity.
His answer has three parts. First, the alternative is not a better investment; the alternative is aggressively waiting by the telephone while Finland sinks. The AI megatrend is here, it requires compute, and Finland could be a compute superpower. Second, the electricity objection inverts: yes, five gigawatt-scale data centres consume power and prices may rise for a period — but demand is what makes new generation worth building. Finland has a roadmap to double clean electricity production by 2035, and that roadmap exists only if there are buyers. This, he says, is the ABC of economic growth: when demand arrives, you can sell. Third, the ancillary benefits are real and stackable — land sales, construction, connections, a couple of hundred jobs in a region that needs them, tax revenue — and the arrival of the operators can be conditioned: on funding professorships at Aalto, on locating design work in Finland, on supporting a country that already holds a remarkable share of the world’s quantum computing capacity. Finland is at the outermost tip of the spear on several of these things and is declining the investment that would let it build on them.
His illustration of the reasoning he is arguing against comes from the Uniper case, where he has been sitting through the parliamentary inquiry: some parties hold that the taxpayer lost nothing on a six-billion-euro disposal, because Fortum is now profitable. Six billion below the line is six billion of production capacity that would otherwise have existed. If it truly costs nothing, he offers, let a consortium take Fortum’s generation capacity off the taxpayer’s hands — the remaining company will still turn a profit, so nobody will have lost anything.
The political section
The episode’s most partisan stretch turns on a single statistic Miettinen attributes to the employers’ confederation: divide Finns into ten income deciles and only the top two are net contributors of tax; the other eight are net recipients. His argument from it is that the people the country most depends on are treated worst, that he can find no logic in it beyond group identity, and that the same instinct produced the option taxation rule the previous episode covered. Paasi widens it into the intellectual history — class antagonism as a constitutive part of the tradition, and the observation, offered from both sides, that the payers stop paying once they stop being able to.
Miettinen’s advice to listeners is the same one he gave in episode 71 and he repeats it as fact rather than opinion: the money has run out, transfers will be cut from here to the foreseeable future regardless of who governs, and you defend your family best by reducing your dependence on cash flows the state decides — which argues for supporting lower taxes funded by a substantially smaller state.
Paasi takes the last word for the AI case instead: Finland has a thousand-billion-euro opportunity in front of it, the money has arrived to build, and the thing simply has to be done. Growth works like stacking Lego — one thing creates the next, which grows the first.
From the same session: Neuvottelija's own episode
Topics: Investing & Markets · AI & Enterprise Tech · Finnish Economy & Policy · ← All media visits · Talks & Events · Markdown mirror: index.md