Author: Henrik Torstensson

Some links to the, mostly, Swedish political/policy debate on economic growth, productivity, and financing of innovation.

There are no easy answers to higher economic growth, but I think there are many things Sweden and the Swedish public sector can do independently of the timing of EU initiatives like the capital markets union.

One thing I believe is that long-term Swedish capital, like pension funds, should invest more in Swedish startups to drive significantly higher GDP growth (with the same or higher direct financial returns to reteriees as today plus the value from higher GDP growth in other investment areas like bonds).

Dagens Industri: Innovation och konkurrenskraft ger ett starkare Europa by Sweden’s Prime Minister Ulf Kristersson and Finland’s Prime Minister Petteri Orpo.

Dagens Industri: Replik: Utse en Teknikminister by Professor Sylvia Schwaag Serger, CEO IVA.

Dagens Industri: Nya superfonden som sätter fart på Sveriges tillväxt by Annika Winsth, Carl Bennet and Björn Rosengren.

DI Digital: Techsektorn står för 8 procent av BNP: “Bärande roll”

McKinsey: The paradoxes of Sweden’s success and struggles – and the path forward

European Commission: The Draghi report on EU competitiveness

A16Z Speedrun: Make ARR Useful Again. There’s a lot of aggressiveness in calculating ARR numbers in the current market.

A16Z writes about how to approach calculating MRR (monthly recurring revenue) and ARR (annual recurring revenue), the most common ARR sins and difference between annual recurring revenue and annual run rate.

On building up ARR from MRR

“MRR, by contrast, is a management metric. This is a point-in-time monthly value of recurring revenue you expect to keep happening. It’s tempting to say “ARR = 12 × MRR” and stop there. But that identity is only useful if “recurring” really means recurring—meaning you exclude one-offs (implementation, training, hardware) and you’re disciplined about usage. The clean approach is simple to state and powerful in practice. You should treat subscription and contracted minimums as MRR. Truly variable usage is its own line item and only when it proves stable or becomes contractually committed you can graduate it into MRR and ARR.”

MRR and ARR specifically for AI companies

“AI inference takes every MRR bad habit and amplifies them. Demand is often bursty (launch weeks, seasonal spikes), units are tiny (per 1,000 tokens, per request), and capacity can be either provisioned or opportunistic. That means you need an MRR policy that respects how the infrastructure is priced.

A founder-practical way to handle this without twelve sub-metrics is to:

Keep MRR for the part of your AI product that is contracted with items like fixed subscription fees, platform access, support tiers, and committed throughput minimums. If a customer signs for a model unit per month or a token floor, that is MRR.

Track Measured Usage separately for everything else: pay-as-you-go tokens, burst capacity, overages. Report it every month with a trailing average. If usage stabilizes for, say, six months, you can promote it into MRR with a straight face.

An older, from 2022, article Choose Good Quests. I found it while reading Build What’s Fundable. Kind of a call to arms to solve important and difficult problems.

I believe there’s a lot of value in driving productivity and economic growth in itself. As an example, a lot of good things could be done today if Sweden and Europe had had 0.5 % higher of annual GDP/capita growth in the last 15 years.

Driving productivity can often be done by replacing current solutions with better and/or cheaper ones in a, relatively, lower risk way for entrepreneurs.

Doing something important that requires fundamental advances in technology (and product) is significantly harder and riskier. Choosing a good quest, as I understand it, is to do that despite it being difficult. Especially for the few people that can gather the resources to do so.

About running Microsoft, AI and more.

Lovable has announced that they have passed $200 million in ARR (up from $100 million in June), 5 million daily visits to Lovable projects and 100,000 new projects per day.

Even as AI is accelerating startup growth (in the extreme cases) I believe that startups that become popular at scale in five years will still be seen as overnight successes. And building large companies will still take about 10 years and there will be a lot to do in year 11 and onwards.

Kyle Harrison has written Build What’s Fundable. It’s a good, and long, read on the development of venture capital and startup funding over the last 20 years. An article for students of startup and venture capital on how startup funding became more professional, but that brought second-order effects of startups being ‘manufactured’ to access follow-on funding.

I don’t think it is wrong to adapt strategy to make fundraising easier (it is really hard for 99 % of startups, so all advantages are worth seeking), but agree with Kyle’s point that one should seek to ‘find beliefs worth living for’. Not only making sure the next fundraising round is a success.

I think I’ll have more thoughts related to the article once I’ve read it again and have let it ponder for a while.

Accel has published its Globalscape report for 2025 titled Race for Compute (PDF to download).

Many interesting data points across the report, but I found these two charts on just how significant the funding of a few companies (OpenAI, Anthropic and xAI) is to overall VC investments and how 25x more capital is invested in US-based AI models than in Europe/Israel.

Einride, the Swedish electric and autonomous freight company, has announced that it will merge with a U.S.-listed SPAC (Special Purpose Acquisition Vehicle). Einride currently has $45 million in annual recurring revenue and contracted annual recurring revenue of $65 million from more than 25 customers and 200 vehicles in operation.

The valuation in the merger is targeted to be $1.8 billion and the new company is targeted to have $219 million in cash (before potential redemptions and transaction expenses) and is looking to raise another $100 million.

It is good that the stock market provides actual risk capital to companies and not only buy shares in already profitable companies and wait for dividend and share buybacks. When making such early-stage investments a public market investor should operate like a venture investor: build a diversified portfolio and don’t invest too much into individual companies.

Yesterday I was at Nordic Founder’s Pitch or Die Trying event at Epicenter in Stockholm.

The highlight was a fireside chat with Charles Maddock, founder of Strawberry, focused on the company’s pre-seed and seed fundraising.

Following Charles was a panel of investors who, among other things, were asked about red flags that would disqualify a company from an investment.

That is a question I find interesting.

I don’t think startup investing is so easy, or companies and founding stories so perfect, that absolute red flags should be used very often or broadly.

The startup investor job is to try and understand each startup and its unique opportunity to become an important, large company.

With the exception of fraud or dishonesty, pretty much all red flag situations (like ownership by founders at Series A, team’s technical or business ability, and previous proof points of founder being great) can be dealt with. The question is if there are signs of exceptional strength in at least one area that make it worthwhile spending the time on it.

By the way, having as a red flag that the founders are married has a pretty poor track record as there are a bunch of outlier outcomes where founders were/are married.