Sweden’s AI Strategy lacks the Pieces That Would Actually Make it Work
- Armin Catovic
- 6 days ago
- 3 min read
Sweden’s current wave of AI startups did not emerge from a well executed national AI strategy.
Successful companies like Lovable, Tandem Health, Neko Health, Legora and many others were not the result of coordinated state programs, sovereign compute investments, or targeted venture initiatives. They grew out of talent, bruteforce will, strong founders, local talent density, and-crucially-the Stockholm AI meetup ecosystem that formed in the pre-LLM era. Long before generative AI became mainstream, that community created a shared technical culture, early experimentation, and founder networks. In other words: the ecosystem produced companies despite the absence of a strong national AI stack strategy, not because of one.
Nearly two years after Timothy Liljebrunn’s and Armin Catovic’s critique of the then-AI strategy, and their emphasis on the strategic AI Stack, Sweden has now released a revised national AI strategy document. While more comprehensive, it still falls too flat, much to the disappointment of the grassroots Stockholm AI community.
This matters when reading Sweden’s newly proposed AI strategy. It is a serious and thoughtful document. But when mapped against a full-stack view of AI competitiveness, it becomes clear where it is strong, where it is incomplete, and where it risks optimizing for adoption without fully addressing all the critical pieces that make the AI strategy actually work.
There are some glaringly obvious absences. The AI strategy acknowledges upstream dependencies on chips, rare earths and supply chains, framing them as national vulnerabilities. But acknowledgment is not a hardware strategy. There is little in the way of: (1) semiconductor participation strategy, (2) long-term compute supply agreements, (3) hardware R&D programs and positioning with the European chip ecosystem (read: ASML).
The strategy also seems to revert to the tried and failed “svensk språkmodeller” of the yesteryear. It seems the nation hasn’t learned anything from GPT-SW3 and is willing to keep pouring more money in this perilous direction. In today’s landscape, training sovereign models from scratch is rarely optimal - quadruply so for “low resource languages”. A more pragmatic path - clearly demonstrated by China - is to leverage strong open foundation models and systematically fine-tune, adapt, and improve them for national language and sectoral needs. This approach moves faster, costs less, and keeps domestic capability aligned with the global frontier.
While the AI strategy highlights skills development, particularly across public sectors and research environments, it doesn’t tackle the elephant in the room - global talent attraction is thin. There is no major program to recruit top AI engineers and researchers internationally. Similarly, while access to capital is acknowledged, there is no mention of national AI venture funds, sovereign co-investment vehicles, startup procurement pipeline, nor scaleup strategy. SMEs are encouraged to adopt AI, but startup scaling is not a central industrial objective. This is striking given that Sweden’s most promising AI startups emerged without such support. The ecosystem has proven it can produce founders and companies. The missing piece is a national framework that helps them scale and stay.
Sweden already has proof that strong AI companies can emerge locally. Lovable, Tandem Health, Legora, Neko Health and others grew out of talent density, founder ambition, and community - particularly the Stockholm AI meetup ecosystem that formed before generative AI hype cycles began.
The new strategy has the opportunity to do something different: not just enable AI use, but build a coherent national AI stack. If Sweden can connect serious and long-term hardware/compute commitments, with model capability that actually makes sense, through software ecosystems and startup scaling, and surround it with talent attraction, then the next generation of companies will not arise despite the strategy, but because of it.
The proposed strategy lays important groundwork. The next step is to turn it into a true stack strategy - one that spans from silicon to startups. Authors:
Comments