Also-ran Sweden bids for AI world leadership

‘Freakin’ ambitious’

Sweden’s cluster initiative was more ambitious than most because it aspired not to mere excellence, but to global leadership, said Norwegian cluster consultant Christian Rangen, while hosting a workshop with cluster bidders for Swedish innovation agency Vinnova in April. More so, even, because it had dual ambition for each cluster to be world-leading in both research and innovation, and then to bring scientific research to market “at scale”.

He told cluster bidders: “Top-five in the world by 2035. That is really freakin’ ambitious! You can’t do this with a small ecosystem mindset. A lot of the clusters that we have in Europe, they struggle because they’re too local.”

Their ambition was admirable and preparations second-to-none, but Swedish planners had made a crucial oversight in their plans for world leadership in fundamental AI, said Mats Nordlund, research director at AI Sweden, who is helping mount a bid to build a cluster for secure and trustworthy AI. AI breakthroughs are driven typically by industry, not academia, he said.

Yet only universities could mount bids to form clusters, added Nordlund, who helped establish a similar cluster programme in Moscow before Russia invaded Crimea.

Other fields of science were “technology-push”, where breakthroughs in academia spurred industry innovation, he said. But AI was technology-pull, and it was evolving so fast that it made 10-year research plans meaningless.

“AI today is not driven by universities,” said Nordlund. “Universities have a really hard time to keep up with what goes on in industry. It’s driven today by the big tech and automotive companies. Industry gets stuck. Then it becomes a research problem.”

World-class researchers

Sweden plans for each cluster to build a team of world-class researchers around a “superstar” who will sit at the centre of such a concentration of companies, entrepreneurs, industry and financiers that it creates a critical mass that sucks in more talent, capital and business.

As it prepared to issue its call for cluster bids in March, French superstar AI research scientist Yann Le Cun launched a robotics AI startup with $1bn seed funding from a long list of notable firms and computing grandees, from the US, France and Asia.

Considered one of the founders of the AI boom by his influential academic work and career-long ties to the US industrial ecosystem, he counted AI chip firm Nvidia, and computing superstars Eric Schmidt and Tim Berners Lee, among his backers. Others included Toyota and the French state.

Swedish startup Superintelligence Computing Systems (Sicsai) was meanwhile recruiting a group of global investors to raise €10m for robotics AI, and like Le Cun, claiming breakthroughs in fundamental AI. Half the money is promised under an EU scheme intended to embolden native private investors to invest in high-risk, high-tech startups they are otherwise shunning.

Sicsai didn’t attempt to raise $1bn because it didn’t need that sort of money, said Karim Nouira, its CEO, who is in talks to participate in a bid to form a Swedish AI robotics cluster. It is already “way ahead” of Le Cun and the rest of the world in physical AI, he said. Yet Swedish AI firms could still use help from clusters to raise capital and build invaluable connections with industry.

“Europe is lagging behind because there’s almost no interest for significant deep tech investments,” said Nouira. “There are some smaller VCs doing small tickets, but it doesn’t take you anywhere. It just dilutes your company.

“It’s really difficult to get pilot projects going,” he said. “The large companies say that they support startups, but when you want to do a pilot project, they have very restricted budgets and very little wiggle room. The management on the top, for them it’s very easy to say, ‘We are doing all kinds of corporate venturing’, but they don’t really allocate resources for it. It’s a huge risk for them in the actual company to take on pilots.”

Dominant US paradigm

Sweden pledged in its landmark AI Strategy in February an ambition to rank among the world’s top 10 countries in applying AI. US dominance of LLM AIs that Sweden would use to that end is so great that it is hard for other countries to be competitors rather than users, according to market research non-profit EpochAI, which charts the race for AI among US computing giants, and the insignificant comparative place of “compute-poor” also-rans in the rest of the world.

That dominance gives US AI firms an advantage in R&D. OpenAI spent €16bn on compute last year, half for R&D. The US big-tech firms that provide that computing power spent $700bn building it last year, as much as Sweden’s entire economy.

One of them, Microsoft, is due to finish building €3bn of AI compute in Sweden in June, comprising 20,000 of the graphics processing units (GPUs) needed to run the LLMs that Sweden will use to power its bid for world status in applied AI. Sweden has however declared an ambition for its clusters programme to help make it, and Europe, independent of foreign AI. Microsoft’s inauguration will coincide with that of Sweden’s own public Arrhenius supercomputer, a €68m, EU-funded AI HPC with 1,500 GPUs for scientific research.

Sweden has meanwhile begun building another €30m ($35m) AI computing centre, with joint funding under the EU AI Factory scheme, to give native small and medium-sized enterprises free access to expensive resources. Called Mimer, it will be powered by 400 GPUs, a tenth the size of other EU centres.

But Sweden’s priority was not, like most AI computing centres, to be biggest, said Bruno Lecointe, who as head of HPC at French state computing firm Bull is responsible for building it. It allocated half its budget for software and expertise that will help small firms in key Swedish industrial sectors re-engineer their algorithms and run AI simulations. It is a stepping stone, to prepare them to operate on Europe’s larger AI Factories in Finland, Germany, France and Italy.

Swedish strengths

Mimer will serve those fields of science and technology where Sweden has strengths and plans to build innovation clusters. It might rank insignificantly in AI, but it does rank second worldwide for innovation. Sweden has more researchers per head of population than any country, and is among the largest in education spending.

R&D spending as a proportion of its economy is fifth in the world. Its strength in bioscience, chemistry and materials is complemented by its hosting two of the world’s most advanced sub-atomic experiments, comparable in size and capabilities to the famous Cern.

German clusters, which Swedish planners used as an exemplar, had proved likely to succeed in fields where they could build upon existing strengths, said Bastian Krieger, an innovation researcher at ZEW Mannheim economic institute, who wrote a landmark study on them. Sweden’s plan to build 10 such clusters and make them world-leading was “ambitious, but not impossible”, he said. Its plan to construct a world-class centre in fundamental AI by top-down diktat might work.

“It is not necessarily a bad idea,” said Krieger. “It might be a really good idea, if they can attract some scientific stars.”

Clusters in France had demonstrated that the presence of top researchers and a concentration of funds tended to pull in other researchers whose work might otherwise have led in other directions. But success tended to gravitate to higher concentrations of funding as well. Sweden might better succeed if it built fewer clusters with more money, he said.

James Wilson, director of the Basque Institute of Competitiveness, and a leading cluster theorist who has worked with Swedish policymakers, said it is important to build on strengths. Policymakers can build clusters in fields where a country is weak, he said, as the Spanish Basque region had done with biosciences and finance. But only if they had existing science institutes to build on, or other clusters with related strengths, and some lead firms to support them.