AI investments

AI Sweden is helping develop OpenEuroLLM, a multilingual, open source GenAI. The EU needs native capacity to run such things because the global race for AI supremacy is driven by geopolitics, it said.

As the consultation concluded, Sweden’s powerful Wallenberg family of industrialists said it is building what will become the country’s largest AI factory, for the sake of national sovereignty and prowess. Four of Sweden’s largest corporations, all of which it owns, will build it: drugs giant AstraZeneca, arms manufacturer Saab, mobile telco Ericsson and SEB bank.

Jenson Huang, CEO of Nvidia, whose market-leading AI chips will power the Wallenberg AI factory, flew in to announce the project – and collect an honorary doctorate from Linköping, the university running and building EU-funded supercomputers around US chip technology.

“No company should outsource their intelligence. No country should outsource their intelligence. Intelligence comes from your life’s work. That means data. The data of companies belongs to the companies. The data of Sweden belongs to its people. It’s a sovereign right. Your data is like land,” Huang said in his acceptance “fireside chat” with family head Marcus Wallenberg.

The Swedish financier and its corporations all refused to answer questions about the size of their investment, the compute and whether small and medium-sized enterprises (SMEs) might get to use it.

A week later, Canadian investment fund Brookfield said it was building the country’s largest AI factory in support of Swedish sovereignty and the national AI strategy: a €10bn, 750MW datacentre of yet undetermined purpose. It is believed to be negotiating with US hyperscalers and other potential tenants.

In February, it declared €20bn for French AI, gunning to build Europe’s largest AI factory. Nvidia said at its financial year end in April that 100 AI factories were being built on its specialist AI chips around the world.

Overshadowed, the EU revealed a €30m budget for Sweden’s Mimer supercomputer, an AI factory intended with 13 others around the continent to build GenAI with which Europe could compete with the US corporations that invented the technology and lead the industry. It said it would prefer bidders to use open RISC-V open source chip architecture, a global initiative born of Silicon Valley from which the EU, China and India have built sovereign technology for general-purpose high-performance computing (HPC) processors.

Last December, when Sweden went 50/50 with the EU on its first publicly funded EuroHPC supercomputer – the SEK500 (£38m) Arrhenius – it was home to none of the 10 such computers built since 2021.

The most powerful – world-ranking Lumi in neighbouring Finland – cost €415m (£350m) to fit out with US supercomputers and chips in a datacentre run by US military cloud corporation CSC. These HPCs, intended for intense, general-purpose scientific computing, differ from AI factories that house chips designed only to develop GenAI alongside cloud computers intended to serve access to them.

Something different

If Europe wants to win the AI race, it must invest in disruptive projects and “stop copying the US”, said Karim Nouira, CEO of Superintelligence Computing Systems (SICS), a Swedish robot brain pioneer. In December, SICS became one of 71 startups to receive a share of €387m in grants and equity finance from the European Innovation Council, in a scheme designed to persuade private financiers to match its investment in high-risk “deeptech” ventures they had eschewed.

“Right now, Europe is just providing more of the same. More large language models, more of the type of AI that’s prevailing already,” Nouira told Computer Weekly. 

SICS considered moving to Silicon Valley for want of investors in Sweden, despite Swedish ABB being one of the world’s largest robotics manufacturers, and against its wishes to be close to Bavaria in Germany, the global centre of industrial robotics, with an ecosystem better than California’s, said Nouira. But after a gruelling competition to prove its worth, an EIC grant and EU equity funding to half the value of €10m, private financiers began to show interest in putting up the rest.

“Sweden has good help for startups. Sweden’s innovation agency gave us half a million euros in 2023. That kickstarted our company. It put us over the threshold for landing our first customers. [But] there’s a significant gap when it comes to deeptech investment. VCs in Europe are afraid of deeptech. They are more risk averse than in Silicon Valley,” added Nouira.

“Venture support for AI in Sweden is mostly for wrappers – companies that build vertical solutions on LLMs,” Jim Dowling, CEO of Hopsworks, a Swedish AI data and app building platform that got most of its €12m investment to date from investors outside Sweden, told Computer Weekly.

Yet its second-largest investor is Swedish VC fund Industrifonden, and Stockholm has a thriving startup scene, with more unicorns per capita than any other city in the world after the San Francisco Bay Area that houses Silicon Valley. Their success shows Swedish investors are willing, given something to invest in, said Dowling. 

Until recently, Sweden had a dearth of the talent that AI startups needed, he said. The Wallenberg AI, Autonomous Systems and Software Program began fixing that in 2017, funding university research in AI.

“WASP helped bootstrap non-existent talent. It established AI researchers in universities. It brought people interested in AI to Sweden,” said Dowling. “Since I taught the first deep learning course in Sweden [in 2016], there has been huge change. Sweden was way behind France and the UK in academia – and even further behind Canada and the US.”

Wallenberg’s investment in AI compute is “noblesse oblige”, he said. It is using the power over Swedish industrial giants it owns to make them invest in AI compute for the sake of the nation.

“Lots of old European money didn’t have the same foresight,” he said.

Marcus Wallenberg said at Huang’s honorary that Sweden (without fossil fuels) is entirely dependent on innovation for its exports. But it has green energy in abundance that it can turn into computing power. It needs sovereign compute because, he implied, data will become an export commodity.