AI-skiftet brings together essays and daily news from an industry-near perspective: what the technology is already doing, which assumptions hold, which don't, and how Sweden can prepare more proportionally.
Read the main essayDaily coverage of AI development — models, research, physical AI, hardware, quantum computers, labour market, regulation and debate.
The Norwegian robotics company 1X has begun delivering its humanoid home robot NEO to the first US homes during 2026.
NEO costs $20,000 to buy, or $499 a month, and is marketed as an assistant for household chores. In practice, though, the robot cannot yet handle the tasks on its own: a remote operator steers it using VR goggles, looking straight into the home – raising clear privacy concerns. In March 2026 1X unveiled a so-called world model meant to let NEO learn by watching people – the company’s path toward real autonomy.
For the Nordics, 1X is one of the region’s heaviest robotics companies. It was founded in 2014 in Moss as Halodi Robotics by Bernt Øivind Børnich, still manufactures in Moss, and raised around one billion dollars in 2025 at a valuation close to ten billion. When camera-equipped robots remotely controlled by humans reach Nordic homes, the question quickly lands on the desks of Datatilsynet and IMY – even as ABB, Denmark’s Universal Robots and NTNU labs push Nordic robotics forward.
Microsoft and Quantinuum have had their quantum error-correction results published in the scientific journal Nature on June 10.
The peer-reviewed study confirms an improvement of up to 800 times lower logical error rate for the logical qubits compared with the physical ones. Using a qubit-virtualization system on Quantinuum’s trapped-ion hardware, four highly reliable logical qubits were created from just 30 physical ones, and in one test case the error rate fell from about 0.8 percent to 0.001 percent. The result was first demonstrated in 2024 but has now undergone independent review.
For the Nordics, news of more robust qubits is central. Through the Wallenberg center WACQT, led from Chalmers, Sweden is building a superconducting quantum computer set to grow from 25 to 100 qubits by 2029, and error correction is the decisive barrier to reaching practical use. In parallel, NordicQCI – with RISE, Ericsson and KTH – is building Nordic quantum communication, and on June 22–24 the industry gathers at the IQT Nordics conference in Oslo.
The research institute SINTEF and the trade union federation Akademikerne on June 12 launched a report on artificial intelligence in knowledge work and its interplay with the Norwegian model.
The report shows that about half of highly educated workers already use AI on the job, but that only 26 percent feel they receive sufficient training, and only one in three feel that management invites them to take part in how the technology is introduced. It lands on eleven recommendations and examines how the core principles of the Norwegian model – autonomy, co-determination and trust – are challenged when AI enters the workplace.
For the Nordics, the training gap is a shared challenge. The same pattern appears in Swedish working life, where unions such as Unionen and TCO and agencies such as Vinnova are pushing the skills agenda, and the report’s point – that the social-partner model with co-determination can become a competitive advantage in the AI transition rather than a brake – applies to both Norway and Sweden.
OpenAI confirmed on June 8 that it has submitted a confidential draft registration statement, a so-called S-1, to the US Securities and Exchange Commission – just days after rival Anthropic did the same.
That puts two of the leading AI labs in the IPO queue at the same time. OpenAI is currently valued at around $852 billion, and some analysts expect a valuation above one trillion dollars at a listing, even though internal documents reportedly point to a loss of around $14 billion this year and profitability only in 2029. Goldman Sachs, Morgan Stanley and JPMorgan are leading the work, and a possible listing window is put at the autumn of 2026.
For the Nordics, the exposure becomes tangible. Norway’s Oil Fund, the world’s largest sovereign wealth fund, and Sweden’s AP pension funds are already heavily invested in exactly these tech companies, and the Oil Fund’s chief recently warned the Storting about record-high concentration. An OpenAI listing would give Nordic pension savers a direct stake in a still-unprofitable company – and make the Nordic savings krona even more dependent on the AI promise being fulfilled.
Just three days after launch, the US Commerce Department ordered AI company Anthropic to immediately disable its flagship models Fable 5 and Mythos 5 for all foreign nationals – wherever in the world they are located.
The directive arrived on June 12, and as of June 15 the models remain offline. The agency cites national security and points to a method of bypassing the safeguards – a so-called jailbreak – that could unlock the models’ cyber capabilities. Anthropic says it disagrees and considers the bypass narrow, but is complying with the order, which also locks out the company’s own foreign employees. A tiered solution is reportedly being negotiated, giving US citizens full access while foreign nationals are blocked or routed to the safer Opus 4.8 model.
For the Nordics, the dependency risk becomes concrete. Nordic companies and developers that had built Fable 5 – launched on June 9 – into their workflows lost access overnight. It strengthens the case for digital sovereignty raised by both Norway’s Oil Fund and Nordic regulators: homegrown Swedish language models, the AI factory in Narvik and the new Nordic data centers look like cheap insurance when a foreign government can switch off a cloud model with the stroke of a pen.
Researchers at Princeton (Princeton Language and Intelligence) have unveiled Goedel-Architect, an AI agent framework that automatically proves mathematical theorems in the Lean 4 proof language.
On the demanding PutnamBench test – 672 problems from the famous Putnam competition – the system solved 75.6 percent for a total of $294 in compute cost. The competing Hilbert system, powered by Google’s Gemini, reached a lower 70 percent but cost around $170,000 – roughly 500 times more. The trick is a compiler-checked blueprint: a dependency graph of definitions and lemmas built up in advance, while the open DeepSeek-V4-Flash model drives the proofs themselves.
For the Nordics, the message is encouraging: smart agent architecture can beat raw compute. Nordic labs with limited GPU budgets – WASP, the Berzelius supercomputer in Linköping, Norway’s Sigma2 and formal-methods groups at KTH, Chalmers and NTNU – can compete on algorithms rather than chip counts, exactly when compute is at its tightest.
The shortage of AI compute is now so acute that competitors are renting each other’s data centers. SpaceX – which now owns Elon Musk’s AI company xAI – is renting out its entire Colossus 1 supercomputer in Memphis to rival Anthropic.
According to Bloomberg, Anthropic is paying around $1.25 billion a month through 2029 for access to 220,000 Nvidia GPUs and more than 300 megawatts of capacity to meet demand for Claude. SpaceX leased out the facility after wrestling with latency and hardware problems while trying to link three data centers together for its Grok chatbot. Meanwhile, Google is set to pay SpaceX roughly $920 million a month for compute.
For the Nordics, it is precisely this shortage that is driving the gigafactories built on cheap hydropower. In Narvik, Nscale and Aker are building an AI factory for 100,000 GPUs – with Microsoft as the engine after OpenAI stepped back – and Nebius is raising a 310-megawatt data center in Lappeenranta, Finland. For Swedish and Norwegian firms, the global capacity crunch means more expensive and harder-to-schedule access to the chips they need, even as the push to keep sensitive data at home grows.
Sweden now has national support to help school operators lead and govern AI in education. The free scenario tool was launched on June 1 by the Swedish Association of Local Authorities and Regions (SKR) and the research institute Ifous.
The backdrop is that there are currently no national guidelines for how AI should be governed in schools, even as the EU AI Act is phased in. The web-based tool is built around three learning tracks – governance and leadership; law and ethics; and technology and procurement – and was developed by 18 national actors, including AI Sweden, the Swedish Internet Foundation, RISE, the National Agency for Education, Linköping University and the University of Helsinki.
For the Nordics, the initiative makes Sweden a frontrunner on an issue just as pressing in Norway and Denmark, where schools face the same AI pressure without clear national rules. The involvement of both the University of Helsinki and the Swedish Research Council signals a Nordic ambition to coordinate how schools govern AI – and hands Norway’s Directorate for Education and the Storting a ready-made model to lean on.
IT services giant DXC Technology and AI company Anthropic announced a multi-year global partnership on June 11 to bring the Claude model into the mission-critical systems that banks, airlines, insurers, manufacturers and government agencies rely on.
DXC becomes one of the few "Global Premier" partners in the Claude Partner Network and will train tens of thousands of Claude-certified engineers embedded directly inside customers' organisations. The company has already run Claude across its 115 000 employees in 70 countries and let the model write more than 95 percent of the code for its new operations platform, DXC OASIS, now used by over 50 customers. The deal is described as a response to companies wanting to move AI quickly from pilots into production – under strict security and compliance requirements.
For the Nordics, the partnership lands squarely in the region's most regulated industries. Banks such as SEB, DNB and Nordea, airlines such as SAS and Norwegian, and engineering giants such as Volvo and ABB face the same question: how to bring AI into systems where every transaction is governed by strict rules. DXC runs large operations across the Nordics, and for unions such as Unionen and Norway's Tekna, the promise of "tens of thousands of certified engineers" raises the question of how consulting and developer roles change when the model writes most of the code itself.
On June 12 Anthropic published the results of its first "Public Record" survey, in which nearly 52 000 Americans were asked what they hope for and fear from AI.
The most common fear was job loss – 64 percent – followed by cognitive dependency (56 percent) and disinformation (52 percent). Anxiety was highest among the well-educated, whose work most resembles what AI is now starting to do. At the same time, nearly half (48 percent) hoped the technology would cure diseases such as cancer and Alzheimer's. Only 15 percent trust the AI companies to oversee themselves, and more than 70 percent – across party lines – want the government to regulate AI.
For the Nordics, the figures mirror a growing regional ambivalence. The Swedish Internet Foundation's annual "Svenskarna och internet" and similar Norwegian surveys show the same mix of curiosity and worry, and the demand for independent oversight puts pressure on Sweden's IMY and Norway's Datatilsynet just as Norway's AI Act is being prepared. For politicians in the Riksdag and the Storting, the message is clear: voters want regulation, not self-regulation – a signal ahead of how the EU AI Act is interpreted nationally.
Sweden's most powerful sovereign AI supercomputer, Sferical AI in Linköping, is being built at a time when the world's fastest AI memory is already sold out – the entire 2026 production run is pre-booked by the big cloud providers.
The backdrop is that Nvidia's next-generation platform, Vera Rubin, entered full production in June and begins shipping to the cloud giants this summer. The platform uses HBM4, the next generation of high-bandwidth memory, for the first time, with South Korea's SK hynix expected to take around 70 percent of supply. Both SK hynix and Micron say their entire HBM4 capacity for the year is already spoken for by hyperscalers such as Microsoft, Google and Meta – and with demand for AI compute more than tripling each year, even TSMC's capacity will not meet demand until 2027 at the earliest.
For the Nordics, this is the bottleneck behind the whole region's AI ambitions. Sferical AI – owned by the Wallenberg sphere together with Ericsson, Saab, SEB and AstraZeneca and built on Nvidia's GB300 systems – and data-centre operators such as Icelandic-Swedish atNorth and EcoDataCenter in Falun compete for the same tight supply chain as the American giants. When memory is sold out and prices rise, Nordic investments in their own sovereign AI capacity become both more expensive and harder to schedule – even as the need to keep sensitive Swedish and Norwegian data at home only grows.
For the first time, Anthropic has passed OpenAI as the most widely used AI provider among US businesses. That is according to the Ramp AI Index, which tracks card spending and subscriptions across more than 50 000 American companies.
According to the index, Anthropic accounted for around 34 percent of companies paying for AI versus OpenAI's 32 percent at the turning point this spring – a sharp rise from just a fraction as recently as 2023. Among first-time buyers, Anthropic won around 70 percent of head-to-head comparisons. At the same time, analysts warn the lead is fragile: rising costs, compute constraints and the token-based pricing model could quickly erase it, and a more cautious IDC survey suggests the picture of Claude's enterprise reach varies depending on how you count.
For the Nordics, the shift turns the choice of AI provider into a strategic question. Banks, industry and the public sector in Sweden and Norway that standardise on one model – Claude, GPT or Gemini – lock themselves in for years, and DXC's recent deal to bring Claude into regulated systems reinforces the trend. For Swedish efforts to build their own language models, such as GPT-SW3, the question becomes how Nordic players weigh openness and digital sovereignty against the American models' rapid lead.
OpenAI said this week that it is acquiring the German developer-tools company Ona – previously known as Gitpod – to strengthen its coding assistant Codex.
Founded in Kiel in 2020 and with around two million developers behind it, Ona builds secure cloud environments where AI agents can write code, find bugs and make changes directly in a customer’s own infrastructure. The technology is meant to let Codex run tasks lasting hours or days without being interrupted when the developer’s computer shuts down. The price was not disclosed and the deal is subject to regulatory approval; Ona’s entire team, led by CEO Johannes Landgraf, will join OpenAI’s Codex group. The move is seen as a response to Anthropic gaining ground among enterprise customers.
For the Nordics, the acquisition sharpens the fight over developer jobs and cloud sovereignty. When Codex agents can run long assignments inside a customer’s own cloud, AI coding becomes more attractive to banks and industrial firms such as SEB, DNB and Ericsson, which are reluctant to send sensitive code to US servers. At the same time, Nordic coding-tool stars such as Stockholm’s Lovable and Strawberry face pressure as OpenAI and Anthropic buy their way to a deeper grip on the entire development chain – and unions such as Unionen and Norway’s Tekna are watching what long-running AI agents do to junior programmer roles.
The EU’s major simplification of its AI rulebook, the so-called Digital Omnibus, is now heading for a final vote in the European Parliament in June, after Parliament and the Council reached a provisional compromise in May.
The package postpones the heaviest obligations for high-risk AI: requirements for stand-alone systems under Annex III of the AI Act move from August 2026 to December 2027, while AI embedded in regulated products gets a reprieve until August 2028. At the same time the law is tightened on other points – a new prohibition is added to Article 5 against AI-generated child sexual abuse imagery and non-consensual intimate images. The transparency rules in Article 50, by contrast, largely remain on the original timeline. After the parliamentary vote, formal adoption by the Council and publication in the EU’s Official Journal are required, expected before August 2.
For the Nordics, the delay gives Swedish and Norwegian companies – which had begun to worry about meeting the requirements in time – some breathing room, but also raises new questions for those who must interpret the rules. In Sweden, the data-protection authority IMY and the Financial Supervisory Authority are proposed to share responsibility for high-risk AI, while the PTS gets lead responsibility for market surveillance, with ten agencies designated in total. Norway, which must incorporate the act via the EEA, follows the same timetable through Datatilsynet and Nkom. The new prohibition aligns with how Swedish and Norwegian police and school authorities are already grappling with AI-generated abuse images among young people.
2026 has become the year humanoid robots begin rolling off the assembly line in earnest. Boston Dynamics’ new electric Atlas has entered serial production, and rival Figure is tuning its BotQ factory to one robot per hour.
Boston Dynamics, owned by South Korea’s Hyundai, showed the production version of Atlas at the CES trade show in January and began serial production in March. All of 2026’s output is already spoken for – committed to Hyundai’s factories and to Google DeepMind – and Hyundai is aiming for 30,000 robots a year by 2028. Figure says its BotQ factory now makes one Figure 03 per hour, with a target of 12,000 a year. The robots are driven by new “world models” that let them learn tasks by watching rather than being programmed step by step.
For the Nordics, the mass-produced humanoids meet a region already strong in robotics – but on different legs. Denmark’s Odense is Europe’s hub for collaborative robots through Universal Robots and MiR, Swiss-Swedish ABB builds industrial robots, and Norway’s 1X, founded in Moss, is one of few European firms betting specifically on humanoids for the home. The question for Volvo, Scania and Nordic manufacturing is when – and whether – humanoids will pay off alongside today’s fixed-mounted robots, while Kongsberg and NTNU drive the Nordic cutting edge in autonomous ships and underwater robots.
The head of Norway’s oil fund, Nicolai Tangen, warned a hearing in the Storting in May that the AI boom has created a stock concentration the fund has never seen before – and that may become even more extreme.
The oil fund, formally the Government Pension Fund Global, is the world’s largest sovereign wealth fund at around $2 trillion and is owned by the Norwegian people. Tangen points out that the fund today has roughly $196 billion locked in a handful of US tech giants – Microsoft, Apple, Nvidia, Alphabet, Amazon, Meta, TSMC and ASML – whose value has been driven up by AI investment. The fund’s manager, NBIM, has at the finance ministry’s request begun analysing concentration risk in the equity portfolio while warning of a heightened risk of an AI-driven market correction.
For the Nordics, the warning makes the risks of an AI bubble concrete for ordinary savers: if the tech giants fall, every Norwegian’s future pension takes a hit, and the question of how much of the people’s savings should rest on a few AI companies is expected to come up when the Storting revises the state budget this autumn. The same concern touches Sweden’s AP pension funds and Norway’s KLP, and the Financial Supervisory Authority and the Riksbank are watching how AI valuations affect financial stability across the region.
AI companies’ hunt for cheap, renewable electricity is making the Nordics one of Europe’s hottest locations for new data centers – but a shortage of grid capacity is becoming the real bottleneck.
In Narvik, northern Norway, Britain’s Nscale is building a 230 MW data center in Kvandal and in May secured $790 million in financing from lenders including DNB, Nordea, SEB, ABN Amro and state-owned Eksfin. Microsoft has taken over the capacity once earmarked for OpenAI’s “Stargate Norway” and is reportedly set to fill the site with around 30,000 Nvidia GPUs. At the same time, the industry warns that high electricity prices could cause Europe to fall behind the US and China in the AI race, and Norway’s Statnett expects data centers’ power demand to potentially triple by 2030.
For the Nordics the question is now acute: the region’s biggest asset – clean hydropower, cool air and low electricity prices, with Norway covering around 98 percent of its power from renewables – only holds if the grids are expanded in step. Statnett and Sweden’s Svenska kraftnät must prioritize between data centers, industry and households, and in towns such as Narvik and Luleå the projects spark both hopes of jobs and fears that the power will flow to foreign server halls rather than local industry. It is a knot for Equinor, Vattenfall and municipal politicians to untangle.
Finnish quantum-computing firm IQM on June 9 unveiled a new family of error-correcting codes, called “barbell codes,” described as an important step toward fault-tolerant quantum computers.
According to IQM, the codes deliver up to a thousand times lower logical error rates than the standard surface code while requiring up to eight times fewer physical qubits for each protected logical qubit. They are a so-called QLDPC code tailored to the company’s own “Constellation” processor architecture, in which each qubit couples directly to twelve neighbors instead of four as in a conventional grid – which, the company says, keeps hardware complexity down and points toward systems with hundreds of logical qubits.
For the Nordics, IQM – based in Espoo outside Helsinki – is one of the region’s most prominent quantum bets, and the company is on its way to a Nasdaq listing via a SPAC merger, with a shareholder vote planned for June 25. Fewer qubits per logical qubit lowers the bar to fault tolerance, which is directly relevant to the Nordic quantum ecosystem: Chalmers and the Wallenberg center WACQT, KTH, NTNU and the Niels Bohr Institute in Copenhagen. If IQM can scale up, it could shorten the path to quantum usefulness for Nordic industry, from pharmaceuticals and materials research to cryptography and defense.
AI data centers’ hunger for memory is now feeding through to the price of ordinary RAM. As manufacturers redirect capacity to the memory that AI accelerators demand, the supply of conventional DRAM is shrinking – and prices are surging for consumers and businesses alike.
Producing one bit of HBM takes, according to industry analyses, about three times as much wafer area as the equivalent DDR5, while an HBM module sells for $60–100 versus $5–10 for DDR5 – so fabs prioritize the pricier AI memory. The result: DRAM prices rose roughly 90 percent in the first quarter of 2026, a cheap 32 GB DDR5 kit has climbed to around $375 from $80–120 a year ago, and Gartner forecasts memory costs could rise by as much as 130 percent. Framework has raised laptop memory prices by 50 percent, Micron has retired its consumer Crucial brand, and at HP memory and storage have grown from 15–18 to around 35 percent of a PC’s component cost.
For the Nordics it shows up directly in the wallet: pricier computers, phones and game consoles at chains such as Elgiganten, Power, Komplett and Dustin, and rising procurement costs for Swedish and Norwegian IT and industry – from Ericsson and Volvo to the public sector. At the same time the same scarce component competes with the Nordic data-center buildout: the server halls in Narvik and Luleå and ventures such as Telenor’s AI factory and Sweden’s Sferical AI are fighting for exactly the memory that is now hardest to get.
Swedish AI company Lovable is reportedly in talks over a new funding round that could value the Stockholm-based firm at up to $12 billion – nearly a doubling since December.
Lovable, which lets anyone build apps and websites by writing in plain language, was valued at $6.6 billion in a round in December 2025 and now says it has passed $400 million in annual recurring revenue. The company recently signed a multi-year agreement with Google Cloud and uses, among others, Gemini models. The talks are reportedly not concluded and no terms are set, so the valuation could change before the round closes.
For the Nordics, Lovable is among the strongest proof that Europe can build its own AI winners – alongside Stockholm’s Strawberry and Klarna. At the same time the company sharpens the question of what “vibe coding” does to jobs: when anyone can build software, the role of junior developers changes – something unions such as Unionen, Akavia and TechSverige are watching closely. That a European star still leans on America’s Google Cloud also underscores the Nordics’ dependence on foreign cloud infrastructure.
Anthropic released Claude Fable 5 on June 9, the company’s first public model in the new “Mythos class” that sits above the Opus line. The model handles text, images and files and has a 1 million token context window.
On the SWE-bench Pro coding test, Anthropic says Fable 5 reaches 80.3 percent, against 69.2 for Opus 4.8, 58.6 for GPT-5.5 and 54.2 for Gemini 3.1 Pro. Pricing is $10 per million input tokens and $50 output – roughly double Opus 4.8 but less than half of Mythos Preview. Fable 5 is included at no extra cost on Pro, Max, Team and Enterprise through June 22. Sensitive queries in cyber and biology automatically fall back to Opus 4.8, and a less restricted sister model, Claude Mythos 5, is available only through a closed trusted-access program.
For the Nordics, the launch sharpens the race over the Swedish and Norwegian language models. As a commercial top model pulls ahead, it raises the pressure on the National Library of Sweden (KB) and national coordinator Katrin Westling Palm’s work on a domestic Swedish model, as well as on WASP and Wallenberg-funded research. At the same time, Nordic banks such as SEB, Handelsbanken and DNB face the same issue as before with Mythos: the strongest cyber models are reachable only through trusted-access programs, leaving access uneven as Sweden’s MSB and Norway’s NSM work to raise their defenses against AI-driven cybercrime.
OpenAI updated GPT-Rosalind in early June, a specialized model for biology, drug development and translational medicine, named after DNA pioneer Rosalind Franklin.
The model combines GPT-5.5’s agentic coding with stronger reasoning in medicinal chemistry and genomics and, according to the company, uses 31 percent fewer tokens than GPT-5.5 on the same tasks. It is released as a research preview to selected organizations globally through a trusted-access setup, and OpenAI is at the same time launching a free Life Sciences plugin for Codex that connects the model to more than 50 scientific tools and data sources.
For the Nordics, the Danish pharmaceutical giant Novo Nordisk is one of the named customers and will use the model to weigh evidence across literature, genomics and structural biology. That puts pressure on Nordic life science more broadly: AstraZeneca, Karolinska Institutet – which this year appointed a Chief AI Officer – and research environments such as SciLifeLab, Uppsala and NTNU’s health AI gain both a new tool and a competitor to reckon with as AI shortens the path from hypothesis to medicine.
Stockholm-based Strawberry, described as Sweden’s next AI rocket after Lovable, has opened its AI browser to everyone and showed it off at Breakit AI Day 2026.
In the browser, teams can unleash AI agents that research several websites in parallel on their own, generate content and handle routines such as data entry and updating CRM systems. The company was founded by Charles Maddock, Arian Hanifi and Sebastian Thunman and has raised around 60 million kronor from investors including General Catalyst and EQT Ventures as well as the founders behind Lovable, Hugging Face and Supabase. “99 percent haven’t even started with AI agents yet,” Maddock says of the competition from the big players.
For the Nordics, Strawberry is proof that Swedish AI has begun shifting from chatbots to agents that take over entire work tasks. That makes the future of office jobs concrete for unions such as Unionen and Akavia and for Sweden’s Arbetsförmedlingen and Norway’s NAV, while it strengthens Stockholm as a Nordic startup hub in the wake of Lovable and Klarna. The challenge will be to combine fast automation with data protection when agents gain access to customer and business systems – an area where Sweden’s IMY and Norway’s Datatilsynet are already raising the bar.
At its congress, the Swedish Writers’ Union (Sveriges Författarförbund) landed a compromise on the sensitive question of whether members’ texts may be used to train the state’s planned Swedish AI language model.
The language model is to be developed within the WASP research program, funded by the Knut and Alice Wallenberg Foundation, after a government decision in February. A motion sought to pause the project and appoint an inquiry. The decision was that the union stays in but will not enter any agreements without first anchoring them with members, possibly via an extraordinary congress, and that an AI council will be set up where translators get several seats. Behind the concern, according to poet Olivia Bergdahl, lay a lack of member influence over the decision to take part at all.
For the Nordics, the dispute puts its finger on a knot both Sweden and Norway are wrestling with: how to build a sovereign language model without overriding rights holders. The same tension exists in Norway after NRK’s revelation about AI text at publisher Cappelen Damm, and it touches organizations such as the Norwegian Association of Non-Fiction Writers and Translators (NFFO) and Kopinor. How national coordinator Katrin Westling Palm and KB handle consent and compensation will be decisive – and will soon be tested against the EU AI Act and its transparency requirements for training data.
IQM, the Finnish company that builds superconducting quantum computers, has been given the green light by the U.S. Securities and Exchange Commission (SEC), which declared the company’s registration statement effective. That clears the way for a public listing, and shareholders in the SPAC Real Asset Acquisition Corp. vote on the deal on June 25.
Through the merger, IQM will go public with a planned listing on Nasdaq in New York under the ticker IQMX, followed by a parallel listing on Nasdaq Helsinki. The company is valued at around $1.8 billion and has strengthened its financing with an upsized $146 million PIPE that drew in investors including Finland’s pension giant Ilmarinen. Founded in 2018, IQM has sold 23 quantum computers and reported €31 million in revenue for 2025.
For the Nordics this is a milestone: IQM becomes one of Europe’s first listed quantum companies and a Finnish counterweight to the American and Chinese giants. The listing gives Nordic research environments such as Chalmers’ quantum initiative WACQT in Gothenburg, along with KTH and NTNU, a regional hardware supplier to lean on. It also aligns with the Norwegian national push on AI and quantum technology signed by Kongsberg, Equinor and SINTEF, and strengthens Nordic technological sovereignty in a field expected to be decisive for both pharmaceuticals and materials research.
On June 9, OpenAI launched a new platform, the OpenAI Economic Research Exchange, intended to fund and support independent research on how AI affects workers, firms and the wider economy.
Selected researchers gain access to the company’s tools under governed, privacy-protected arrangements in order to build credible evidence on AI’s effects on employment and productivity. Applications close on July 5 and decisions will be announced by July 31. The effort builds on the OpenAI Foundation’s pledge of $250 million for research into how AI is transforming working life.
For the Nordics the question lands in the middle of an ongoing debate about AI and entry-level jobs. Swedish research bodies such as IFAU, SNS and Ratio, as well as Norway’s Frisch Centre at the University of Oslo, hold register data unusually well suited to exactly this kind of study. At the same time, the setup raises the question of how dependent independent labor-market research should be on the very AI companies it is meant to scrutinize – a trade-off that concerns the public employment services Arbetsförmedlingen and NAV as well as unions such as LO when AI’s effects on jobs are to be measured and addressed.
New York-based Standard Bots raised $200 million in a Series C on June 9 and is now valued at $1 billion. The money will scale up its manufacturing of AI-driven industrial robots in the United States.
The company’s robot arms can learn a new task by watching it performed just once, thanks to AI that interprets the demonstration – an example of the “physical AI” now drawing investors. The round was led by RoboStrategy and General Catalyst and marks a sharp step up from the $63 million the company raised in 2024. The capital will go toward an expanded factory on Long Island and more engineers.
For the Nordics, industrial robots are home turf. Denmark’s Universal Robots and MiR in Odense are Europe’s leaders in collaborative robot arms and mobile robots respectively, and Sweden’s ABB Robotics in Västerås is among the world’s largest. When American challengers raise billions to bring robot manufacturing home to the United States, the pressure grows on Nordic manufacturers and on engineering-industry subcontractors, from Småland to Jutland, to keep pace in the race for robots that teach themselves.
In early June the White House issued an executive order, “Promoting Advanced Artificial Intelligence Innovation and Security,” meant to strengthen U.S. leadership in AI while addressing the security risks that come with ever more capable models.
The order relies on voluntary measures rather than binding rules: federal agencies are to harden their systems with AI-based defenses, prioritize criminal enforcement against AI-driven cyber activity, and build a framework in which developers can voluntarily give the government early access to so-called “covered frontier models” for up to 30 days before release. The deliverables are due on July 2 and August 1.
For the Nordics, the question of who gets to see the most powerful models first is highly relevant. If U.S. agencies routinely test frontier models before release, it could affect when and how Swedish and Norwegian users, companies and agencies such as FMV, the Swedish Armed Forces and MSB gain access to the same technology. It cuts to the same nerve as the EU’s DMA and AI Act: Nordic actors risk ending up last in line, which strengthens the case for the homegrown language models and sovereign AI clouds that Sweden and Norway’s Telenor are already investing in.
On June 8 at WWDC, Apple unveiled a completely rebuilt assistant, Siri AI, built on a new in-house foundation model developed together with Google. According to multiple reports, Siri runs on a custom Gemini model under a multi-year deal worth around one billion dollars a year.
At the same time, Apple is opening the platform: with iOS 27 Extensions, users can choose Claude, ChatGPT, Gemini or Grok as their preferred AI in Siri, Writing Tools and Image Playground. Apple thus abandons its previous single-provider model, lets different models have their own voices, and rolls everything out with iOS, iPadOS and macOS 27 in September. Claude and Gemini become the first third-party partners alongside ChatGPT.
For the Nordics, the catch is significant: Apple says that because of the EU's Digital Markets Act it cannot ship Siri AI on iPhone and iPad in the union at launch. Swedish, Danish and Finnish iPhone users will therefore go without the feature, while it is available on Mac, Apple Watch and Vision Pro. For Norwegian users the situation is unclear, since Norway sits outside the EU but often follows via the EEA. The question affects both consumers and Nordic developers building on Apple Intelligence, and again highlights how the DMA and the AI Act shape which AI services Nordic citizens actually get access to.
The US Department of Defense is now testing models from OpenAI, Google and xAI's Grok to replace Anthropic's Claude, which has been the primary AI in the Pentagon's classified systems and part of the Maven Smart System analysis platform.
The dispute stems from two lines Anthropic has held since its founding: no mass surveillance of Americans and no fully autonomous weapons that attack people without human control. After Anthropic refused to remove the guardrails, Defense Secretary Pete Hegseth declared the company a supply-chain risk, and the department had about 25 "power users" begin evaluating the competitors. Anthropic is challenging the decision in court and warns of losses in the billions.
For the Nordics, the question is highly concrete. Saab and Kongsberg are building AI into their systems, and agencies such as FMV, the Swedish Armed Forces and Norway's Forsvaret are buying ever more AI amid the NATO build-up. The case shows how a vendor's ethical guardrails can collide with military demands, and raises the question of how dependent the Nordic defense sector dares to become on US AI models whose terms can change overnight.
Broadcom warns that the limited capacity at contract manufacturer TSMC is holding back the supply of AI accelerators into 2026. TSMC's CEO has said supply is not expected to meet demand until 2027.
The bottleneck is above all memory: HBM, the fast high-bandwidth memory that AI chips need. Samsung, SK Hynix and Micron, which together control more than 95 percent of the world's DRAM production, have shifted capacity to HBM, and prices have, according to industry estimates, risen sharply – some increases of more than 200 percent since early 2025. On top of that, TSMC's advanced packaging, CoWoS, is a second bottleneck.
For the Nordics, the region's AI ambitions rest on the same scarce chips. Telenor's AI factory in Norway, Ericsson, Volvo and the new Nordic data centers all compete for GPUs and HBM. When supply tightens, both Swedish sovereignty efforts – such as the Berzelius supercomputer in Linköping and the national language model – and industry roll-outs risk higher costs and longer lead times. Those who plan compute early win.
In late May, Sweden's innovation agency Vinnova decided to invest 350 million kronor in six projects in which advanced digital technology and AI are to strengthen Swedish industry and society. The projects tackle challenges such as energy efficiency, climate benefits in the forestry industry and secure digital decision support.
The initiative is part of the Advanced Digitalisation partnership, a collaboration since 2021 between the state and industry in which Vinnova finances up to half of the budget. Industry is represented by ABB, Ericsson, Saab and Teknikföretagen, and among the projects is an Expert Learning Lab run by ABB, Ericsson, Saab and the Volvo Group together with Chalmers, KTH, Linköping University and Örebro University.
For the Nordics, it is a concrete example of how the state, industry and academia can share the bill to keep regional industry competitive as AI moves from pilot to production. The model is directly relevant to jobs and skills in Swedish industrial regions, and resembles the choices facing Norway's NHO, the Research Council and NTNU, as well as Danish counterparts, as AI is to deliver measurable value in manufacturing.