AI · Nov 9, 2026
OpenAI’s Control Room puts a fail-closed kill switch inside the inference engineA DeepMind veteran bets fusion can be won in simulation, and Daniel Ek is buying in
· Updated · Signal Desk newsroom
Analysis of reporting by TechCrunch. How we work

Europa Fusion’s founder left Pacific Fusion’s machine learning team convinced the field leans too hard on hardware. The Spotify founder’s backing makes it a test of whether AI-era methods transfer to the hardest physics problem in energy.
Key takeaways
- Spotify founder Daniel Ek is backing Europa Fusion, a stealthy UK fusion startup founded in February, according to TechCrunch; Bloomberg first reported the investment.
- Founder Nantas Nardelli previously worked at Google DeepMind, Facebook AI Research and Microsoft Research Cambridge, and was head of machine learning at Pacific Fusion.
- He reportedly believes fusion startups should lean more on simulation and data than on physical test devices.
- Europa Fusion will pursue pulsed power, the same broad approach as Pacific Fusion.
- No funding amount, team size or timeline has been disclosed.
What is actually known
This is a thin story about a potentially interesting idea, and it is worth being clear about which is which. TechCrunch reports that Daniel Ek, the Spotify founder, is backing Europa Fusion, a UK fusion power startup that was founded in February and has kept quiet since. Bloomberg broke the investment. Sources told TechCrunch that the founder, Nantas Nardelli, had been working on the concept for longer than the company has existed. There is no disclosed round size, valuation, team or target date.
What makes it notable is Nardelli’s path. His resume runs through Google DeepMind, Facebook AI Research and Microsoft Research Cambridge, and most recently he led machine learning at Pacific Fusion, the California startup that has broken ground on a demonstration facility in New Mexico. He says his original background is electronics and control engineering, and that he used to build robots. His stated reason for the switch is climate.
The contrarian bet: fewer machines, more modeling
The interesting claim is methodological. Like most fusion companies its size, Pacific Fusion relies heavily on physical devices to test theories and refine simulations. According to a source familiar with Nardelli’s thinking, he disagreed with that emphasis and believes fusion startups should invest in a more simulation- and data-heavy approach. Europa Fusion will still chase pulsed power, in which huge, precisely timed surges of current compress a fuel target until its atoms fuse. So the bet is not a new physics concept; it is a different way of searching the design space.
We find the logic easy to understand and hard to verify. AI-era tooling has shown real gains where simulators are trusted and iteration is cheap; Nvidia’s recent science push, for instance, explicitly pitched new CUDA-X tools at fusion and other physics workloads. But fusion has a long record of plasmas behaving differently from the models that predicted them. Simulation can narrow the search. It cannot replace the moment a target is actually compressed and either yields energy or does not. A startup that skimps on hardware may move faster on paper and discover its errors later and more expensively.
Why tech money keeps flowing toward fusion
The broader pull is obvious. No one has yet built a fusion plant that supplies electricity to the grid, but a working one would be a carbon-free source drawing on an essentially limitless fuel supply, and it would ease the growing worry about AI’s appetite for power. That is why the same people who profit from compute keep writing checks for exotic energy. The grid strain is not abstract; as we noted in our look at AI’s power crunch, data-center architecture alone can knock large amounts of load off a system.
There is something fitting about an AI researcher leaving the field to work on the energy problem that AI is sharpening. It is also a reminder that fusion timelines are measured in decades, while the data-center build-out is happening now. Backing Europa Fusion is a long-dated option, not a near-term answer to anyone’s power bill. The responsible reading is curiosity, not celebration, until the company shows what its simulations can predict and how it plans to test them.
Who it matters to
- Fusion startups
- A credible ML-first entrant challenges the hardware-heavy playbook and will compete for the same simulation talent.
- AI researchers
- Another example of machine learning specialists moving into physical-science problems where data is scarce and stakes are high.
- Energy-hungry AI companies
- A distant but symbolically important hedge against the power constraints shaping data-center growth.
What to watch
- The size and lead of the round once Europa Fusion comes out of stealth.
- Whether it publishes simulation results or benchmarks against experimental pulsed-power data.
- Hiring of plasma physicists and experimental engineers, which would signal how much hardware it really plans to build.
- Any partnerships for access to existing pulsed-power test facilities.
FAQ
Who founded Europa Fusion?
Nantas Nardelli, a former researcher at Google DeepMind, Facebook AI Research and Microsoft Research Cambridge, who most recently led machine learning at Pacific Fusion.
What approach will it use?
Pulsed power, which uses massive, precisely timed electrical surges to compress fuel targets, with a heavier emphasis on simulation and data than most rivals.
How much did Daniel Ek invest?
The amount has not been disclosed.
More on AI
AI · Nov 11, 2026
OpenAI’s $2 million for children’s AI safety research matters for one promise: publishing bad newsAI · Nov 17, 2026
OpenAI says its models now shave area off production chips, including AMD’s next flagshipsAI · Nov 5, 2026
NVIDIA’s four open Cosmos 3 Nano models let robot and driving teams simulate worlds that reactAI · Nov 3, 2026
OpenAI’s GPT-5.6 Cyber reaches AWS GovCloud through Daybreak Red, with two approval gates