My Conscium co-founder Daniel Hulme spoke to Tech.eu about DeepSeek, the Chinese lab whose R1 model release in January 2025 rattled assumptions about how much capital and compute you actually need to compete at the frontier of AI. Daniel's take was more optimistic for Europe than most of the commentary I'd seen at the time, and I think it's held up well since.
Why DeepSeek changed the calculation for European AI
The dominant story about frontier AI for the past few years has been that you need enormous capital and enormous data centres to compete, which structurally favours a handful of well-funded American labs and, increasingly, Chinese ones. DeepSeek's release undercut that story by showing a model competitive with the best in the world could be built with far less brute-force spending than assumed. Daniel's argument, and the piece's central one, is that this reopens the door for European AI companies that were never going to out-fund OpenAI or Google, but might be able to out-think them.
Daniel's specific bet: agentic AI is the wide-open lane
Where Daniel got concrete was on agentic AI, the shift from models that answer questions to systems that carry out multi-step tasks with some autonomy. His view, as he put it to Tech.eu, is that "the competition is now moving towards agentic AI and the field is wide open." I read that as a direct challenge to European founders: the frontier-model race may already be decided for now, but the agent layer built on top of these models is still up for grabs, and that's a race Europe can actually enter.
Machine consciousness as an underdeveloped frontier
Daniel also flagged something closer to Conscium's core work: that machine consciousness research remains underdeveloped relative to how fast capability is advancing, and that 2025 could be a turning point for taking it seriously as a field rather than a curiosity. I'd add that this is exactly the gap we started Conscium to close, and it's encouraging to see it raised in a piece aimed at European startup founders rather than only in academic circles.
What other voices in the piece got right
The article pulled in other perspectives worth noting alongside Daniel's. Dataconomy's Elena Poughlia made the case that DeepSeek's approach is cheaper and more sustainable than the brute-force alternative. Block Dojo's Alex Ball argued companies should compete on user relationships rather than chasing a marginally better model. Savery's Anders Ibsen made a point I agree with, that Europe's advantage was always more likely to come from algorithmic sophistication than from chip manufacturing capacity we don't have. Not every voice was purely optimistic, WhomLab's Carol Constant raised the geopolitical and regulatory risks that come with building on a Chinese open-source model, which is a fair caveat to Daniel's otherwise upbeat read.
Read the full piece on Tech.eu>>