DeepSeek’s open-source release of R1 and the research underpinning its design is an essential milestone for AI’s growth. To date, high costs have made numerous AI applications unsustainable. Now, this new architecture has made it possible to deploy AI models at a fraction of the cost, permitting a whole new wave of businesses to launch AI profitably.
That these advancements emerged from a venture in China is proof positive of the adage that “necessity is the mother of invention”. When DeepSeek was forced to navigate compute constraints caused by chip export restrictions, it leveraged its regional strengths by using a team of Ph.D.s from China’s top universities to design and implement novel techniques.
Markets have reacted accordingly, questioning the value of nine-, ten- or 11-figure investments into AI. The question for European investors is twofold. One, if compute and model design are no longer the barriers to entry they once were, what are the remaining set of barriers? And two, how should Europe think about participating in the AI race going forward?
The moat problem: cost-efficiency and the question of compute
Traditionally, many have believed compute access and model design to be key moats (i.e. competitive advantages accrued through a unique usage of AI). The first was perceived as a structural advantage to better model performance, underpinned by massive GPU clusters designed to train models with hundreds of billions of parameters and run inferencing at scale.
The second recognised moat consists of proprietary architectures. It is worth noting that while moves such as the publication of research papers and the release of open-source models had already shallowed this moat, the underlying data remains a source of defensibility.
In response, DeepSeek has built upon publicly available research from major US and European institutions and companies. Its use of a mixture-of-experts architecture aligns with research done by companies such as DeepMind, and it appears to have implemented strategies such as “fill-in-the-middle” or YaRN originally proposed by OpenAI and other Western institutions. It has managed to apply these techniques where others have struggled but has also introduced novel innovations, such as Multi-head Latent Attention.
Now, with DeepSeek-R1 released as an open-source initiative, it will benefit the wider AI community, contributing to the erosion of traditional AI moats. Greater compute efficiency will reduce the need for access to large hardware capabilities and bring down AI’s high running costs, while open-sourcing the models and openly sharing their research will further weaken the advantage of proprietary architectures.
With a key bottleneck eliminated, more players will enter the field and lower prices are expected to stimulate demand. The biggest impact will be on applications that were previously locked out due to prohibitively high op-ex costs. As such, we can expect a new wave of competition and applications in the AI market.
What DeepSeek means for European investors
With the erosion of traditional moats, the obvious question is what defensibility looks like going forward. For investors and founders searching for ways in which AI companies can differentiate themselves, there are some initial areas to focus on.
First, proprietary data remains a significant advantage, especially in domains where data is limited and hard to access such as healthcare. Next, vertical AI applications, where startups control both the distribution and development of models could offer a more defensible path to scale. Finally, differentiation may come less from building foundational models and more from fine-tuning and application-layer innovations that drive real-world adoption.
For European investors, this reinforces the importance of a deep technical understanding of AI and its rapidly evolving surroundings. This will be crucial in identifying the right investment opportunities with genuine moats and lasting defensibility.
Europe has historically lagged behind global rivals in the AI race, and this breakthrough could level the playing field. This continent has great strengths in industries such as aerospace and financial services, world-leading education with five out of the top ten universities in the QS World University Rankings, and a fantastic store of data from various European initiatives.
To capitalise on this moment, investors should focus on where they have leverage: backing companies that exploit data advantages, dominate high-value verticals, or build the infrastructure layers that AI adoption depends on.
A post-DeepSeek investment playbook
This moment is both a challenge and an opening. Now, defensibility lies in who controls the critical data, distribution, and infrastructure layers: not just who trains the biggest models. The winners will be those who recognise where scarcity still exists and how AI can be leveraged within high-value, defensible markets.
This demands a more technical investment approach. The strongest opportunities will come from companies with privileged data access, deep integration into industry workflows, and clear monetisation. The democratisation of AI does not mean all competitive advantages will disappear – they will just be found elsewhere.
Crystal van Oosterom is an AI Venture Partner at OpenOcean
Read more: We urgently need to bridge the IT-OT cybersecurity divide
