We do not know if the emergence of Ineffable Intelligence on the European technological scene is still an exception to the classic sequences of venture capital, or if it announces the emergence of a new investment standard, of which certain recent operations, like AMI Labs, could constitute the first milestones.
With a Seed round of 937 million euros, valuing the company at 4.3 billion euros, the London startup is not just entering the market and is challenging the hegemony of language models by banking on intelligence built through experience.
Since the emergence of major language models, the industry has gradually aligned itself with a paradigm that has become dominant. The systems developed by OpenAI, Anthropic or Google DeepMind are based on a massive accumulation of data from human production. Their effectiveness lies in the ability to model large-scale statistical regularities, supported by a continuous intensification of computing resources. This approach allowed a rapid diffusion of use cases, from software co-pilot to conversational agents, and contributed to installing LLMs as a standard infrastructure.
This framework, as stabilized as it appears, nevertheless reveals structural limits. Dependence on existing data raises legal and economic, but also epistemological, questions. These systems, as sophisticated as they are, remain fundamentally dependent on the corpus on which they are trained. Their capacity to produce truly new knowledge, beyond recomposition or extrapolation, remains uncertain. Alternative approaches are certainly being investigated.
It is one of them that the project led by David Silver addresses. Former head of reinforcement learning at Google DeepMind, he distinguished himself through work that marked a turning point in the discipline. Systems like AlphaGo or AlphaZero have demonstrated that an agent can achieve unprecedented levels of performance by learning through interaction, without depending on annotated databases. By replacing imitation with a logic of exploration and optimization, this work has opened an alternative path, which for a long time remained confined to closed environments.
The ambition of Ineffable Intelligence is to design a system capable of acquiring skills and producing knowledge from its own experience. Where language models learn by observing human traces, the approach adopted assumes direct interaction with an environment, real or simulated, in which the agent tests and refines its behaviors.
The distinction between these two paradigms refers to two conceptions of artificial intelligence. On the one hand, systems which reproduce and recompose existing structures from accumulated data. On the other, agents who gradually build their understanding through action. This second path, as promising as it may be, still faces significant constraints. Modeling sufficiently rich environments, managing computational costs linked to exploration, or even the stability of large-scale learning processes constitute all unresolved challenges.
The composition of Ineffable Intelligence’s funding round reflects the nature of this bet. Sequoia Capital and Lightspeed Venture Partners include leading technology players, including NVIDIA and Google, as well as British public institutions. This convergence, unusual at this stage of development, suggests that the project is seen less as an immediate commercial opportunity than as a strategic option on the future evolution of AI.
The involvement of the British Business Bank and the Sovereign AI Fund introduces a political dimension and is part of a desire to position the United Kingdom in a still unstructured segment of the market, by focusing on a differentiating approach rather than on catching up with existing models. In a context of increased competition between large economic zones, this strategy reflects a reading where the mastery of learning architectures becomes an issue of sovereignty.
Competitively, Ineffable Intelligence is more in the wake of advanced research laboratories, such as DeepMind or OpenAI, which combine fundamental research and industrial deployment, than in that of the startup ecosystem in the strict sense.
For investors, the operation also deviates from the usual logic of venture capital. The valuation of 4.3 billion euros is based less on execution indicators than on the credibility of a scientific thesis and the ability of a team to demonstrate its viability. In this regard, David Silver’s profile gives particular legitimacy to the project. But the bet remains uncertain and takes place over a long period of time, where the economic benefits are neither immediate nor guaranteed.
Basically, the Ineffable Intelligence initiative does not call into question the relevance of language models, but questions their definitive nature. As their limitations become apparent, the question of learning less dependent on human data gains importance. It remains to be determined whether this approach can be made operational on a large scale. Between consolidation of an established paradigm and exploration of alternative trajectories, artificial intelligence today seems engaged in a phase of bifurcation whose outcome remains open.
