DeepMind CEO: The West Needs to Open Source AI Stack, Already Losing to China
DeepMind CEO Demis Hassabis stated in an interview with Y Combinator that he hopes to see the West establish a complete open-source AI technology stack, as it is currently lagging behind China in the AI race.
Google's computing power is insufficient to support two cutting-edge models (open-source and closed-source) simultaneously, so the Gemma series has chosen to develop a smaller family of models.
Western AI developers and companies are increasing their investment in open-source stacks, shifting funding from a single closed-source giant to diversified open-source infrastructure development. DeepMind/Google and Western open-source projects will benefit, while China's closed-source AI leaders face intensified geopolitical competition pressure.
Source: Public Information
ABAB AI Insight
Demis Hassabis has long led the breakthroughs of the Alpha series since DeepMind was acquired by Google in 2014. This public call for a Western open-source stack reflects his recent concerns about the AI competition between China and the U.S. He has previously pushed for Gemma to be open-sourced internally to counter China's full-stack advantage.
In terms of capital strategy, Google prioritizes limited computing power for the closed-source Gemini flagship while using Gemma's open-source to attract community contributions and external data, forming a dual-track strategy of "closed-source leadership + open-source ecosystem buffer." However, Demis candidly stated that the computing power bottleneck has limited the simultaneous advancement of both lines, forcing the West to collaborate through open-source to mitigate risks and accelerate catch-up.
Similar to how the Meta Llama series quickly attracted global developers through open-source, Google/DeepMind is in the midst of transitioning Western AI from "closed-source monopoly" to "open-source full-stack competition." The limited scale of Gemma is a direct result of computing power allocation.
Essentially, this reflects capital concentration: the traditional single closed-source path of Google has been disrupted by the competition for computing power and talent between China and the U.S. Demis calls for the West to concentrate dispersed computing power, talent, and data into a common technology stack through open-source collaboration, reconstructing AI development from "company island competition" to a resource allocation mechanism of "Western alliance full-stack confrontation."