Google AI Adjustments or Recalculation of Computing Power Diffusion
Recent adjustments in Google's AI leadership indicate a possible prioritization of AI diffusion over leading-edge model advancements, betting on AI computing power as a larger economic opportunity.
Demis Hassabis has stepped down from daily operations at DeepMind to become chairman and chief scientist of Alphabet, while Jeff Dean and his team have left to create Discovery Loop, with Koray Kavukcuoglu taking over operations; SemiAnalysis estimates that Gemini's annual recurring revenue is about $12 billion, while GCP's third-party AI revenue and TPU sales could exceed $200 billion by the end of 2027.
Resources are shifting towards cloud infrastructure and computing power sales, with the focus of R&D moving from pure frontier competition to commercial diffusion, benefiting Google Cloud and TPU businesses, while independent frontier labs face pressure.
Source: Public Information
ABAB AI Insight
Historically, Google has balanced research and products through internal restructuring, from the merger of Google Brain and DeepMind to the unification of Gemini, and now this executive diversion, continuing its path of monetizing technology through infrastructure advantages rather than merely chasing model rankings.
Funding and talent are shifting from daily frontier model operations to cloud services and external TPU sales, motivated by the need to capture the explosive demand for computing power post-AI implementation, positioning itself as a "computing landlord"; similar strategies can be seen in its early monetization of search ads and expansion of the Android ecosystem.
In comparison to Microsoft's collaboration with Azure and OpenAI, and Amazon's AWS investments in AI infrastructure, the current AI industry is transitioning from a model arms race to large-scale deployment and commercialization.
Essentially, this represents a transfer of pricing power: as frontier models become homogenized, the supply and distribution of computing power become the core profit pool, driven by the high costs of large-scale training and inference, allowing giants with self-developed chips and cloud networks to secure long-term revenue through diffusion.
ABAB News · Cognitive Laws
Diffusion can harvest more profits than invention
Computing landlords are the ultimate winners
Leading-edge does not equal commercial victory