Background: Google Limits Meta's Gemini Access
According to sources, Google has quietly imposed usage caps on Meta's access to its Gemini model since March 2026, due to its own AI computing power shortage. This restriction directly caused delays in multiple internal AI projects at Meta. Although both parties had previously signed a computing resource agreement, Google prioritized its own business needs amid tight supply, slashing Meta's access.
Meta's Countermeasures: Efficiency Boost and Model Self-sufficiency
Faced with the bottleneck, Meta adopted a two-pronged strategy. Internally, it optimized computing efficiency through scheduling and model compression to reduce dependency on Gemini. Externally, it accelerated the development of its in-house Muse Spark model, designed for efficient AI inference while minimizing reliance on external computing resources.
Google's Supplementary Move: High-value Computing Lease with SpaceX
To ease the crunch, Google simultaneously signed a substantial computing lease agreement with SpaceX, leveraging Starlink's edge computing capacity to augment its data center supply. This move signals that traditional cloud providers' computing reserves are insufficient to meet surging AI inference demand, prompting the industry to seek novel computing sources.
The incident underscores the severe imbalance between supply and demand for AI inference computing. For crypto projects and enterprises dependent on third-party cloud computing, the stability of the computing supply chain has emerged as a critical risk, elevating the importance of self-developed chips and decentralized computing networks.