Google has reportedly limited Meta’s access to its Gemini AI models after it was unable to provide all the Gemini computing capacity the company sought, according to the Financial Times.
Google, owned by Alphabet, told Meta around March that it could not provide all the Gemini computing capacity the company wanted to purchase, as per the report. This highlights the increasing pressure on AI infrastructure as demand for the computing power needed to support advanced AI models and services continues to outpace available capacity.
The Financial Times reported that the restrictions have disrupted and delayed some of Meta’s internal AI projects and remain in place. Several other Google customers have also reportedly been affected by the capacity constraints, although Meta has been among the hardest hit because of its exceptionally high demand for Google’s models.
Meta has encouraged employees to use AI resources more efficiently by limiting unnecessary AI token usage due to the restrictions and a broader effort to streamline AI costs.
Despite spending tens of billions of dollars on chips, data centres and power infrastructure, the world’s largest AI providers continue to face growing demand for advanced AI models and services. Reflecting those pressures, Google earlier this month signed a $920 million-a-month agreement with SpaceX to lease computing capacity as it works to meet rising demand.
During Google’s first-quarter earnings, CEO Sundar Pichai said the company was “compute-constrained” in the near term and that Google Cloud revenue would have been higher if it had been able to meet customer demand.
Meta has begun to prioritize its new Muse Spark model, which is viewed as more competitive with Gemini and reduces the company’s dependence on external models for some applications, said FT report.
Gemini has been used internally at Meta to help automate safety processes, including scam detection and harmful content moderation. It is also used internally for some workflows and coding, alongside other models such as Anthropic’s Claude.
Unlike Google, Meta does not have a cloud business and is racing to build out its fleet of data centres for its own training and inference needs. As part of the push, Meta has committed to investing $600bn in the US by 2028.
The reported restrictions come as companies across the AI industry search for ways to address growing compute constraints. While cloud providers are expanding data centres and chip capacity, others are pursuing alternative approaches. Earlier this year, Huawei said its Tau Scaling Law and LogicFolding architecture could improve AI chip performance by changing how chips are designed, rather than relying only on shrinking transistors.
