Google restricts Gemini access amid power shortages while Meta’s development stalls

As generative AI adoption accelerates, computing power has emerged as the critical asset driving competition among major technology firms. Recently, Google has had to restrict resource allocation due to surging demand for its Gemini AI platform, responding to the escalating workload from developers and enterprise clients.
Starting in spring 2025, requests for the Gemini API doubled, creating a scarcity of essential computing resources. To ensure equitable access across the ecosystem, Google introduced a quota-based usage limit on May 17, 2026, employing a rolling update system akin to mobile data plans. Requests exceeding these quotas are subject to rate limiting, ensuring that all partners retain access to basic API functionality despite constrained resources.
Meta has been the most significantly affected by these changes. Reports indicate that Meta's demand for model capacity far exceeded that of other clients, leading Google to explicitly state it could not fulfill Meta's requested computing quota. This abrupt restriction disrupted several internal AI initiatives at Meta. In response to these constraints and to align with a strategic focus on reducing AI research and development costs, Meta has urgently instructed employees to optimize API calls, enhancing the efficiency of code execution and token utilization.
Industry analysts note that this resource adjustment highlights a critical challenge in global AI infrastructure: while large model capabilities continue to advance, the development of foundational computing infrastructure supporting their operation has not kept pace with the explosive growth in application demand. For companies like Meta, which rely heavily on cloud computing, maintaining research and development velocity under these constraints will be a decisive factor in future competitive landscapes.
Related article
ByteDance Unveils Seedance 2.5, Marking the Start of 30-Second Single-Take Film Production
ByteDance Seed team has officially unveiled Seedance 2.5, a next-generation video creation model designed for audio-visual joint generation with a focus on "one-shot filming." This latest iteration delivers breakthrough enhancements across three key
Yanyu Technology Eyes Hong Kong IPO After 3-Year, 3B USD Valuation
As global enthusiasm for large AI models intensifies, can an "intermediary" bridging models and end-users sustain a massive capital narrative without building foundational models themselves? Recent media reports indicate that Evoken has begun prepari
What 3 Top VCs Really Think About the AI Frenzy
During TechCrunch’s StrictlyVC event in Athens, part of the Panathēnea festival, I spoke with Niko Bonatsos of Verdict Capital, Andreas Stavropoulos of Threshold Ventures, and Ben Blume of Atomico about the current venture landscape, the impending Sp
Related Special Topic Recommendations
Comments (0)
0/500

As generative AI adoption accelerates, computing power has emerged as the critical asset driving competition among major technology firms. Recently, Google has had to restrict resource allocation due to surging demand for its Gemini AI platform, responding to the escalating workload from developers and enterprise clients.
Starting in spring 2025, requests for the Gemini API doubled, creating a scarcity of essential computing resources. To ensure equitable access across the ecosystem, Google introduced a quota-based usage limit on May 17, 2026, employing a rolling update system akin to mobile data plans. Requests exceeding these quotas are subject to rate limiting, ensuring that all partners retain access to basic API functionality despite constrained resources.
Meta has been the most significantly affected by these changes. Reports indicate that Meta's demand for model capacity far exceeded that of other clients, leading Google to explicitly state it could not fulfill Meta's requested computing quota. This abrupt restriction disrupted several internal AI initiatives at Meta. In response to these constraints and to align with a strategic focus on reducing AI research and development costs, Meta has urgently instructed employees to optimize API calls, enhancing the efficiency of code execution and token utilization.
Industry analysts note that this resource adjustment highlights a critical challenge in global AI infrastructure: while large model capabilities continue to advance, the development of foundational computing infrastructure supporting their operation has not kept pace with the explosive growth in application demand. For companies like Meta, which rely heavily on cloud computing, maintaining research and development velocity under these constraints will be a decisive factor in future competitive landscapes.
ByteDance Unveils Seedance 2.5, Marking the Start of 30-Second Single-Take Film Production
ByteDance Seed team has officially unveiled Seedance 2.5, a next-generation video creation model designed for audio-visual joint generation with a focus on "one-shot filming." This latest iteration delivers breakthrough enhancements across three key
Yanyu Technology Eyes Hong Kong IPO After 3-Year, 3B USD Valuation
As global enthusiasm for large AI models intensifies, can an "intermediary" bridging models and end-users sustain a massive capital narrative without building foundational models themselves? Recent media reports indicate that Evoken has begun prepari
What 3 Top VCs Really Think About the AI Frenzy
During TechCrunch’s StrictlyVC event in Athens, part of the Panathēnea festival, I spoke with Niko Bonatsos of Verdict Capital, Andreas Stavropoulos of Threshold Ventures, and Ben Blume of Atomico about the current venture landscape, the impending Sp





Home






