Google capped Meta's access to its Gemini models in March, disrupting internal AI projects. This restriction, still in place, is the sharp end of a compute shortage now impacting everyday Gemini users. Google now rations AI usage by prompt complexity, model chosen, and chat length for all consumers.
How We Got Here
At its first-quarter earnings in April, Google CEO Sundar Pichai admitted the company was "compute-constrained in the near term." This followed signed-but-undelivered cloud contracts nearly doubling to over $460 billion, indicating demand far outstripping supply.
The Numbers
- Google limited Meta's Gemini capacity in March after the social media giant sought more than Google could supply.
- Google signed a $920 million-a-month deal to lease computing capacity from Elon Musk's SpaceX to cope with demand.
- Anthropic, maker of Claude, struck a similar leasing arrangement with SpaceX, highlighting a market-wide structural shortage.
- Google's Gemini Apps help page states free user limits may be cut before paying subscribers if capacity changes.
- The highest cost in running AI models now comes from inference—the work of processing user prompts—not just initial training.
What Happens Next
🇮🇳 Why This Matters for India
For Bangalore's AI startup founders relying on free tiers for rapid prototyping, Google's move means sooner-than-expected graduation to paid plans or architecting for leaner models.
The Take
Google's move clarifies the unsustainability of free, high-compute AI inference for mass users. This forces developers to optimize prompts and models from day one, accelerating the race for efficient, smaller AI solutions.
Source:
MediaNama ↗