Google has imposed strict usage limits on free access to its newly launched Gemini 3 Pro model and Nano Banana Pro image generator, citing overwhelming demand and capacity constraints.
The shift marks a significant departure from the generous allocations promised just days after the models' November 18 and November 20 rollouts, respectively.
Upon launch, free users received straightforward daily allowances: up to five text prompts with Gemini 3 Pro and up to three image generations with Nano Banana Pro, matching the previous allocations for Gemini 2.5 Pro. Within a matter of days, these fixed limits vanished.
Free users now receive only "basic access" to Gemini 3 Pro with daily limits that "may change frequently" depending on server load. In practical terms, users report reaching their limits after just three prompts. The Nano Banana Pro image generation limit has been reduced from three images to two per day, with Google explicitly noting that image generation is "in high demand".
Google's infrastructure chief, Amin Vahdat, disclosed internally that the company must double its serving capacity every six months to sustain the exponential growth in demand for AI services. The company faces the challenge of achieving a thousand-fold increase in capacity over the next four to five years.
This reflects a broader shift in the AI industry's priorities: while compute capacity for training models dominated previous years, serving capacity—the ability to deliver responses to millions of concurrent users—has emerged as the critical bottleneck.
The capacity strain extends beyond Gemini's free tier. NotebookLM, Google's document-analysis product, temporarily rolled back access to newly launched Nano Banana Pro-powered Infographics and Slide Decks features for free users entirely.
Pro subscribers face unspecified "additional limits," suggesting that even paying customers are experiencing restrictions.
Paid subscribers remain insulated from these constraints. Users with Google AI Pro subscriptions retain their 100 daily prompts, while Google AI Ultra subscribers maintain their 500 daily prompts.
Image generation limits for paid tiers have similarly remained unchanged.
The demand surge reflects the computational intensity of the new models. Gemini 3 Pro represents a substantial leap in multimodal reasoning and performance capabilities.
Nano Banana Pro, built on the same foundation, excels at rendering accurate text within generated images and maintaining character consistency across multiple images—capabilities that demand significantly more computational resources than their predecessors.
OpenAI faces similar infrastructure challenges. The company announced that free users can generate only six videos per day with Sora 2, down from previous allocations, with OpenAI's head of Sora acknowledging that "our GPUs are melting".
Like Google, OpenAI has prioritized paid subscribers, leaving their limits unchanged.
The physical infrastructure required to support AI services at scale presents a formidable constraint. Google has committed $75 billion to capital expenditure in 2025, with the majority directed toward data centers and computing equipment.
Despite these investments, power consumption, cooling capacity, networking bandwidth, and the sheer time required to construct and energize new data center facilities remain limiting factors. These are not issues that capital spending alone can quickly resolve; physical infrastructure deployment operates on timelines measured in years rather than weeks.
The move to variable daily limits rather than fixed allocations allows Google to respond dynamically to server load without the need for ad hoc announcements.
As demand fluctuates or new capacity comes online, limits can adjust automatically. During periods of peak usage, free users may find themselves with minimal access; during off-peak hours, allocations might expand.
Google has committed to restoring unrestricted access to free users once serving capacity increases sufficiently. The timeline for such restoration remains undefined, contingent on the company's ability to deploy new infrastructure and optimize existing systems.
For users unable to work within the constrained free tier, the upgrade path to Google AI Pro at $19.99 per month or Google AI Ultra at $249.99 per month offers a route to predictable, higher-volume access.
The situation underscores a fundamental challenge facing the entire AI industry: the gap between computational capability and infrastructure for serving that capability to users.
While generative AI models continue advancing in capability and efficiency, the physical systems required to deliver them to users worldwide remain difficult to scale rapidly.
