The days of seemingly unlimited free access to cutting-edge generative AI are suddenly over. Two giants in the artificial intelligence field, OpenAI and Google, have simultaneously imposed strict new daily usage limits on their most computationally intensive tools: OpenAI’s video generator Sora and Google’s latest image model Nano Banana Pro.
The sudden restrictions, which disproportionately impact non-paying users, highlight a harsh reality. Running cutting-edge generative AI systems consumes enormous amounts of computing power and energy, taxing even the world’s most sophisticated data centers.
Sora restricts Spark Viral’s “GPU melting” comments
The most dramatic cut came from OpenAI. Free users of Sora 2, the highly anticipated video generation model, are currently limited to six video generation per day. This represents a significant drop from the estimated 30+ generations that users reportedly received in the weeks following the model’s rollout.
The severity of this limitation was confirmed by Bill Peebles, Head of Sora at OpenAI, who explained the changes on social media platform X. “As our GPUs start to melt, we want to make sora accessible to as many people as possible,” he posted, adding, “We are setting a usage limit of 6 generations per day for free users.”
While the “GPU melting” comment is an exaggeration, it points to a real infrastructure crisis caused by AI video creation. Producing even a few seconds of high-fidelity video requires an exponential increase in graphics processing unit (GPU) time and power compared to producing simple text or still images. The new limits are an obvious attempt to allocate this valuable resource, especially during the holiday period when usage spikes.
Importantly, the new caps do not apply to paid ChatGPT Plus and Pro subscribers, confirming the company’s two-pronged strategy to protect the premium experience for its paying customer base while using the free tier as a powerful demonstration platform. Paid users can also purchase additional video generation tokens as needed.
Google strengthens access to Nano Banana Pro
Google, on the other hand, took a quieter but equally important path of tightening restrictions across its Gemini AI ecosystem.
Free users of Nano Banana Pro, an image generation and editing tool powered by the new Gemini 3 Pro architecture, now generate just two images per day. This is a decrease from the previously reported 3 images per day, and a significant decrease compared to the older Nano Banana model (with Gemini 2.5), which still lets you store up to 100 images per day for free.
The company’s support page offers a less dramatic but equally convincing explanation, citing “high demand” for image generation and editing and warning that “limits may change frequently and are reset daily.”
Restrictions apply to more than just images. Access to Gemini 3 Pro’s core conversation features has also been reduced for free users. While Google initially offered a clear quota of free prompts per day, the company has now replaced this guarantee with vague language, describing free usage as “basic access” with “daily limits subject to frequent change.”
Economics of generative AI
This simultaneous comeback by two market leaders highlights the economic frictions inherent in offering cutting-edge AI for free.
“The most direct reason for these limitations is computational cost,” explains Dr. Lena Chen, AI economist at the Future of Tech Institute. “Using Sora to generate a single complex video is effectively subsidizing a huge portion of a free user’s GPU time. As demand grows, that cost becomes unsustainable for the company if it wants to maintain quality of service for everyone.”
The move is widely seen as a commercial pivot to transform generative AI from an experimental “free buffet” into a service that can be more clearly monetized. By limiting the free tier, the companies are creating a strong incentive for heavy-user artists, content creators, and professionals to move to a paid subscription model that offers substantially higher and more stable usage limits. The value proposition for these users is simple. You pay for priority access and guaranteed computing resources.
For casual users, the new limits will require you to allocate your daily allowance more carefully. Two images or six videos may be enough for light experimentation, but creators who rely on heavy iterations will find their creative workflow severely limited unless they upgrade their subscription.
This new limit clearly signals that the huge operational costs associated with powerful generative models are finally beginning to outweigh the benefits of providing unlimited free access, ushering in a more sustainable, if less generous, phase in the consumer AI landscape.
