OpenAI and Google are imposing new daily limits on their most widely used generative-AI tools as soaring demand strains the companies' underlying GPU infrastructure, marking one of the most significant capacity-related pullbacks since the current AI wave began. The restrictions, which took effect in November 2025, apply largely to free users of video and image generation services such as OpenAI's Sora and Google's Nano Banana Pro, while paid customers retain broader access. The moves come as companies acknowledge that the rapid expansion of AI image and video features is pushing hardware to its limits.
OpenAI now caps free Sora users at six video generations per day, a shift that the company framed as necessary to preserve system stability amid unprecedented usage. Paid ChatGPT Plus and Pro subscribers remain unaffected and may buy additional video generations. The decision follows a surge in use linked to the rollout of native image and video capabilities within ChatGPT, which significantly increased the computational load across OpenAI's data-center network.
Google has implemented similar constraints. The company reduced the free daily allowance for Nano Banana Pro users to two generated images, down from three. Other services within Google's broader Gemini lineup are being placed into a "basic access" tier during periods of high demand, giving the company flexibility to adjust resource availability dynamically. The adjustments reflect the same underlying pressure affecting OpenAI-mass adoption of generative media tools that rely on heavy GPU lifecycles and extensive cooling systems.
Sora's engineering leadership described the situation bluntly in a public post explaining the changes. "our gpus are melting," the team wrote, adding, "we're setting usage limits for free users to 6 gens/day. chatgpt plus and pro users have unchanged limits, and everybody can purchase additional gens as needed. our gpus are melting, and we want to let as many people access sora as possible!" The statement underscores the scale of the operational challenge, as the hardware required to generate high-resolution AI video consumes exponentially more power and computational capacity than text-based queries.
Industry analysts note that the rapid commercial expansion of generative-media tools has outpaced early infrastructure planning. GPUs optimized for training large models remain scarce and expensive, while inference workloads-what users tap every time they request a video or high-fidelity image-continue to rise. The growing mismatch has forced companies to reassess how much free output they can sustainably provide without degrading service quality for paying customers.
The limits also signal a strategic business shift. Free tiers that once served as generous gateways into AI ecosystems are tightening, nudging heavy users toward subscription models or microtransactions. Executives at both companies have privately acknowledged that the economics of large-scale video and image generation require more disciplined capacity management as energy consumption and compute costs rise.
The changes pose a new challenge for creators, educators and small businesses that relied heavily on free tools to generate content at scale. While casual users may find six daily videos or two images sufficient for experimentation, frequent content producers face new constraints that could either push them toward paid tiers or lead them to scale back production.
