Google's Veo 3 has quickly become one of the most visible tests of consumer demand for AI video. According to Google Deepmind CEO Demis Hassabis, users generated millions of AI videos in just a few days after launch, creating a surge that tested Google's systems and forced the company to move fast on availability, infrastructure, and labeling.
A fast start for Veo 3
Veo 3 did not ease into the market quietly. The model drew enough activity in its first days that Google expanded it to 71 countries, with the UK recently added. The EU is still outside the current rollout, although Google says access there is on the way.
The central signal is simple: demand exceeded what the team expected. Josh Woodward, VP at Google Labs and GeminiApp, said the team spent the entire week keeping systems stable and described the level of interest as, "Way, way, way more demand than we expected!"
That kind of early usage matters because video generation is not a light workload. Every prompt can require substantial compute, and when millions of clips are being produced in a compressed window, interest becomes an infrastructure event as much as a product launch.
Hassabis said the sudden rise in usage pushed Google's infrastructure to the limit. He credited the infrastructure, chip, and SRE teams with preventing "our wonderful TPUs from melting." Similarweb also reported a noticeable "Veo 3" effect on Deepmind's website traffic, suggesting that curiosity around the model reached beyond existing users.
Why the model spread so quickly
One reason Veo 3 gained traction is that users shared examples across social media. The source of that attention was not only video quality, but the model's ability to produce videos with high-quality, matching audio tracks.
That detail changes how people experience AI video. A silent clip can still be impressive, but matched audio makes the output feel more complete. For casual users, it reduces the number of extra steps needed to create something shareable. For filmmakers and creators, it points toward a workflow in which the first version of an idea can arrive with both picture and sound already connected.
The rapid spread also shows how AI video launches can become feedback loops. Users create clips, those clips circulate, more people search for the tool, and the product's own visibility drives more load onto the system. In Veo 3's case, Google appears to have seen that cycle almost immediately.
Watermarks arrive as adoption grows
Google has now added visible watermarks to every video generated with Veo 3, with one notable exception: Ultra users working with the Flow tool can create watermark-free results. Google describes watermarking as an initial step to make AI-generated content clearer to viewers.
The company is also working to broaden access to its SynthID Detector, which is designed for detecting synthetic media. Together, visible marks and detection tools are meant to address the same problem from two directions: helping people recognize AI output when they see it, and giving systems a way to identify synthetic media after it has moved through the web.
The timing raises questions. The source article notes that it seems odd Google did not include watermarks from the start, given that the company understood Veo 3 could generate highly convincing fake videos. The exception for Ultra users working in Flow adds another concern because it means watermarking is not universal across paid access levels.
This pattern is not unique to Google. OpenAI's Sora video model follows a similar approach, allowing users to pay for unwatermarked content. That broader comparison matters because it suggests the AI video market is still balancing product appeal, creator flexibility, and responsible deployment.
From web-only to mobile and production tools
Veo 3 began as a web-only model, but Google has already expanded access for Pro and Ultra members through the Gemini app for Android and iOS. That move brings AI video generation into a more everyday surface, where users can try prompts from a phone instead of relying only on a desktop workflow.
Filmmakers also have access to Flow, a dedicated tool built around specialized workflows for AI-powered video production. That distinction matters because casual prompting and production work are not the same use case. A social media user may want a quick clip, while a filmmaker may need a more controlled process for developing scenes and iterations.
Google is also preparing additional features for Veo 3. Woodward said development is underway on image-to-video generation within the Gemini app, more consistent audio, faster output, and support for Google Workspace accounts.
Those planned updates point to the product's next priorities. Image-to-video would give users another way to guide the model. More consistent audio and faster output would improve reliability. Google Workspace support would broaden access into accounts used for work, although the source does not specify timing or exact availability.
The larger signal from Veo 3
Veo 3's launch shows that demand for AI video can arrive faster than expected, even for a company with Google's infrastructure. The early numbers are broad rather than precise, but "millions" of generated videos in a few days is enough to show that users are not treating AI video as a niche experiment.
At the same time, the rollout highlights unresolved questions. Availability is still uneven, with the model expanded to 71 countries while remaining unavailable in the EU. Watermarking is now present for most outputs, but not every output. Detection tools are being broadened, but the source does not say how widely they are available today.
The practical takeaway is that Veo 3 is no longer only a model demonstration. It is now a live consumer and creator product under real demand, with Google working to keep systems stable, expand access, improve features, and define how synthetic video should be labeled as it spreads.