Google has lifted the curtain on a Vids feature that lets anyone create business videos starring a digital avatar of themselves. Just a prompt and a reference image are enough, and the system — powered by Gemini Omni — generates realistic clips in which the user appears in first person. At first glance, the announcement seems like another step in the race for generative AI at work. But a closer look reveals a structural shift: advanced video personalization now runs exclusively in the cloud, pushing local data control further out of reach.

For organizations bound by digital sovereignty rules — public bodies, banks, regulated industries — this feature reinforces a familiar reality: the most advanced AI capabilities remain the domain of hyperscalers. Gemini Omni demands computing power and memory resources that are hard to replicate on-premise without massive investments. As a result, anyone evaluating an on-premise deployment for productivity tools faces a dilemma: accept a functional gap around personalized video generation, or surrender significant biometric and identity data to the cloud.

Google embedding personalized avatars into an enterprise product like Vids also carries strategic weight: it forces competitors (from Microsoft to specialized startups) to reply or differentiate, while widening the moat between cloud solutions and those that could run on private infrastructure. We can expect growing regulatory pressure in the coming years, especially in Europe, over how such technologies handle digital identity. Google may anticipate this by offering data residency guarantees, but the very nature of training a personal avatar means remote servers remain unavoidable.

For the on-premise hardware landscape, Google’s silence on concrete computational costs is no surprise. Yet the implicit message is that multimodal video generation will stay a centralized computing privilege for the foreseeable future. This could fuel interest in hybrid architectures, where some inference tasks on smaller models happen locally, while high-end generative features stay in the cloud. For those who want everything in-house, the road is still long.

The Vids AI avatar isn’t just a gadget — it’s a thermometer of growing dependence on big provider infrastructure, a signal to read carefully for anyone designing IT architectures that must balance innovation and autonomy.