Shadow, a French cloud PC provider, has launched Blade, a platform that lets developers, startups and businesses run AI models and deploy applications in production without buying or managing dedicated hardware. The idea is simple: the same fleet of twelve thousand GPUs that Shadow uses for cloud PCs and gaming, which is busy with players at night, is put to work on professional AI workloads during the quieter residential hours, that is, during the day.
The platform has two services. The first, Blade Models, is an online interface from which users select a language model: it offers the main open-weight models for text, images and other uses. The second, Blade AI Fabric, is aimed at those who build and deploy applications. The goal is inference, the execution of models in production, more sovereign and with more predictable prices than those of the large U.S. hyperscalers.
Image: nztoys.vn — https://nztoys.vn/products/blade-shadow-acy-studio
The economic point is the reuse of capacity that would otherwise sit idle. Optimising the utilisation rate of the infrastructure, the company says, allows compute costs to be cut by between twenty and thirty percent. Because the capacity depends on how much of the fleet is free, pricing follows real availability.
Why it matters. Inference, the stage where a model answers users, is according to HackwareNews concentrated mostly with large U.S. providers that often offer very high-end GPUs over-engineered for smaller models. Costs and complexity have held many companies back from moving from prototype to production. A European provider offering a simpler alternative closer to the data addresses exactly that obstacle, at a time when Europe is debating how dependent it is on foreign hardware and cloud. AlexTech has already reported that the EU controls less than 10% of the chips and servers for data centres needed to triple AI capacity.
Image: nztoys.vn — https://nztoys.vn/products/blade-shadow-acy-studio
The backdrop is crowded. In Italy, Fastweb and Vodafone are expanding their Nvidia supercomputer to offer compute to businesses and public administrations with data staying in the country, while Mistral has chosen to host third-party models alongside its regional inference. Blade takes a different route: not new data centres to build, but capacity already present in the remote gaming rooms.
Who benefits? The announcement names developers, startups and businesses that want to run models without managing their own hardware. For them the concrete advantage is compute close to their data, with open models they can customise and serve at scale. How much free capacity holds up under professional loads at peak hours, which depends on player traffic, is the test the platform must pass in the coming months.
No comments yet. Be the first!