The rising global demand for concrete poses a significant challenge to reducing carbon emissions. Recycling waste glass offers a sustainable alternative by reducing landfill burden and conserving resources. However, conventional mix design methods are inefficient when recycled materials are involved, and many existing machine learning approaches overlook the materials genome and lack experimental validation. This study in-troduces an inverse design methodology using a Conditional Invertible Neural Network to generate concrete mixtures containing waste glass that meet target compressive strengths. By integrating physical and chemical properties of raw materials into the generative model, the proposed approach enables efficient and accurate mixture design. Experimental validation shows 93.5% accuracy for a 55 MPa target strength within one minute. This method can reduce carbon emissions by up to 92.4% through the recycling of global waste glass. This scalable, cost-effective strategy supports the development of high-performance, low-carbon concrete aligned with broader circular economy goals.