Advancements in single cell genomics have led to improved understanding of cellular plasticity and evolution in gliomas over the course of the disease. However, the reorganization of the tumor architecture and cellular interactions within the tumor microenvironment remains elusive. To address this knowledge gap, we employed spatially resolved multi-omic technologies on glioma specimens from various treatment modalities and timepoints. Our efforts have resulted in a map of longitudinal architectural reformation in glioblastoma, which offers insight into the molecular mechanisms responsible for therapy resistance. By using a combination of unbiased array-based (Visium) and single-cell (in-situ sequencing) transcriptomics on de-novo and recurrent glioblastomas under various treatment modalities, including radio-chemotherapy (STUPP +/- Tumor Treating Fields), immunotherapy, or target therapy, we have uncovered shared and unique key transcriptional programs driving therapy resistance. Post-standard treatment, only minor shared modifications of the transcriptional architecture was found after adjusting for confounders such as sample collection region and histology. However, significant inter-patient heterogeneity was noted, including many unique transcriptional programs observed in recurrent samples. Exploration of spatial genomic and metabolic alterations identified hypoxic-driven adaptations in distinct niches as a source for the development of resistant subclones. After immunotherapy, we found significant re-education of the microenvironment, leading to anti-inflammatory responses from myeloid cells and the accumulation of regulatory T cells, resulting in CD8 dysfunction. By integrating spatial transcriptomics, imaging, and clinical parameters into graph-based convolutional neural networks with contextual learning strategies, we identified recurrent patterns that can forecast transcriptional responses to treatment, derived from the de-novo tumor architecture. The spatial perspective on glioma evolution through descriptive and predictive models will enhance accuracy of the decision-making process in precision oncology in the future.