We present a novel, data-driven approach for global ionospheric mapping that fuses GNSS-derived vertical total electron content (VTEC) with complementary VLBI observations. Building on a daily multi-layer perceptron framework, we explore two fusion strategies: direct assimilation of station-based VLBI VTEC and end-to-end incorporation of VLBI differential TEC (DTEC). Models are trained with a weighted loss to balance the disparate data volumes and evaluated against independent Jason-3 altimetry measurements over the full year 2023. We find that uncorrected residuals are dominated by a large global bias; after bias removal, the VLBI VTEC fusion reduces median residuals by nearly 50% relative to GNSS-only maps, with most improvements localized in the vicinity of VLBI sites. In contrast, the DTEC-based approach offers minimal gains. Our results demonstrate that assimilating absolute VLBI VTEC improves local mapping accuracy, while its impact on global-scale TEC maps remains modest.