Vision-Language-Action (VLA) models have recently emerged as powerful generalists for robotic manipulation. However, their reliance on visual modalities limits physical reasoning and precise force regulation in contact-rich tasks. Existing attempts to incorporate vision-based tactile sensing into VLA models typically treat tactile inputs as auxiliary visual tokens, thereby overlooking the underlying correlation between surface deformation and interaction dynamics. To bridge this gap, we propose a shift from tactile-vision alignment to tactile force alignment. We introduce TaF-VLA, a force-aware VLA framework that integrates tactile representations grounded by physical force and pressure supervision into action generation. To support this, we develop an automated tactile-force acquisition device and construct TaF-Dataset, containing over 10 million synchronized tactile observations, 6-axis force/torque measurements, and force maps. To achieve tactile-force alignment, we introduce the Tactile-Force Adapter (TaF-Adapter), which learns force-grounded tactile representations by aligning temporal tactile embeddings with discretized force and pressure codes. This mechanism ensures that the learned representations capture history-dependent, noise-insensitive physical dynamics rather than merely static visual textures. Finally, we integrate the pretrained adapter into a VLA backbone to condition action generation on force-grounded tactile feedback. Real world experiments across contact-rich manipulation tasks show statistically significant improvements over vision-only and tactile vision-aligned baselines. These results demonstrate that explicit tactile-force alignment provides effective physical grounding for VLA policies, enabling more robust force-aware manipulation.