Previous studies have reported inconsistent associations between platelet count (PLT) and cancer survival. However, whether there is linear causal effect merits in-depth investigations. We conducted a cohort study using the UK Biobank and a two-sample Mendelian randomization (MR) analysis. PLT levels were measured prior to cancer diagnosis. We adopted overall survival (OS) as the primary outcome. Cox models were utilized to estimate the effects of PLTs on survival outcomes at multiple lag times for cancer diagnosis. We employed 34 genetic variants as PLT proxies for MR analysis. Linear and non-linear effects were modeled. Prognostic effects of gene expression harboring the instrumental variants were also investigated. A total of 65 471 cancer patients were included. We identified a significant association between elevated PLTs (per 100 × 109/L) and inferior OS (HR: 1.07; 95% CI: 1.04-1.10; p < .001). Similar significant associations were observed for several cancer types. We further observed a U-shaped relationship between PLTs and cancer survival (p < .001). Our MR analysis found null evidence to support a causal association between PLTs and overall cancer survival (HR: 1.000; 95% CI: 0.998-1.001; p = .678), although non-linear MR analysis unveiled a potential greater detrimental effect at lower PLT range. Expression of eleven PLT-related genes were associated with cancer survival. Early detection of escalated PLTs indicated possible occult cancer development and inferior subsequent survival outcomes. The observed associations could potentially be non-linear. However, PLT is less likely to be a promising therapeutic target.