Introduction: Fasting plasma ketone bodies (KB) are elevated in individuals with type 2 diabetes (T2D) and could affect glycemic control and disease progression. Prolonged KB exposure may result in adaptive beneficial responses, counteracting glycemic dysregulation. In the current proof-of-concept study in adults with T2D, we hypothesized that fasting plasma KB are cross-sectionally associated with poorer glycemic control but prospectively with better glycemic control. Materials and Methods: Fasting plasma KB were measured via nuclear magnetic resonance spectroscopy in patients with T2D treated in primary care (Zodiac cohort; The Netherlands). We analyzed the associations between KB and HbA1c at baseline using linear regression analyses and HbA1c changes over time using linear mixed models. We adjusted for potential confounders, including risk factors for poor glycemic control. Individuals with T2D participating in the general population-based PREVEND study were used as a replication cohort. Results: We included 271 individuals with T2D with a total of 859 HbA1c measurements during a follow-up period of 3.0 (2.0–3.2) years. At baseline, the total amount of fasting plasma KB was independently and positively associated with HbA1c levels (regression coefficient in the fully adjusted analysis = 0.31; 95% CI 0.06–0.57, per doubling of KB; p = 0.02). In contrast, in the longitudinal analyses, fasting plasma KB were associated with a yearly HbA1c (%) decrease of −0.10 (95% CI −0.19 to −0.00 per doubling baseline KB; p = 0.05). Results were replicated in 387 individuals with T2D from a general population cohort with a total of 1115 glucose measurements during a follow-up period of 7.5 (7.2–8.0) years. A yearly decrease in fasting plasma glucose (mmol/L) of 0.09 was found per doubling of baseline KB. Conclusions: This study is the first to suggest a paradoxical role of circulating KB on glycemic control in T2D: elevated KB are associated with cross-sectionally poorer glycemic control but longitudinally with better long-term glycemic control.