Phosphorus (P) transformation in soil is jointly governed by abiotic conditions and microbial functional genes, particularly in semi-arid grassland ecosystems where water and nutrient limitations restrict productivity. However, the integrated effects of P, nitrogen (N), and water addition on soil P fractions and microbial P-cycling genes remain poorly characterized. In this study, we evaluated the responses of inorganic and organic P fractions and the abundance of functional P genes (phoD, gcd, pqqC, ppx, phnK) and their interactions under five treatments implemented over 18 years in a semi-arid grassland in North China. The treatments included: control (CK), P addition (P10), N + P addition (NP), water + P addition (WP10), and water + N + P addition (WNP). Results revealed that NaHCO3-Pi was significantly increased under WNP (17.71 mg kg(-1)), followed by WP10 (16.10 mg kg(-1)), P10 (15.04 mg kg(-1)), NP (11.40 mg kg(-1)) and CK (5.53 mg kg(-1)), while NaOH-Pi was highest under WNP (96.85 mg kg(-1)), followed by WP10, P10, NP and CK, indicating enhanced inorganic P availability with water addition. For organic P, NaHCO3-Po peaked under WNP (46.34 mg kg(-1)), followed by WP10, and P10, whereas NaOH-Po showed a similar trend, with the highest concentration in WNP (64.52 mg kg(-1)), then WP10 (58.01 mg kg(-1)) and P10 (49.75 mg kg(-1)). In contrast, HCl-Pi and residual P were greatest under P10, suggesting that P-only application promotes accumulation in recalcitrant pools. Gene abundance patterns mirrored P availability: both phoD (involved in organic P mineralization) and gcd (responsible for solubilizing inorganic P) were highest under WP10, aligning with increased labile P fractions, while pqqC was downregulated under enriched conditions (WNP < NP < WP10 < P10 < CK). phnK, associated with phosphonate metabolism, peaked in WP10, whereas ppx was maximally expressed in P10. Redundancy analysis (RDA) demonstrated strong alignment of phoD and phnK with labile P forms (NaHCO3-Pi, NaHCO3-Po), while pqqC and ppx correlated with total and stable P pools. Mantel correlations confirmed that WP10 and WNP treatments exhibited the strongest positive associations between microbial gene abundance, bioavailable P fractions, microbial biomass phosphorus, carbon, nitrogen (MBP, MBC, MBN), and trace elements (Zn, Fe, Mn). Collectively, our findings show that combined water and nutrient inputs synergistically enhance labile P pools and stimulate functionally diverse microbial pathways, thereby improving phosphorus turnover in semi-arid grassland soils. This study provides mechanistic insight into the interactions between nutrient management and microbial functional dynamics, offering guidance for sustainable soil fertility practices under dryland conditions.