Data valuation has increasingly been recognized as a critical mechanism for determining fair compensation for data contributors in ML tasks. Despite its significance, the game-theoretic aspects of the existing valuation metrics have not been studied. In this paper, we study a data marketplace where sellers incur private heterogeneous costs for sharing their data. Perhaps surprisingly, we discover that existing methods (Data Shapely and Leave-One-Out) are not incentive-compatible. Specifically, they encourage sellers to misreport their true costs, leading to market inefficiencies. To address this, we propose a new pricing rule that we theoretically prove simultaneously satisfies: i) incentive-compatibility, ii) market-efficiency, iii) individual rationality, and iv) budget balancing, while also v) being the lowest possible price under these constraints. Our results underscore the importance of game theoretic considerations while designing data valuation metrics.