Background/Objectives: Tuberculosis (TB) remains a global health challenge due to long treatment durations, poor adherence, and growing drug resistance. Inhalable co-amorphous systems (COAMS) offer a promising strategy for targeted pulmonary delivery of fixed-dose combinations, improving efficacy and reducing systemic side effects. Methods: Our in-house-developed machine learning (ML) tool identified two promising API-API combinations for TB therapy, rifampicin (RIF)–moxifloxacin (MOX) and RIF–ethambutol (ETH). Physiologically based pharmacokinetic (PBPK) modeling was used to estimate therapeutic lung doses of RIF, ETH, and MOX following oral administration. Predicted lung doses were translated into molar ratios, and COAMS of RIF-ETH and RIF-MOX at both model-predicted (1:1) and PBPK-informed ratios were prepared by spray drying and co-milling, followed by comprehensive physicochemical and aerodynamic characterization. Results: RIF-MOX COAMS could be prepared in all molar ratios tested, whereas RIF-ETH failed to result in COAMS for therapeutically relevant molar ratios. Spray drying and ball milling successfully produced stable RIF-MOX formulations, with spray drying showing superior behavior in terms of morphology (narrow particle size distribution; lower Sauter mean diameter), aerosolization performance (fine particle fraction above 74% for RIF and MOX), and dissolution. Conclusions: This study demonstrated that PBPK modeling and ML are useful tools to develop COAMS for pulmonary delivery of active pharmaceutical ingredients (APIs) routinely applied through the oral route. It was also observed that COAMS may be less effective when the therapeutic lung dose ratio significantly deviates from the predicted 1:1 molar ratio. This suggests the need for alternative delivery strategies in such cases.