The integration of healthcare algorithms into clinical practice represents a significant transformation. Deployed as clinical decision support tools, these algorithms promise a new era of medical precision and efficiency. However, concerns persist that they may absorb, codify, and even amplify historical biases in medical care. We investigate this trade-off by quantifying the downstream consequences of algorithm adoption. Using a difference-in-differences analysis of 308,137 inpatient admissions for acute coronary syndrome (ACS), we find that algorithm adoption changes care delivery: high-severity patients receive more diagnostic testing and greater access to treatment, while care for low-severity patients is de-escalated. These shifts translate into a significant reduction in in-hospital mortality for high-severity patients, while outcomes for low-severity patients are not adversely affected. The survival benefits, however, are not equitably shared. Within the high-severity group, the mortality reduction accrues exclusively to White patients, while non-White patients experience no significant improvement. Consistent with this disparity, non-White patients are less likely than White patients to receive critical diagnostic tests and intensive treatments after adoption. Our study provides large-scale evidence that technologies designed for efficiency can deepen health inequities, underscoring the critical need to manage algorithms for equitable outcomes.