Hematopoiesis is a highly dynamical and stochastic process, challenging our understanding of homeostasis. Clinical studies of leukemia or neutropenic patients revealed that multiple blood cell types fluctuate spontaneously with large yet regular oscillations of their frequencies. Yet the stability of hematopoiesis in healthy individuals remains understudied. Here we report on both cross-sectional and longitudinal studies of dozens of healthy mice, through high-dimensional mass and spectral cytometry, to understand hematopoiesis at homeostasis. We found that all cell types in the bone marrow, blood, and spleen exhibit large variations of frequency (e.g., with coefficients of variation larger than 1). While the frequencies of individual cell type fluctuate, there existed extensive and robust correlations/anti-correlations between cell types, exemplified by the pronounced anti-correlation between blood neutrophils and B cells. Through longitudinal study of the blood content of healthy mice, we found that leukocyte fluctuations are ergodic yet subject to chaotic behaviors characterized by a broad spectrum of characteristic timescales. We then built a minimal mathematical model to capture these dynamical features of hematopoiesis (fluctuations, correlations, and chaos) and explain how the accumulation of B cells (e.g. during lymphoma development) would transition the blood cell dynamics from chaos to oscillations (as observed clinically). Finally, we demonstrated the ubiquity and consistency of the correlated fluctuations in hematopoiesis by comparing mouse cohorts of different genetic backgrounds and ages. To conclude, we discuss how study of hematopoiesis must factor in the newfound chaotic dynamics at homeostasis, towards better modeling the responses to perturbations.