Genome sequencing of pathogens is becoming ubiquitous in microbiology, and automated diagnostics will soon appear. The genome sequence archives, already growing exponentially, are currently not searchable for arbitrary sequence. Such an ability would unlock this resource for science and could underpin global real-time genomic epidemiology and surveillance. We combine knowledge about bacterial genetic variation with ideas used in web-search, to build a DNA search engine for microbial data that can grow incrementally. We index the complete corpus of bacterial and viral whole genome sequence data as of December 2016 (447,833 genomes, 176 Terabytes), using four orders of magnitude less storage than previous methods, making the global archive for the first time accessible to search, and scaling to millions of genomes. We demonstrate its usefulness with three applications: ultra-fast search for resistance genes MCR1-3, host-range determination for 2827 plasmids, and quantification of the rise of antibiotic resistance prevalence in the archives.