Background. Antibiotic heteroresistance (HR), a resistant subpopulation within a clonal isolate reported susceptible by routine testing, can drive unexplained treatment failure and on-treatment emergence of resistance, yet prospective prevalence data for first-line beta-lactams in Gram-negative bloodstream infection are lacking. We assessed the prevalence and genomic basis of HR to piperacillin, piperacillin/tazobactam, or meropenem among bloodstream isolates reported susceptible by standard-of-care testing. Methods. In this prospective two-centre prevalence study we screened consecutive monomicrobial Gram-negative positive blood cultures at the University Medical Center Hamburg-Eppendorf (UKE), Hamburg, Germany, and the Fondazione Policlinico Universitario A. Gemelli IRCCS (FPG), Rome, Italy. Enterobacterales and Pseudomonadales reported susceptible by standard-of-care AST were allocated to one investigated beta-lactam (piperacillin, piperacillin/tazobactam or meropenem) according to their routine susceptibility phenotype (Vitek2), and entered a two-tier phenotypic workflow: an agar filter at 0.5x the EUCAST breakpoint in triplicate, followed - if positive - by population analysis profiling (PAP) up to 4x the breakpoint. Heteroresistance was defined as a clonal resistant subpopulation at a frequency of at least 0.5x10-6 surviving in the 4x breakpoint zone. Prevalence was estimated with Wilson-score 95% CIs and compared between centres with Fisher's exact test. Confirmed heteroresistant isolates and their paired susceptible bulk populations underwent short- and long-read whole-genome sequencing, and each resistant/susceptible pair was tested for tandem gene amplification, differential subclonal point variants and structural variation. Findings. 279 isolates were screened (93 at UKE, 186 at Gemelli). The 0.5x breakpoint filter was positive in 22 of 93 isolates (23.7%) at UKE and 66 of 186 (35.5%) at Gemelli. PAP confirmed heteroresistance in 6 of 93 isolates at UKE (6.5%, 95% CI 3.0-13.4) and 27 of 186 at Gemelli (14.5%, 10.2-20.3), a pooled prevalence of 33 of 279 (11.8%, 8.5-16.1); the between-centre difference did not reach significance (Fisher's exact test p=0.051). Heteroresistance was essentially confined to piperacillin/tazobactam: 32 of 150 isolates screened against piperacillin/tazobactam were heteroresistant (21.3%, 15.5-28.6), versus 1 of 71 for piperacillin (1.4%) and 0 of 58 for meropenem. Escherichia coli accounted for 17 of the 33 heteroresistant isolates. In the resistant/susceptible pairs sequenced to date a candidate mechanism was resolved in 18: 13 carried a tandem amplification of a beta-lactamase locus (copy-number ratio 2.4-33.0; amplified units 5-136 kb, most often plasmid-borne and centred on blaTEM-1, blaCTX-M or blaOXA-1), and 5 carried point variants in the AmpC circuit (ampD, ampR, dacB/PBP4) or in efflux components, three of them fixed loss- or gain-of-function changes and three subclonal. Interpretation. Roughly one in eight Gram-negative bloodstream isolates called susceptible to a first-line beta-lactam harboured a resistant subpopulation, and the burden fell almost entirely on piperacillin/tazobactam - the agent most often used empirically in this setting. The dominant genomic substrate was unstable amplification of an existing beta-lactamase gene rather than acquisition of a new resistance determinant, which explains why the phenotype is reversible, invisible to AST and, in principle, quantifiable by copy-number assays at the point of care.