© 2018 Elsevier B.V. A long-standing conceptual model for deep submarine eruptions is that high hydrostatic pressure hinders degassing and acceleration, and suppresses magma fragmentation. The 2012 submarine rhyolite eruption of Havre volcano in the Kermadec arc provided constraints on critical parameters to quantitatively test these concepts. This eruption produced a >1 km3raft of floating pumice and a 0.1 km3field of giant (>1 m) pumice clasts distributed down-current from the vent. We address the mechanism of creating these clasts using a model for magma ascent in a conduit. We use water ingestion experiments to address why some clasts float and others sink. We show that at the eruption depth of 900 m, the melt retained enough dissolved water, and hence had a low enough viscosity, that strain-rates were too low to cause brittle fragmentation in the conduit, despite mass discharge rates similar to Plinian eruptions on land. There was still, however, enough exsolved vapor at the vent depth to make the magma buoyant relative to seawater. Buoyant magma was thus extruded into the ocean where it rose, quenched, and fragmented to produce clasts up to several meters in diameter. We show that these large clasts would have floated to the sea surface within minutes, where air could enter pore space, and the fate of clasts is then controlled by the ability to trap gas within their pore space. We show that clasts from the raft retain enough gas to remain afloat whereas fragments from giant pumice collected from the seafloor ingest more water and sink. The pumice raft and the giant pumice seafloor deposit were thus produced during a clast-generating effusive submarine eruption, where fragmentation occurred above the vent, and the subsequent fate of clasts was controlled by their ability to ingest water.