Purpose-built modular hardware delivers 2x inference throughput at 30-90% lower cost than traditional inference providers, with capacity coming online starting with 160 locations in the United States and Europe in H2 2026. Read more