Latitude.sh

Latitude.sh is a bare-metal and GPU cloud operator with Brazilian roots that pitches itself as a global alternative to the hyperscalers for AI inference and low-latency workloads. Its platform layers API-driven provisioning on top of owned servers in 25 locations across five continents, from São Paulo to Tokyo, and it has spent 2025 and 2026 pivoting hard from generalist bare-metal into what it now calls the Global AI Inference Cloud.

In November 2025 Australian network fabric provider Megaport acquired Latitude.sh for US$150M upfront plus up to US$150M in earn-outs. In 2026 Megaport raised about $594M, much of it earmarked for AI inference build-out on Latitude.sh, including a commitment to bring 2,048 NVIDIA B300 GPUs online for on-demand use by autumn 2026. Public pricing puts the on-demand H100 at $1.68 per hour, with reserved commitments discounted 50 to 65 percent.

Unlike the training-scale neoclouds (CoreWeave, Nebius, Crusoe) chasing frontier lab contracts, Latitude.sh targets the long-tail of inference platforms and developer-facing infra. Named customers cluster around AI infrastructure resellers (Together AI, Baseten, Modal, Prime Intellect, Koyeb) plus Web3 and gaming operators (Magic Eden, Hashgraph, Boosteroid, BTCS), rather than named hyperscaler or frontier-lab anchors.