Latitude.sh

What Latitude.sh actually is

Latitude.sh rents physical servers with the workflow people expect from cloud VMs: pick a plan, pick a site, and the machine is provisioned in minutes through a dashboard, API, Terraform provider, or CLI. The company started life as Maxihost out of São Paulo, rebranded to Latitude.sh, and in November 2025 became part of Megaport, the Australian network-fabric operator. That ownership change is the most important context for this review: the compute platform now sits on top of a private-network business spanning 1,100+ data centers in 31 countries, and in 2026 the combined company raised roughly $594 million explicitly to build a global AI inference cloud, including 2,048 NVIDIA B300 GPUs promised for on-demand use.

The product splits into three lines. Metal is dedicated bare metal — Gen 4 machines on AMD EPYC and Ryzen with DDR5, NVMe storage, and up to 200 Gbps networking. Metal GPU is dedicated GPU capacity, from single H100 servers up to 8x NVIDIA HGX B300 clusters. Virtual Machines are the newer, cheaper tier for lighter workloads. All three share the same account, API, and billing model: hourly on-demand, or monthly and yearly prepayment at steep discounts.

What sets Latitude.sh apart from the average dedicated-server seller is geography. Its site list covers the US, Europe, Asia-Pacific, and — the real differentiator — Latin America: São Paulo, Mexico City, Buenos Aires, Bogotá, and Santiago alongside the usual Dallas, Ashburn, Miami, Los Angeles, New York, Chicago, London, Frankfurt, Amsterdam, Singapore, Tokyo, and Sydney. If you need physical hardware in South America with cloud-style automation, the shortlist of serious providers is very short, and Latitude.sh is on it. For the full provider map in this segment, see our bare-metal-cloud category.

Pricing: what you really pay

Latitude.sh publishes its price list, which already puts it ahead of most bare-metal competitors. Every plan includes 20 TB of outbound traffic per month and unmetered inbound traffic; the per-plan network lines read "20 TB free out / infinity in" with 10, 25, or 100 Gbps uplinks depending on tier.

On the Metal side, entry pricing sits around $190 per month for a c3.small.x86 — an Intel E-2386G with 6 dedicated cores (12 threads), 32 GB RAM, and 800 GB of NVMe — as listed in the vendor's own price sheet. The Gen 4 line scales from there: m4.metal.small at $296 per month (AMD 4244P, 6 cores, 48 GB RAM, 1,920 GB NVMe) is the entry to DDR5 hardware, and the large configurations run $2,920 to $3,971 per month — an rs4.metal.xlarge with an AMD 9554P, 64 cores, 1,536 GB RAM, and 2 x 100 Gbps links starts at $5.44 per hour, $3,971 per month prepaid monthly.

GPU pricing is transparent too. A single H100 80GB (g3.h100.small, 2 x 3.8 TB NVMe) starts at $1.68 per hour, or $1,230 per month. An 8x RTX PRO 6000 node runs $24 per hour ($17,520 per month), and the flagship 8x B300 cluster — 2,304 GB of VRAM on dual-plane RoCE — starts at $64 per hour, $46,720 per month. The VM tier is where the low end lives: vm.small (4 vCPU, 16 GB RAM, 160 GB local storage) is $0.09 per hour, $69 per month, and vm.medium (8 vCPU, 32 GB) is $0.18 per hour, $128 per month.

The billing math rewards commitment in a big way: Latitude.sh states monthly prepayment is about 50% cheaper than hourly, and yearly about 65% cheaper. So the honest way to read the headline rates is as the price of flexibility. A g3.h100.small run hourly for a month would cost far more than $1,230; the monthly figure is the reserved price. Some plans also cost more in Latin America, Asia, and Australia than the US price, and popular plans go out of stock per-site — on one mid-October 2026 check, the cheapest Metal plan was in stock only in Mexico City among the US-priced sites. Egress beyond the included 20 TB is billed, and separate block storage is $0.10 per GB-month with filesystem storage at $0.05 per GB-month.

Performance & infrastructure

Bare metal here means bare metal: single-tenant physical servers, no hypervisor between your workload and the CPU, with NVMe storage and dedicated network interfaces. The Gen 4 lineup spans 6-core entry machines up to 96-core EPYC boxes with 384 GB to 1.5 TB of RAM, and the network fabric is the post-Megaport selling point — private networking between servers, VXLAN-based segmentation, and interconnection into Megaport's wider fabric for customers who want private links to other clouds or sites. Servers come with managed IPv4 and IPv6 addresses, included DDoS mitigation, and support for Linux or Windows with RAID configuration and user-data scripts at provisioning time.

Provisioning speed is the cloud-like part: the company's headline is sub-5-second deployment, and everything is API- and Terraform-driven, which matters more than it sounds if you are rebuilding fleets rather than ordering one server a quarter. Storage products (block and filesystem) exist for capacity beyond local NVMe, though there is no S3-compatible object store in the current catalog.

The service-level agreement is published in the docs and is worth reading literally. For Bare Metal, the metric is data-center infrastructure — power and network uplink — at 99.5% availability. Virtual machines and GPU virtual machines are covered by 99.5% instance availability. That is a modest number: it measures the infrastructure layer, excludes scheduled maintenance and customer-managed issues, and effectively says "your hardware and its uplink will be there 99.5% of the charging period." Third-party trackers quote higher figures (99.9% or 99.95% in some reviews), and reserved contracts reportedly negotiate 99.9% with service credits of 5-30%, but the published baseline in the vendor's own service schedule is 99.5%. We would negotiate the SLA explicitly rather than trust marketing pages.

Support runs from a free community tier up to premium 24/7 support with 15-minute response SLAs that reviewers price at $1,000 per month and up. For unmanaged bare metal this is the normal shape: you get hardware replacement and network guarantees, and everything above the OS is yours.

Where it falls short

Unmanaged is unmanaged. There is no administration layer, no patching, no managed database, and no backups included with servers — the backup story is bring-your-own onto the storage products or off-platform. If your team wants a control panel that fixes things, this is the wrong category.

The 99.5% published availability target is lower than we expect from infrastructure priced at $190 to $46,720 per month. Hyperscaler compute SLAs run higher, and even within bare metal the number is measured narrowly (power and uplink for Metal). Combined with per-site stock limits — where the plan you want may exist only in a subset of locations, sometimes a single city — the operational reality is that capacity planning is part of your job.

Regional pricing differences are another trap. The US price is the advertised one; Latin America, Asia, and Australia can cost more for the same plan, which is worth knowing before you anchor a budget on the price table. And the company is mid-transformation: the Megaport merger closed in late 2025, the AI-inference pivot and B300 rollout are still landing through 2026, and product direction is moving fast. That is more likely to bring capacity than risk, but long-term contracts should be read against the current entity's terms.

Finally, transparency has limits despite the published list: there is no public changelog of per-site stock, the SLA page mixes Megaport-network and compute metrics in one document, and egress overage pricing per GB is not spelled out on the pricing overview itself.

Who should pick this

We would pick Latitude.sh when you need dedicated hardware — not shared vCPUs — in specific cities, especially in Latin America, with provisioning automated like cloud. It fits latency-sensitive game backends, AI inference serving close to users, ad-tech and trading workloads that want the whole machine, and teams standardizing on Terraform who are done with ticket-based dedicated-server ordering. The single-H100 and 8x GPU nodes are credible options for teams that want reserved GPU capacity with published prices rather than auction markets.

We would not pick it for managed hosting, for workloads that demand a 99.99% contractual SLA, or for organizations without an ops function — the platform gives you hardware and a network, and the operating system above it is your problem. For budget-first general-purpose hosting in Europe, cheaper options exist in our bare-metal-cloud category; for hard-to-reach geographies and automated bare metal, Latitude.sh earns its place on the shortlist.

Key takeaways

• Bare metal, GPU metal, and VMs from one API-driven platform; sub-5-second provisioning, Terraform and CLI support
• Published pricing: Metal from about $190/mo, VMs from $0.09/hr ($69/mo), 1x H100 at $1.68/hr; monthly prepay is ~50% below hourly and yearly ~65% below
• Every plan includes 20 TB outbound traffic per month with unmetered inbound
• Latin American coverage (São Paulo, Mexico City, Buenos Aires, Bogotá, Santiago) is the standout; US, Europe, and APAC sites too
• Published SLA is 99.5% (data-center power/uplink for Metal; instance availability for VMs) — negotiate higher for reserved contracts
• Acquired by Megaport in Nov 2025; $594M raised in 2026 for an AI inference cloud with 2,048 B300 GPUs — backing is real, but the platform is mid-pivot and unmanaged by design

Sources

• https://www.latitude.sh/pricing
• https://latitude.sh/docs/sla
• https://stackrates.ai/hosting/latitudesh
• https://itechguides.com/best/dedicated-server-hosting/latitude-sh-metal
• https://linuxdork.com/hosting/dedicated/latitude-sh
• https://pitchbook.com/profiles/company/490846-60
• https://thecompanycheck.com/company/b/latitudesh/xq45aufxis8jjluhe
• https://cloudgpuprice.com/provider/latitudesh
• https://registry.deploy.report/companies/latitude-sh