Latitude.sh (@latitudesh) / X

Latitude.sh

878 posts

Latitude.sh

@latitudesh

Deploy and manage fully automated bare metal servers globally.

latitude.sh

Joined September 2009

Latitude.sh
@latitudesh
Jun 11
🟢 New on Latitude.sh VMs 𝗥𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 𝗺𝗲𝘁𝗿𝗶𝗰𝘀: CPU, memory, network, disk. Query windows from 5m to 24h. 𝗨𝘀𝗲𝗿 𝗗𝗮𝘁𝗮: attach cloud-config scripts that run on first boot, before you ever SSH in. Check out VM metrics and set up user data →
GIF
114
Latitude.sh
@latitudesh
Jun 3
Big news. Together with @megaportnetwork, we're helping build the next generation of AI infrastructure - A$827M raised - A$459M in new contract TCV - Latitude.sh grew from A$60M ARR to A$385M ARR in just 6 months following the acquisition. Over the last 60 days
681
Latitude.sh
@latitudesh
Jun 1
We just shipped 𝗢𝗯𝗷𝗲𝗰𝘁 𝗦𝘁𝗼𝗿𝗮𝗴𝗲 🪣 S3-compatible, no request, retrieval or egress fees and no minimum terms. Two classes: high-performance for AI workloads, standard for cold storage. Point your existing S3 tooling at it and go! → For locations and pricing:
227
Latitude.sh
@latitudesh
May 28
Replying to @latitudesh
If you want to understand how modern infra teams are running K8s closer to the hardware layer, this one is worth the read. Read the full post: latitude.sh/blog/kubernete…
83
Latitude.sh
@latitudesh
May 28
Replying to @latitudesh
Bare metal Kubernetes makes the most sense for workloads like: → validator infrastructure → inference serving → HFT / real-time APIs → multiplayer backends → platform engineering → compute-heavy batch pipelines Anywhere p99 latency matters more than averages.
112
Latitude.sh
@latitudesh
May 28
Replying to @latitudesh
Why does this matter? Because modern CPU inference is getting very good. For many workloads: → lower cost → simpler ops → no GPU queueing → easier scaling → predictable performance Especially for lightweight agents, embeddings, routing, and internal AI services.
46
Latitude.sh
@latitudesh
May 28
Replying to @latitudesh
Architecture is intentionally simple: → nginx frontend → API backend → Redis → MetalLB exposing services externally Classic 3-tier architecture but now the workload is actual LLM inference instead of hello-world containers.
61
Latitude.sh
@latitudesh
May 28
Replying to @latitudesh
So @parazito deployed something real. Hermes 4 + Qwen 3.6 side by side on a single bare metal node. No GPUs. No external inference APIs. No managed AI platform. Just Kubernetes + CPU inference running directly on hardware. Try here: <github.com/latitudesh/exa…>
72
Latitude.sh
@latitudesh
May 28
Replying to @latitudesh
One of the hardest problems in bare metal K8s is networking. Clouds hide this from you, bare metal doesn’t. Every LKS cluster ships with MetalLB + BGP already configured. Create a LoadBalancer service → get a real external IP automatically announced to the network. No hacks.
63
Latitude.sh
@latitudesh
May 28
Replying to @latitudesh
Under the hood, LKS runs RKE2. Why? → embedded etcd → automated TLS rotation → sane defaults → standard Kubernetes APIs → predictable networking with Calico No proprietary abstractions. kubectl still works exactly the way you expect.
441
Latitude.sh
@latitudesh
May 28
Replying to @latitudesh
Bare metal K8s usually fails before workloads even start, not because of hardware, because cluster ops are painful. etcd TLS bootstrapping network policies load balancing BGP control plane HA LKS automates all of it while staying fully standard Kubernetes.
512
Latitude.sh
@latitudesh
May 28
Most Kubernetes workloads belong on VMs. but if you care about deterministic latency, direct hardware access, predictable networking or running inference close to the metal, virtualization becomes overhead. Our Software Engineer wrote about running Kubernetes on bare metal
1.1K
Latitude.sh
@latitudesh
May 25
what's your storage backbone look like right now? (asking for a reason)
332