AI Colocation in São Paulo
Home for your Artificial Intelligence workstation.
Bring your upright chassis or rack and get power and cooling sized for multiple ultra-high-performance GPUs.
Cancel whenever you want.
Choose the ideal plan for your AI machine
All plans support Workstation (Tower) chassis or rackmount servers and include dedicated power. A tower PC/chassis is not required - rackmount is fully accepted.
AI Builder
AI Next-Gen
AI Enterprise
AI Cluster
CustomNote: Servers may have power peaks above the contracted amount (power burst), as long as they do not exceed 5% of the time in the monthly aggregate.
Hardware suggestions by plan
AI Builder (1000W)
Entry plan for inference and fine-tuning of light models.
Ideal for high-performance machines running one top-tier GPU or two previous-generation GPUs at full load without overloading the power circuit.
Recommended PSU: Recommended for 850W to 1000W PSUs.
Configuration examples:
- 1x NVIDIA RTX 4090 (TDP ~450W) or 1x RTX 5080.
- 2x NVIDIA RTX 3090 or 2x RTX 4080.
- 1x to 2x RTX 4000 Ada Generation.
- AMD Ryzen 9 (7900X/9900X series) or Intel Core i9.
- Common use cases: Local LLM inference (such as Llama 3 8B), intensive image generation (local Stable Diffusion/Midjourney) and Blender/Premiere rendering.
AI Next-Gen (1500W)
The ideal choice for the new GPU generation and parallelism.
Designed specifically to handle the high draw of new flagship cards or dual-GPU setups.
Recommended PSU: Recommended for 1200W to 1600W PSUs.
Configuration examples:
- 1x NVIDIA RTX 5090 (projected TDP around 600W).
- 2x NVIDIA RTX 4090 running together at 100%.
- 2x NVIDIA RTX 6000 Ada.
- AMD Ryzen Threadripper (7000) or dual Intel Xeon.
- Common use cases: Continuous computer-vision model training, fine-tuning larger LLMs (70B with quantization) and batch data processing.
AI Enterprise (2500W)
Infrastructure for bare-metal clusters and enterprise data science.
This plan has high power capacity. It targets large chassis (Full-Tower or E-ATX) that hold multiple cards and need cooling no office can provide.
Recommended PSU: Recommended for extreme 1600W/2000W PSUs or redundant supplies (dual 1200W).
Configuration examples:
- 1x or 2x NVIDIA H100 (PCIe), A100 or L40S.
- 4x NVIDIA RTX 4090 (liquid cooling or blower).
- 4x NVIDIA RTX 6000 Ada Generation.
- AMD EPYC (dual socket) or Threadripper PRO with 512GB+ RAM.
- Common use cases: Training LLMs from scratch, 24/7 enterprise AI solutions with zero tolerance for failure, and complex scientific simulations.
Location and infrastructure in São Paulo
Our main high-density infrastructure is in the Tier III datacenter in Osasco, built for extreme connectivity and ultra-low latency with direct peering at IX.br.
206.0.29.1 to measure latency directly to our network.
The definitive home for your AI infrastructure
Forget traditional datacenters that cannot handle the draw of new GPU generations. Our environment is optimized for high thermal dissipation and high power consumption, ideal for setups with multiple RTX 4090, A100 or H100 cards.
- Optimized spaceWe accept tower-format chassis or rackmount servers (a tower PC/chassis is not required). We place your equipment safely in the rack.
- Guaranteed powerCircuits sized for peak load during Deep Learning training.
- Extreme connectivityPorts up to 10Gbps dedicated for fast dataset transfers.
Frequently asked questions (FAQ)
See the most common questions about our AI Colocation.
What is AI Colocation?
AI colocation is the hosting model where you bring your own AI PC, workstation or GPU server (tower or rack chassis) into a specialized high-density datacenter. A tower PC is not required - standard rackmount servers (1U, 2U, 4U and larger) are fully supported and welcome. Instead of running high-draw, high-heat hardware at home or in an office, you place the equipment in professional infrastructure optimized for GPUs such as RTX 4090, 5090, A100 or H100 - paying only for space, power and high-capacity network.
Challenges of running AI PCs and tower chassis in ordinary spaces
Explosive AI growth demands a high-performance AI PC with robust GPUs. Running a tower PC or tower chassis of that power in homes or typical offices brings critical issues: extreme heat overwhelms ordinary AC, creates excessive noise and risks hardware integrity through overheating and thermal throttling.
HypeHost high-density solution
HypeHost in São Paulo offers the definitive colocation solution for moving your AI PC out of the workplace into a High Density environment. Our infrastructure safely houses any tower chassis (Mid-Tower or Full-Tower) or rackmount servers. Bringing a tower chassis is not required - rackmount is fully supported. Uninterrupted power and precision cooling are guaranteed. Airflow is tightly controlled in contained aisles, forcing cold air through your workstation, eliminating thermal throttling and ensuring maximum 24/7 performance.
Support for tower chassis, workstations and rackmount AI servers
Besides custom gamer or workstation tower builds, we specialize in hosting rackmount servers and enterprise AI equipment. A tower PC is not required - standard rackmount servers are fully accepted. Our facility is ready for high-density models such as Dell PowerEdge T640, the new PowerEdge T740, Supermicro SuperWorkstation (e.g. 741GE-TNRT), HP Z8 Fury G5 and Lenovo ThinkStation P8, plus traditional rack servers. These machines are powerful but noisy and power-hungry - our Tier III datacenter is the perfect place to run them.
10Gbps connectivity and long-term savings
Connectivity is another pillar of the service. Hosting your tower PC, workstation or rackmount server with us includes 10Gbps ports for data traffic - essential for neural-network training and large datasets. Compared with the prohibitive cost of public-cloud GPU instances, keeping your own hardware in colocation is a huge long-term saving, with the benefit of owning a company asset.
No lock-in and upgrade flexibility
For full agility in Deep Learning projects, we offer a 100% no lock-in model. Keep your AI PC, workstation or rackmount server in our infrastructure as long as you need, free to scale or do physical upgrades via Smart Hands. Whether for heavy 3D rendering, LLM training or real-time inference, bring your hardware (tower or rackmount) to HypeHost.
Payment methods
We accept credit cards (Visa, Mastercard, Elo, Amex), Pix with instant activation, PayPal, bank slip and bank transfer.
SWIFT subject to prior review. Contact us for details.
Visa
Mastercard
Elo
Amex
Pix
PayPal
Banco do Brasil