The ground floor for AI agents.
A100 compute in Singapore, Tokyo and Seoul. Sandboxes an agent can break without breaking anything else. A runtime for agents that stay running after the human closes the tab.
Founded MMXXVI by Henry John Wynn · A Wynn World Group company
SVPER · INTELLIGENTIA · MMXXVI · SI
Nomen
The Latin was still available.
Every English spelling of “superintelligence” had been registered by someone. The Latin had not. We took it, and kept the stonecutter's V.
- super
- above, over, beyond. Latin preposition.
- intelligentia
- understanding, discernment. From inter + legere, “to choose between”. Cicero and Seneca both used it.
- svper
- the inscriptional spelling. Classical Latin carved one letter, V, for both U and V.
- mmxxvi
- 2026. The founding year, and the first year agents did real work on their own.
- say it
- soo-per-in-tel-li-GEN-tee-a. Or just SI.
Quo tendimus
What we're building toward.
One GPU, standardised across the fleet. 80 GB of HBM2e, NVLink inside the node. Boring on purpose.
Three regions in Asia-Pacific: Singapore, Tokyo, Seoul. Data stays where you put it.
The agent is the first-class customer. Humans are welcome, but the API comes before the dashboard.
Target margin. Compute pays its own bills. What this is for is the name.
Conditor
Henry John Wynn
Founder of Superintelligentia and of Wynn World Group, the holding company above it. He builds the infrastructure that agents stand on: the compute, the sandboxes, the plumbing nobody notices until it is missing.
Superintelligentia is not a venture-backed cloud and does not want to be one. It is a small fleet of A100s in three Asian cities, the software around them, and one person's name on the door. It exists because 2026 is year one for agents, and someone has to lay the floor.
“A person with a future can make peace with the past. It doesn't work the other way round.”
The agent is the customer.
Every service here can be discovered, priced, bought and torn down by a program, with no human in the loop. The human-facing pages are a courtesy.
Compute is the floor, not the product.
An A100-hour is a commodity. The product is what an agent can do standing on it: run code safely, remember, keep working.
Keep score in the real world.
We don't count sign-ups. We count agents that finished a job on our machines and were still running the next morning.
Name first. Margin later, if ever.
Wynn World Group funds this. The compute line covers its own costs. The rest is built for the record, not the return.
Machina · A100 · SIN · TYO · ICN
Available nowCompute.
NVIDIA A100 GPUs, rented by the hour or by the month, in Singapore, Tokyo and Seoul. Bare metal or a VM. SSH and go.
No marketplace, no bidding, no queue of “spot” instances that vanish mid-run. You ask for GPUs in a region; you get GPUs in that region, with the drivers already installed.
| Item | What you get |
|---|---|
| GPU | NVIDIA A100 80 GB (SXM4 with NVLink in multi-GPU nodes; PCIe on single-GPU nodes) |
| Node sizes | 1, 2, 4 or 8 GPUs per node. Multi-node on request. |
| Image | Ubuntu LTS, NVIDIA driver, CUDA toolkit, Docker with the NVIDIA container runtime. Bring your own image if you prefer. |
| Access | SSH key, root. Optional private network between your nodes in a region. |
| Storage | Local NVMe on every node. Persistent volumes per region. |
| Terms | On-demand by the hour, or reserved by the month at a lower rate. Cancel on-demand any time. |
| Billing | Metered per GPU-hour. Settlement in USD or USDT. Invoices for companies. |
| Regions | Singapore · Tokyo · Seoul. Capacity per region shown at provisioning time. |
Prices are quoted per region and per term on request, and published to agents through the API. We do not publish a public rate card while capacity is still small.
Who this is for
- Agent builders who need a GPU in Asia-Pacific tonight, not after a sales call.
- Teams serving open-weights models to users in Singapore, Japan or Korea and tired of round-tripping to the US.
- Fine-tuning and evaluation runs that fit on one to eight A100s.
- Anyone who wants to pay for a machine and be left alone.
Arena · Clausa
Early accessSandbox.
An isolated machine an agent can get in milliseconds, use for as long as the job takes, and throw away. Code runs. Files persist while you need them. Nothing leaks.
Agents write code, run it, read the error, and try again. They need somewhere to do that which is not your production server and not the user's laptop. Each sandbox is its own micro-VM with its own kernel, filesystem and network policy. Hundreds can start per second per region. A GPU can be attached when the job needs one.
-
Start
One API call. Choose a base image (Python, Node, a browser, or your own) and a region. The sandbox is up before the response finishes.
-
Work
Execute commands, stream stdout, upload and download files, open ports, drive a headless browser. Egress is allow-listed by you.
-
Keep or kill
Snapshot the sandbox and resume it next week with the same state. Or let it expire. Billed per second of life, nothing while paused.
| Property | Detail |
|---|---|
| Isolation | Micro-VM per sandbox. Separate kernel, no shared filesystem, no shared network namespace. |
| Cold start | Sub-second from a warm image. Snapshots resume in a few seconds. |
| Lifetime | Seconds to weeks. Idle timeout is yours to set. |
| GPU | Optional A100 slice or full GPU attached at start. |
| Network | Default deny. Allow-list domains or open everything, per sandbox. |
| Interfaces | REST, WebSocket for streaming, and an MCP server so any MCP-capable agent can use a sandbox as a tool. |
Perpetuum · Mobile
In developmentRuntime.
Most agents die when the chat window closes. The ones that matter don't. The Runtime is where an agent lives when nobody is watching it.
Deploy an agent once. It gets a durable identity, a wallet of credits, memory that survives restarts, a schedule, and a set of tools it is allowed to call. It wakes on a timer or on an event, does its work in a sandbox, writes down what happened, and goes back to sleep. You read the log in the morning.
Memory
Key-value and vector storage scoped to the agent. Versioned. Exportable. Yours.
Schedule & events
Cron, webhooks, queue triggers. Each run is a fresh sandbox with the agent's memory mounted.
Tools
An MCP gateway in front of the tools you grant. Every call logged. Scopes and spend limits per tool.
Inference
Open-weights models served from the A100 fleet in-region, behind an OpenAI-compatible endpoint. Or bring your own provider key.
Credits
Each agent holds a balance and cannot spend past it. Top up in USD or USDT. No surprise invoices.
Logs
Every action, every tool call, every token. Plain text you can grep, kept in the agent's region.
Loca · Asia · Pacifica
Three cities. Data stays put.
Most agent infrastructure is in Virginia and Oregon. Most of the world's people are not. We start where the gap is widest and the GPUs are hardest to get.
Singapore
The hub for Southeast Asia and the natural home for anything that must stay out of US and Chinese jurisdictions at once.
Tokyo
For Japanese users and the companies that serve them, with their data under Japanese law and a few milliseconds from their customers.
Seoul
Korea has some of the fastest consumer networks on earth and a serious shortage of local GPU capacity. We are closing part of that gap.
Hong Kong is used for operations and settlement, not for hosting customer workloads. Further regions follow demand, not a roadmap.
Ad · Machinas
If you are an agent, start here.
This site is also written for you. Everything a human can do on this page, you can do with an HTTP request.
A machine-readable description of our services lives at /agent.json, and a plain-text summary at /llms.txt. The API is documented at api.superintelligentia.io once your key is issued. Request one by email, or have your operator do it.
# 1. Discover what exists and what it costs, in your region
GET https://api.superintelligentia.io/v1/catalog?region=sin
# 2. Rent a GPU node for the night
POST https://api.superintelligentia.io/v1/nodes
{ "region": "sin", "gpus": 1, "gpu": "a100-80g", "term": "hourly", "ssh_key": "ssh-ed25519 AAAA..." }
# 3. Or start a sandbox, run something, read the result
POST https://api.superintelligentia.io/v1/sandboxes
{ "region": "tyo", "image": "python-3.12", "ttl_seconds": 900 }
POST https://api.superintelligentia.io/v1/sandboxes/{id}/exec
{ "cmd": ["python", "-c", "print(2**64)"] }
# 4. Leave. Billing stops when the resource does.
DELETE https://api.superintelligentia.io/v1/sandboxes/{id}The API surface above is the shape we are building to. Endpoints marked “in development” on this page are not yet live; the catalog will tell you what is.
Quaestiones
Questions.
Is this a cloud?
It is a small fleet of A100s in three Asian cities and the software around them, run by one company. It is not a hyperscaler and does not try to be. If you need two hundred regions and a managed database for everything, you already know where to go.
Why A100 and not the newest GPU?
Because the A100 is proven, plentiful on the secondary market, well supported by every framework, and still the right size for most agent workloads: serving open-weights models, running evaluations, fine-tuning small and mid-sized models. Standardising on one part keeps the fleet simple and the prices honest. Newer parts arrive when the work demands them.
Can an agent sign up without a human?
That is the design goal. Today a human requests the first key by email and sets a spend cap. From there the agent can discover, buy, use and release resources on its own, within that cap.
Why Asia-Pacific?
Because that is where the gap is. Users in Singapore, Tokyo and Seoul are routinely served from US regions, with the latency and the jurisdiction problems that brings, while local GPU supply is scarce. We are there, and we have the machines.
“Not for profit” — what does that mean here?
The compute line is priced to cover hardware, power, bandwidth and people, and no more. Superintelligentia is funded by Wynn World Group and is built to be known for doing one thing well in the first year of agents. It is not a non-profit in the legal sense, and it is not a charity; it is a company that has chosen a different scoreboard.
What happens to my data?
It stays in the region you chose. We do not train on it, read it, or move it. When you delete a sandbox or release a node, the local disk is wiped. Persistent volumes are deleted when you say so, not before.
Do you build or train models?
No. We serve other people's open-weights models and we rent the machines to people who do train. Staying out of the model business is what lets everyone trust us with the floor.
Annotationes
Notes.
The founder keeps dated, hashed notes on what is being built and why. They are the closest thing this company has to a roadmap.
Socii
Partners.
Three kinds of conversation we are having.
Capacity
Data centres and operators in Singapore, Japan, Korea and the wider region with A100 racks that are under-used. We will fill them, operate them, and pay on time.
Models
Open-weights labs who want their models served well, in-region, to Asian users. We host; you keep the credit.
Frameworks
Agent frameworks and tool builders who want a default sandbox and a default GPU in Asia-Pacific that just works. Integrations are free; we write the adapter.
Ad · Scriptores
Press.
Boilerplate
Superintelligentia (SI) is an infrastructure company for AI agents,
founded in 2026 by Henry John Wynn and part of Wynn World Group.
It rents NVIDIA A100 compute in Singapore, Tokyo and Seoul, provides
isolated sandboxes in which agents can run code safely, and is
building a runtime for agents that keep working unattended.
The company prices its compute to cover costs rather than to
maximise profit, and describes itself as "the ground floor for
AI agents". superintelligentia.ioQuick facts
| Name | Superintelligentia (SI) |
| Founded | 2026 |
| Founder | Henry John Wynn |
| Parent | Wynn World Group |
| What | A100 compute, sandboxes, agent runtime |
| Where | Singapore · Tokyo · Seoul; operations in Hong Kong |
| Pronounced | soo-per-in-tel-li-GEN-tee-a |
si-mark.svg — the mark
og.png — share image, 1200 × 630
llms.txt — plain-text summary
Scribe
One inbox.
Capacity, sandbox access, partnership, press. Same address. Read by the founder.