Now in public launch
A green neocloud.
Run anything. Billed in joules.
Powered by the grid you already trust.
Run anything. Billed in joules.
Powered by the grid you already trust.
Drop-in APIs for everything you already build with. Plus a few you didn't expect on a cloud.
OpenAI-compatible. Routes to the cheapest capable silicon, every call.
OCI containers, Firecracker microVMs, ARM, x86. Set an energy budget. Get alerts.
Wasmtime serverless. Sub-millisecond cold start. Per-invocation joules.
SQL, Cypher, GraphQL, Cassandra, time-series — one engine. Joules per query.
S3-compatible. Zero egress within the mesh. BLAKE3 dedupe.
Video calls. Same room link. Lower-energy encoders. Bring your own minutes.
Point your OpenAI SDK at api.greenjoules.cloud. Same calls, the floor price, and a joule receipt on every response.
Or describe a whole stack in invisible.hcl and ship it with one command.
from openai import OpenAI
client = OpenAI(
base_url="https://api.greenjoules.cloud/v1",
api_key="jc_…"
)
r = client.chat.completions.create(
model="auto",
messages=[{"role":"user","content":"hi"}],
)
# r.energy_joules → 0.31 (X-Energy-Joules header)
workload "site" {
image = "nginx:alpine"
region = "auto" # carbon-aware placement
energy_budget = "10 kJ/day" # hard ceiling
}
route "api.example.com" {
to = workload.site
}
Every other cloud charges for time you reserved — the VM is yours whether your code is running or asleep. We measure the energy your workload actually consumed at the silicon, and you pay for that. Idle is near-zero. Inefficient code shows up on the bill. Efficient code shows up on the bill. The unit of account is the same unit physics charges.
The router picks the cheapest capable silicon every call — across the green-grid nodes of our 101-node mesh — so the floor of what you pay is the floor of what the computation physically required.
Same workload — three different ways to pay for it. The mesh keeps your cost flat with what physics required; the others price the variance.
| Big cloudAWS · GCP · Azure | Inference shopsOpenAI · Anthropic · Bedrock | Joule Cloud | |
|---|---|---|---|
| Unit of account | vCPU-hour, GB-month, per-request | Per-token, per-image, per-minute | Joules consumed |
| Idle workload | Reserved cost continues | n/a — call-based | Near-zero |
| Egress | $0.05–$0.09 / GB out | n/a | Free on the mesh |
| Multi-region | Configure per region, replicate manually | Single region per call | region = "auto", carbon-aware |
| Per-call energy / carbon receipt | Aggregate estimate, monthly | Token count, no energy | Signed receipt, every call |
| SDK migration | Service-by-service rewrite | OpenAI shape (vendor-specific quirks) | Two lines: base_url + api_key |
| EU residency / CSRD evidence | Possible; manual | Limited; varies by vendor | Region-pinned + signed receipts |
| Pricing surprise | Over-provisioning, RIs, egress lock-in | Per-token math at the worst tier | $5 to start; you pay for the joules |
Numbers are illustrative for a typical mid-size mixed workload. Run your own week in parallel; the receipts will tell you.
We don't put logos on the homepage we haven't earned. These are the customer shapes we work with at v1 — once a case study is signed off, it lands in customers.
"The unlock wasn't the cost. It was getting the per-call joules so our pricing model could be defensible to our own customers."
"We picked Joule Cloud because the receipts were auditor-ready out of the box. The bill being lower was a bonus."
If you're running a workload that fits and want ongoing credits in exchange for a public case study, write to [email protected]. We co-author; you keep editorial.
A small hold lets you spin up workloads immediately. After that you pay for the energy your code actually consumed at the silicon — nothing for idle, nothing for over-provisioning, nothing for time you didn't use.
No sales call. No setup wizard. No 14-day trial waiting for a credit card decision. You make an account, you mint a token, you ship.
Sign in with a passkey, put $5 on file, and you're shipping. Top up with cards, USDC, or reward points — whatever you've got.