Skip to content
Odasper

Early access · onboarding a few design partners

Same jobs. Same deadlines. Cheaper, cleaner electricity.

Odasper is a multi-cloud API that tells AI teams when and where to run their flexible jobs, training and batch, to pay the cheapest, lowest-carbon electricity without missing deadlines.

See how it works
  • Read-only API
  • Multi-cloud, provider-neutral
  • You keep control of execution

One flexible job. One day. One grid.

Training job · 3 h

  • Electricity price
  • Carbon intensity
  • Deadline · 22:00

Illustrative 24-hour curve. The deltas are computed on this synthetic day, not measured on a customer workload.

01/Why now

The constraint on AI compute is moving from chips to power.

Training and batch inference run on electricity whose price and carbon intensity change every hour and differ between regions. Most schedulers ignore both. They run jobs when they are submitted, wherever capacity happens to be, and pay whatever the grid charges at that moment.

~945TWh

Projected data center electricity demand by 2030

The IEA projects data center electricity demand roughly doubling to about 945 TWh by 2030. Available power, not silicon, has become the practical limit on AI expansion.

IEA, Energy and AI, 2025

One region, one week

Hourly electricity price, illustrative

Wholesale electricity prices and grid carbon intensity swing through the day as wind, solar and demand shift. The cheapest hours are often also the cleanest.

Three regions, the same day

Hourly electricity price, illustrative

At the same hour, prices and carbon intensity differ between countries and cloud regions, especially in Europe. A job that can run in more than one region has a choice most schedulers never make.

Not every job can move. Many can: training runs, evaluations, batch inference, data processing. Anything with a deadline but no fixed start time has slack, and slack has a price.

02/How it works

Three steps. Nothing touches your infrastructure.

Odasper sits beside your scheduler, not inside it. It reads signals and job metadata and returns a recommendation. Execution stays with you.

Where Odasper sits

Price and carbon signals
Your flexible jobs
Odasper
Recommendation
Your scheduler
  1. 01

    Connect the signals and the jobs

    Odasper ingests real-time and forecast electricity price and carbon intensity for the regions you use. You describe your flexible jobs: expected duration, eligible regions, deadline. Read-only, through the API or SDK.

  2. 02

    Get a window and a region for each job

    For every job, Odasper finds the start time and region that minimize electricity cost and carbon within the deadline and your SLA constraints. It returns a plain recommendation with the estimated cost and CO₂ difference.

  3. 03

    Run it and capture the savings

    You apply the recommendation in your own scheduler: Kubernetes, Slurm or cloud batch. Odasper records what was recommended and what ran, and reports the cost and carbon difference for finance and scope 2 reporting.

GET /v1/jobs/train-eval-7b/recommendation200 OK
{  "job": "train-eval-7b",  "duration_hours": 3,  "deadline": "2026-09-25T22:00:00Z",  "as_submitted": {    "start": "2026-09-25T08:00:00Z",    "region": "eu-west-3"  },  "recommended": {    "start": "2026-09-25T12:00:00Z",    "region": "eu-west-3",    "cost_delta": -0.39,    "co2_delta": -0.26  },  "mode": "recommend_only"}
The shape of a recommendation, on the same synthetic day as the chart above. Field names may still change before the beta.

No agents on your nodes. No credentials to your clusters. Odasper never starts, stops or moves a job. In the beta, it recommends and you decide.

03/Outcomes

Lower cost first. Reportable carbon as the bonus.

Lead

Lower cost on flexible compute

Flexible jobs move to the hours and regions where electricity is cheapest, inside their deadlines. The saving is the price gap between when a job would have run and when it actually ran, on the share of your compute that can move.

−39%

electricity cost, on the illustrative day

  • As submitted100%
  • Odasper window61%
  • Applies to training, evaluation, batch inference and data processing
  • Largest where electricity is paid at market price rather than on fixed reserved capacity
  • Measured against the counterfactual: what the job would have cost as submitted
Bonus

Carbon avoided, reportable

The same shift lowers the grid carbon intensity of every kWh consumed. Odasper quantifies the avoided emissions per job and aggregates them into a scope 2 view you can put in a CSRD report, with the method and signal sources stated.

−26%

CO₂, on the illustrative day

  • As submitted100%
  • Odasper window74%
  • Hourly, location-based accounting per job and per period
  • Signal sources and method documented for auditors
  • No offsets, no certificates: less consumption at dirty hours, nothing else

Bars use the same synthetic day as the hero chart. They show the mechanism, not a measurement. How much you save depends on your grid, your regions and how much slack your jobs have. We do not publish a headline percentage. Any figure we show is modeled on historical grid data and labeled as such.

04/Why Odasper

Neutral, multi-cloud, and built around an optimization engine.

Clouds are adding carbon-aware options for their own regions. Signal providers sell raw data. Odasper sits between the two and turns signals into a schedule you can act on across providers.

  • Chooses across clouds and regions

    A cloud's option only moves jobs inside that cloud.

    Cloud-native carbon option
    No
    Signal provider
    No
    Odasper
    Yes

  • Forecasts price and carbon

    Cloud-native carbon option
    Partly
    Signal provider
    Yes
    Odasper
    Yes

  • Schedules jobs under deadlines and SLAs

    Cloud-native carbon option
    Partly
    Signal provider
    No
    Odasper
    Yes

  • Works read-only, outside your infrastructure

    Cloud-native carbon option
    No
    Signal provider
    Yes
    Odasper
    Yes

  • Reports carbon per job for scope 2

    Cloud-native carbon option
    Partly
    Signal provider
    No
    Odasper
    Yes

Category-level comparison with the two adjacent approaches, not with named products.

  • Neutral across clouds

    Odasper is not a cloud, so it has no region to favor. It compares your providers and regions on price and carbon alone. A single provider's carbon-aware option can only ever move a job inside that provider.

  • A real optimization engine

    Forecasting price and carbon, then scheduling many jobs across time and regions under deadlines and SLA constraints, is an optimization problem, not a lookup. That engine, and the usage data it improves with, is the product.

  • Not a wrapper over public signals

    Public price and carbon signals are the input, not the output. Anyone can read a carbon intensity API. The value is in deciding what to do with it for your specific jobs, deadlines and regions.

  • Recommend, not control

    Read-only by design. Odasper never touches your clusters or your cloud accounts. It fits beside whatever you already run and is as easy to remove as to add.

05/Roadmap

Recommendations today. Closed-loop optimization next.

We are deliberately starting read-only. The beta has to prove real savings on real workloads before anything automates.

  1. Beta · now

    Recommend

    Read-only recommendations and reporting.

    • Price and carbon signals, real-time and forecast
    • Per-job recommendation: window, region, estimated cost and CO₂ difference
    • Reporting module for scope 2 and CSRD
    • API and SDK, read-only, no access to your infrastructure
  2. Advanced · roadmap

    Optimize

    Closed-loop scheduling across many jobs, time and regions.

    • Forecasting engine for price and carbon
    • Multi-job scheduling across time and regions under SLA constraints
    • Optional automated execution through Kubernetes, Slurm and cloud batch connectors, only if you turn it on
    • Forecasts and schedules that improve with accumulated usage data

06/Who it's for

Built for the people who run the compute.

Primary

ML infra and platform engineering leads

You run a lot of training and batch work, some of it with slack, and you pay for electricity at market price through your cloud or your own data center. You want the bill down without touching SLAs or adding another agent to the cluster.

Good fit if

  • Large or frequent training and batch workloads
  • Jobs with deadlines but flexible start times
  • Electricity priced at market rates, ideally in Europe
  • Multi-region or multi-cloud footprint
Secondary

Sustainability and ESG leads

You need a scope 2 reduction you can defend in a CSRD report, with a method and data sources an auditor can follow. Odasper gives you avoided emissions per job and per period, computed from the same shift that lowers the bill.

Good fit if

  • Scope 2 reporting under CSRD or similar frameworks
  • Hourly, location-based accounting
  • Documented signal sources and method

Not sure how much of your compute is flexible? That is usually the first thing we work out together.

Early access

Onboarding a few design partners.

We are looking for teams that train a lot, have flexible or batch workloads, and want to see what the price and carbon gap is worth on their own jobs. Leave your work email, or write to us directly, and we will come back to you about early access. A human reads every message.

Request early access

Design partners work directly with the founding team and shape what gets built.