The trusted execution layer for AI compute

A new frontier stack for enterprise AI compute.

Covextra brings verifiable governance and GPU mutualisation together, so enterprises can run AI agents and other sensitive workloads on the compute they need, with proof of where and how every workload ran.

  • Verifiable governance

    Workloads run in confidential computing environments, with attested evidence of where and how they ran.

  • GPU mutualisation

    Capacity from independent providers, pooled on one exchange and priced in real time.

  • Orchestration for AI agents

    Every workload placed by policy: jurisdiction, certification, cost, and performance.

Why now

AI compute is bought on trust. It should be bought on evidence.

Enterprises that want to run AI on their most sensitive data have to take a provider's word for where it runs, who can see it and what they are paying for. Providers with capacity to spare have no common market in which to offer it on those terms.

Confidential computing changes what can be proven. Covextra builds the market on top of it: requirements stated precisely, capacity offered against them, and a workload that carries its own evidence.

How it works

From a requirement to a settled contract, with evidence at every step.

  1. 01

    The buyer states the requirement

    Model, throughput, jurisdiction, certifications, confidential computing and service level go into one request. Each constraint is marked as a must or a preference, so nothing is silently dropped.

  2. 02

    Providers answer with binding quotes

    An agent in each provider’s data centre knows its own capacity and energy cost, and answers the requests it can meet with a price it will honour.

  3. 03

    The workload runs where it was promised

    The accepted quote is deployed into a confidential computing environment, which proves by attestation what is running and where.

  4. 04

    The evidence settles the contract

    Payment is held until the instance is attested as specified, and usage is metered against the quote. The record that settles the contract is the one both sides can check.

The stack

One layer between the buyer and the hardware.

Covextra sits above the infrastructure and below the workload. It places each job by policy, runs it where it can be proven, and returns the evidence to both sides.

Workloads

AI agents, inference, fine-tuning, sensitive data pipelines

Covextra

Orchestration · exchange · metering · settlement

Confidential computing

Attested environments on CPUs and GPUs

Infrastructure

Independent data centres, sovereign and hybrid clouds

Who it is for

Built for the parties who have to trust each other least.

Enterprise buyers

Run AI agents and sensitive workloads on the compute they need, with proof that the data stayed where the policy says and that no operator could read it.

Compute providers

Fill idle GPU capacity with workloads that fit it, priced against your own costs, and get paid against evidence rather than disputes.

Public sector and defence

Place workloads under jurisdiction and certification constraints that are checked by the hardware, not stated in a contract alone.

Products

What we are building.

In pilot

Covextra Exchange

The order book for AI compute: requests for proposals from buyers, binding quotes from providers, matched on the constraints that matter and settled on attested evidence.

Available to design partners

Covextra Harness

An agent workspace whose agent, model and tools each run in an attested confidential environment, reached from the browser over a channel the operator cannot read.

Launching 2027.

We are working with a small group of buyers and compute providers ahead of launch. If you would like to be one of them, write to us.

contact@covextra.com