Partnerships

Put capital and capacity to work.

A compute contract is a commitment to what someone can build next. We work with capital partners, operators, and trading desks to put more capacity into use—and give the next wave of builders more room to grow.

Explore the partnerships
A lighthouse above small boats on turquoise water
Our starting point

Our compute operations

Epilogue contracts for its own pool of compute, running inference and serving startups in its network and independent researchers. We take forward positions to provision that pool and trade capacity through our book.

Community rootsAGI House (opens in a new tab)

Our team includes founding members of AGI House, a community dedicated to advancing the technological frontier. AGI House housemates have founded companies worth over $1 trillion, including OpenAI, Perplexity, and Lovable.

See the Orderbook

Our bids and offers, alongside third-party listings for colocation and larger clusters.

Investors and partnerships
Epilogue is an official Cisco partner and compute partner for Deloitte, backed by a multibillion-dollar investment firm.
A named counterparty
When we buy for our book, Epilogue enters the contract and takes the payment obligation.
An operating use for capacity
A shared pool for inference, startups, and independent research, with onward use agreed with the supplier.
Terms you can examine
GPU configuration, location, delivery window, payment schedule, and usage rights, set out for each position.

Writing & talks

Multicoin CapitalOctober 8, 2026

Compute Capital Markets

Shayon Sengupta’s essay on physical delivery, principal desks, and compute financing credits discussions with Epilogue cofounder Kyle Morris.

Read the essay (opens in a new tab)
AI Infra Summit 5May 1, 2026

The phone call economy

Behind the AI buildout is a compute market still run by phone calls and private negotiations. Kyle Morris explains what’s broken, why large GPU deals are so hard to close, and what needs to change.

Watch the talk (opens in a new tab)
For lenders, funds & capital providers

Capital partners

Compute needs capital before it earns revenue. Buyers face deposits; operators need commitments; lenders need to understand the capacity and cash flows behind the paper.

We’re developing financing structures with partners around our own positions and transactions originated by others. Bring an existing compute contract, a lending mandate, or a project seeking offtake.

Prepayment & contract financing

Fund an upfront deposit or a schedule of compute payments against an identified contract.

What the structure needs to answer: Who owes what, when funds are released, when service starts, and how the lender is repaid.

Operator & buildout finance

Pair a capacity commitment with funding for hardware or a deployment. We can assess an offtake already in place or discuss a purchase for our book.

What the structure needs to answer: Delivery milestones, contracted revenue, hardware and capacity rights, and the plan if commissioning slips.

Buyer working capital

Explore credit for teams whose compute bill arrives before their customer revenue.

What the structure needs to answer: Usage, customer receipts, payment timing, credit support, and the size of the commitment.

Revenue forwards

Develop contracts that fix a price for future compute revenue, giving an operator more visibility over what its capacity will earn.

What the structure needs to answer: Revenue being hedged, reference price, volume, delivery period, settlement method, and counterparty obligations.

Hedging instruments

Develop price protection for buyers’ future compute spend and operators’ revenue, including the period after an offtake ends.

What the structure needs to answer: The exposure and reference price, differences from actual compute prices, collateral, settlement, and any premium or fees.

Pricing data

Compare capacity by configuration, location, and term to inform underwriting and the value of remaining commitments.

What the structure needs to answer: Data provenance, comparable transactions, quote versus executed price, and the assumptions behind a valuation.

For datacenters, clouds & capacity holders

Offer capacity

We have standing orders to purchase compute contracts that meet our pricing, terms, and technical specifications. We build ongoing relationships with suppliers as our inference needs and network grow.

We purchase capacity for our own inference workloads and for the startups and independent researchers in our network. Bring us existing capacity or a new buildout, including edge compute and other novel deployments. We agree the workloads, access arrangements, and onward-use rights directly with you.

Existing capacity
Dedicated nodes or clusters with a defined specification, location, start date, and service commitment.
New buildouts & edge compute
New operators, edge sites, and upcoming deployments. We can discuss forward purchases against your specifications, commissioning schedule, and proposed terms.
Unused reservations
A remaining term or a portion of a cluster, where the provider permits the proposed onward use or transfer.

From specification to a purchase

  1. Scope the capacity

    Hardware, interconnect, storage, location, availability, and the rights you can offer.

  2. Agree the commercial terms

    Volume, duration, price, prepayment, service levels, and delivery obligations.

  3. Define acceptance & access

    Tests, handover, and remedies agreed for the configuration before capacity is accepted.

View acceptance tests

We agree the hardware specifications and performance targets before handover. An initial test window is followed by final checks on the full cluster.

Hardware health
GPU inventory and DCGM Level 3 diagnostics, with NIC link health, shared storage mounts, and management access checked across nodes.
Sustained operation
A 48-hour cluster stress test, with logs checked for node crashes, GPU errors, and thermal throttling.
Compute performance
HPL and HPL-MxP benchmarks, plus an agreed multi-node reference workload, measured against the configuration’s performance targets.
Network scaling
NCCL all-reduce tests from a single node to the full allocation, checking NVLink bandwidth, inter-node throughput, and stable scaling.
Storage under load
Concurrent read, write, and metadata benchmarks across nodes, including storage tests alongside collective communication.
For trading desks & commercial partners

There is more than one side to a position.

We welcome counterparties who buy and sell physical compute, source difficult allocations, or see a price and term that our book can work with.

Trade physical capacity
Bring a bid or offer for a defined configuration and delivery window. We can assess it against our positions and purchasing needs.
Work an allocation together
Bring supplier access or a customer requirement. Establish the contracting roles, disclosure, and fee arrangement before the work begins.
Revenue forwards & hedging
Work with us on structures for future compute revenue or costs, supported by pricing, contract standards, and verification.

Get in touch

View bids & offers

Before we work together

Are we contracting with Epilogue?

Yes, we contract for our own clusters. The exception is some listings on the Orderbook from third parties, as we do not purchase colo or large-scale clusters ourselves.

Can you support a deal we already have?

Bring the contract or project. Capital partnerships can concern transactions originated by others as well as our own book. The role, financing availability, and terms depend on review of the specific deal.

Can I offer capacity reserved from another cloud?

Yes, where your agreement permits the proposed use. We need to establish the relevant resale, subletting, or assignment rights and obtain any required provider consent before proceeding.

Can you buy a portion of a cluster or term?

We can assess a defined subset of nodes or a delivery window. It needs to work operationally and fit the supplier’s permitted allocation and access arrangements.

Do you work with brokers?

Our terms are set by our internal pricing models. We are happy to work with brokers that bring us listings that meet those, and can pay a referral fee.

The next conversation

Bring us something worth building around.

A cluster coming online. A contract that needs capital. A position with room on the other side.

Get in touch