Company
Every commodity that mattered eventually got a venue, a reference rate and instruments written against it — grain, crude, power, freight, bandwidth. Compute has the volume and the volatility and none of the machinery. We are building the venue, and we are building it neutral, because the alternative is another reseller with a price list.
Mission
The four largest US cloud providers have guided to roughly $725 billion of capital expenditure for 20262, and data-center electricity demand is projected to more than double by 20301. An input that large should not trade across a 4.7× spread depending on which vendor you ask3, and it should not be sold in year-long blocks to buyers whose demand moves weekly.
A venue fixes both. One book, one clearing price, visible to both sides, with forwards so a lab can cost a training run before committing to it and an operator can fix revenue before energising a hall. That is the whole company.
Values
Every externally-checkable figure on this site carries a citation to its source, with the sentence it rests on. Where there is no public evidence, the page says so. This is a venue — the moment people stop trusting our numbers, we do not have a product.
We do not take proprietary positions on our own book, and we do not favor our own capacity over a partner's. An exchange that trades against its members is not an exchange. This constrains our revenue model on purpose.
Every decision gets tested against buyer and seller. A feature that is good for demand and bad for supply is a feature that thins the book, which eventually makes it bad for demand too.
Settlement, custody and metering should be the least surprising parts of the company. We would rather ship a dull ledger that reconciles than an interesting one that does not.
Decisions of consequence get written down with the reasoning and the alternatives, before the decision. It is slower, and it is the only way a team that doubles stays coherent.
What we are for
Three statements that are ours rather than sourced, and one piece of history. They are here because a counterparty is entitled to know what a venue thinks it is doing before routing an order to it.
Compute is now an input cost on the scale of energy, and it is still bought the way enterprise software was bought in 2005: bilaterally, at a price neither side can check, on terms that depend on who you know. Every other input at this scale — power, freight, memory, metals — eventually got a market, because at scale a private price is a tax on everyone who cannot negotiate.
A venue is neutral or it is a dealer. Exascale takes no proprietary position in the instruments it lists: the exchange does not trade against the participants on it. That is the whole purpose, and everything else is downstream of it.
To make one hour of a named accelerator, in a named region, a quoted and tradeable instrument — so that a buyer can budget against a published number, an operator can finance against a forward curve, and a lender can mark a data center to a market instead of to a business plan.
The measure of whether this works is not our volume. It is whether somebody who has never traded here can look up what compute costs and be right.
These are written as commitments rather than adjectives, because an adjective cannot be breached.
The company started from an observation rather than a product idea: two parties who both wanted the same trade — an operator with idle accelerators and a team that needed them for six weeks — could not do it, because there was no price either would accept as fair and no mechanism to settle it if they had agreed one.
That is a market-structure problem, not a supply problem. Building the venue was the only way to find out whether it was the binding one.
NoteThe dates, the people and the specifics of that first period are held in the company deck and are not reproduced here. This page states the reasoning; it does not narrate a history it has not sourced.
Team
Founder · Vision & Technology
Architected the exchange, the AI platform and the GPU cloud: the matching engine, the model routing layer, and the bare-metal orchestration stack on Slurm and Kubernetes. Working in AI and machine learning since 2014; previously in private equity.
At ExascaleTechnical vision, product, and the engineering organization.
Founder · Chairman · Capital
Vice Chairman at HSBC and Head of its Global Banking & Markets division for twenty years. Created Morgan Stanley’s Leveraged & Acquisition Finance business in Europe. Thirty years structuring capital for the world’s largest private equity firms.
At ExascaleInstitutional capital — private equity, sovereign wealth, enterprise relationships.
Founder · People & Operations
Sold cloud infrastructure at enterprise scale — more than $130M in AWS contracts to three Fortune 100 global enterprises. Then private equity, on a fund investing in alternative finance, power generation and industrials.
At ExascaleGlobal data-center partnerships, AI infrastructure sales, capital, and talent.
Four from Anthropic, OpenAI and DeepMind. Six PhDs. Engineering comes out of Google, Microsoft, AWS, Tesla, AMD, TSMC and Scale AI.
Offices
Markets, clearing, institutional coverage
Where the counterparties are. Desks, funds and the banks that will eventually clear for them sit within a mile of each other.
Trading, risk, settlement operations, institutional sales.
Engineering, research, product
Where the demand is written. Our largest buyers are building models a few blocks away, and proximity to them shapes the instrument set.
Matching engine, consoles, metering, ML systems.
Regional venue, supply origination, MENA coverage
Abu Dhabi Global Market is a financial free zone operating under its own common-law framework with an independent regulator, and it sits next to some of the fastest data-center build-out and cheapest firm power in the world. Supply origination and regional clearing belong here rather than in New York.
Supply partnerships, regional clearing, energy-adjacent capacity.
NoteEntity structure, regulatory permissions and which activities are conducted from which entity are set out in our counterparty pack, available under NDA to institutions in onboarding. Permissions differ by jurisdiction and not every instrument is offered from every entity.
Careers
| Team | Role | Location | Type | What it is |
|---|---|---|---|---|
| Exchange | Matching engine engineer | New York · SF | Full-time | C++ or Rust. Price-time priority, deterministic replay, sub-millisecond. |
| Exchange | Clearing & settlement engineer | New York | Full-time | Ledgers that reconcile. Escrow, netting, T+0. |
| Exchange | Market risk analyst | New York | Full-time | Margin models for an asset with no historical curve. |
| Supply | Supply origination lead | ADGM · New York | Full-time | Bring data centers onto the book. Power, contracts, attestation. |
| Supply | Infrastructure engineer, metering | SF | Full-time | Per-second metering across heterogeneous fleets, auditable end to end. |
| Demand | Forward-deployed engineer | SF · New York | Full-time | Sit with buyers, move their workloads, bring back the instrument they actually needed. |
| Demand | Inference platform engineer | SF | Full-time | Serving across twelve modalities without twelve codebases. |
| Research | Quantitative researcher, compute markets | New York · London | Full-time | Build the reference rate. There is no precedent for this instrument. |
| Company | Regulatory counsel | ADGM · New York | Full-time | Venue permissions across three jurisdictions. |
| Company | Technical writer | Remote | Full-time | Own the citation discipline on this site and in the docs. |
We reply to every application within five working days, including the rejections. Write to careers@hyperlink.org.
Internships
Ship into the matching engine or the settlement ledger. Paid, with housing.
Work on the reference rate and the forward curve alongside the research team.
Power markets, capacity origination, and the economics of siting.
For PhD students working on compute markets, scheduling or energy systems.
Interns work on production systems with a named owner and ship to the book. We do not run shadow projects. Applications open twice a year; write to careers@hyperlink.org at any time.
Every figure above is numbered to an entry here. Links last read 27 July 2026.
Energy and AI — Executive summary
Electricity demand from data centres worldwide is set to more than double by 2030 to around 945 terawatt-hours (TWh) … slightly more than the entire electricity consumption of Japan today.
Google, Amazon, Microsoft, and Meta collectively plan to allocate $725 billion to capital expenditures in 2026 — up 77% from last year's $410 billion.
A sum of separate company guidance ranges, not a reported figure. Individual guidance: Amazon ~$200B, Google $175–185B, Meta $115–135B, Microsoft $110–120B.
H100 Rental Prices Compared: $1.49–$6.98/hr Across 15+ Cloud Providers
ParaphraseOn-demand H100 list prices span $1.49 to $6.98 per GPU-hour across more than fifteen providers, a spread of roughly 4.7× for the same silicon.
Survey of published list prices. Committed and reserved rates are negotiated and are not represented here.
Where a claim rests on a third-party estimate rather than the party that owns the number, the entry says so. Figures that are Exascale’s own — our rate card, our fee schedule — carry no citation, because they are ours to set rather than facts about the world.