# Cold-Start Strategy

**Compute for Equity — investor memo**
*Status: strategy memo. Prepared for the data room, June 2026.*

---

## 1. The honest problem

Every marketplace dies the same death. It needs supply to attract demand and demand to attract
supply, and on day one it has neither. The graveyard of "compute marketplaces," "GPU exchanges,"
and "tokenized infrastructure" projects is mostly companies that built a beautiful two-sided
platform and then stood in an empty room waiting for liquidity that never came. We have studied
that failure mode closely, and our entire go-to-market is organized around not repeating it.

The single most important sentence in our strategy is this: **we do not need 1,000 participants on
day one. We need 2 great ones.** One large, energy-rich datacenter with compute to deploy, and one
high-quality AI startup that needs that compute and would rather pay in equity than cash. One deal,
done well, with real money and a real instrument, is worth more than a thousand sign-ups. We
therefore reject the "launch the platform and scale users" plan in favour of a deliberate,
three-phase sequence that builds liquidity the way successful exchanges actually built it:
manually, narrowly, and from the supply side first.

## 2. Phase 1 — "Fat deals," brokered by hand (Q3–Q4 2026)

In Phase 1 we are not a platform. We are a **dealmaker that happens to be building a platform.**
We originate and broker a small number of large, high-conviction single-pair transactions — one
datacenter, one startup — entirely by hand. No self-serve. No open marketplace. A human structures
each Compute-SAFE, prices the compute with the Oracle, papers the deal under the ADGM RegLab
perimeter, and shepherds it to settlement.

This is deliberate, and it does five things at once:

1. **It proves the instrument.** A Compute-SAFE that has actually been signed, drawn against, and
   converted is worth infinitely more than a term sheet on a website. Phase 1 produces real
   reference deals.
2. **It builds the muscle.** Every manual deal teaches us where the friction is — in pricing, in
   custody, in the SLA clause, in onboarding. That learning becomes the product spec for Phase 2.
3. **It generates revenue early.** Origination and clearing fees on a handful of large deals fund
   the build without waiting for marketplace scale.
4. **It de-risks the regulator.** Small, supervised, high-quality deals are exactly what RegLab is
   for. We earn the supervisory track record that the full licence requires.
5. **It creates gravity.** The first credible deal is the hardest. Once a respected datacenter and
   a respected startup have transacted, the second and third conversations start from "who else has
   done this" instead of "has anyone done this."

The bar in Phase 1 is quality, not count. We would rather do three excellent deals than thirty
mediocre ones. The target is roughly **ten reference transactions** by the end of Phase 1 — enough
to standardize from, not so many that we are running an unlicensed marketplace before we are ready.

## 3. Phase 2 — Standardize and open to vetted participants (2027)

Phase 1's ten deals are the raw material for Phase 2's core asset: a **standardized Compute-SAFE
term sheet.** Bespoke deals do not scale; a standard contract does. In Phase 2 we take everything
the manual deals taught us and crystallize it into a default instrument — fixed cap/discount
mechanics, a standard SLA-suspension clause, standard draw windows, standard custody and conversion
terms. The standard is what turns a series of private contracts into the beginning of an asset
class.

With the standard in hand and a full Financial Services Permission in flight, we open the **clearing
platform to vetted participants** — not the open internet, but a curated cohort of qualified
startups, investors, datacenters and energy producers who clear against the standard instrument.
The platform now does in software what a human did in Phase 1: price, clear, custody, settle. We
still gate entry hard. Every participant is vetted; every instrument is the standard or a
controlled variant. This is the phase where the company stops being a dealmaker and becomes a
venue — but a members' club, not a public square.

Why supply-first, again, in Phase 2? Because compute is the constraining side. Energy-rich
datacenters with capacity to deploy are fewer, larger, and slower-moving than AI startups that need
compute — and they are the harder, stickier relationship to win. Anchor the supply, and the demand
side (cash-constrained, dilution-sensitive AI startups) is comparatively easy to attract: they have
an acute, recurring pain and a clear incentive to switch. We spend our scarce early credibility on
locking in supply, and let demand pull itself in.

## 4. Phase 3 — Open marketplace and secondary (2028+)

Only in Phase 3 do we resemble the thing the pitch describes: an **open marketplace** with a
**secondary market for Compute-SAFEs.** By now the instrument is standardized, the licence is full,
custody is audited, and there is a real book of primary deals whose holders want liquidity. The
secondary market is not a launch feature — it is an *earned* feature that becomes possible only once
there is enough standardized primary volume for positions to trade. Trying to launch a secondary
market on day one, against zero standardized inventory, is the classic way to build an exchange with
no trades. We sequence it last on purpose.

In Phase 3 the network effects we were careful not to depend on early finally start to compound:
more supply tightens pricing, tighter pricing attracts more demand, more demand attracts more
supply, and a liquid secondary market makes every primary position more valuable. The flywheel is
real — we simply refuse to assume it exists before we have hand-built its first revolution.

## 5. Why supply, specifically, and why the Gulf

Our cold-start has an unfair advantage most marketplace startups lack: **a structural supply
relationship.** We are building in Abu Dhabi, beside sovereign and infrastructure capital and some
of the cheapest energy and largest planned compute clusters on earth. The hardest side of our
market — energy-rich compute supply — is exactly the side our geography and relationships make
easiest. A San Francisco marketplace has to *convince* datacenters to participate; we start the
conversation already inside the room where that capacity is being deployed. That is why the company
is being built here first, and it is the reason our Phase 1 supply-acquisition risk is materially
lower than a generic marketplace's.

## 6. What success looks like at each gate

- **End of Phase 1:** ~10 brokered Compute-SAFE deals settled under RegLab; a proven instrument; a
  standardized term sheet drafted; early origination revenue; full-licence application filed.
- **End of Phase 2:** full FSP granted; clearing platform live for a vetted cohort; the standard
  instrument in repeated use; supply anchored by multiple datacenter relationships.
- **Phase 3:** open marketplace + secondary market live; network effects compounding; the venue is
  the default clearing house for compute-for-equity in the region.

## 7. The risks we are explicitly managing

- **Phase 1 takes longer than planned.** Big deals are slow. *Mitigation:* we fund Phase 1 as a
  dealmaking operation, not a scaled platform, so burn is low and survival does not depend on speed.
- **Premature scaling.** The temptation to "open it up" before the standard and the licence are
  ready. *Mitigation:* the phase gates are explicit and we hold them.
- **Supply concentration.** Early dependence on one or two datacenters. *Mitigation:* Phase 2 widens
  the supply base before the marketplace opens.
- **Demand without supply (or vice versa).** *Mitigation:* supply-first sequencing and single-pair
  brokering mean we never advertise a side of the market we cannot fill.

## 8. Anatomy of a Phase 1 deal (what "by hand" actually means)

To make Phase 1 concrete, here is the shape of a single brokered transaction. A regional datacenter
has a block of H200 capacity coming online with cheap, contracted energy behind it. We have an AI
startup — vetted, venture-backed, dilution-sensitive — that needs a six-month training run and would
rather not burn cash on it. We sit in the middle. Our Oracle prices the compute into a fair GPU·h
quote. We structure a Compute-SAFE: a capped, discounted instrument with a draw window and an
SLA-suspension clause, papered under the RegLab perimeter and custodied in a segregated account. The
datacenter commits compute, not cash; the startup draws it as it trains; nothing converts until the
startup's next priced round. We earn an origination and clearing fee. Every step that was manual
here — pricing, papering, custody, settlement — is a line item in the Phase 2 product spec. Ten of
these, done well, *are* the standardization corpus. This is why "by hand" is not a weakness of the
plan; it is the data-collection mechanism for the platform.

## 9. What we will deliberately NOT do

Discipline is the strategy. We will **not** open a self-serve marketplace before the instrument is
standardized and the licence is in flight. We will **not** chase sign-up vanity metrics — a
thousand registered users who never transact is failure dressed as traction. We will **not** launch
a secondary market against zero standardized inventory. We will **not** market to retail. Each "not"
removes a classic marketplace failure mode. Saying no to premature scale is how we earn the right to
scale.

## 10. The one-line summary for the partner across the table

We are not betting that "if we build it, they will come." We are hand-building the first ten deals,
standardizing what we learn, opening to a vetted cohort, and only then opening the doors and the
secondary market. **Two great participants, then ten great deals, then a market.** That is a
sequence a careful investor can underwrite — not a leap of faith.