The budget we take first
The CIO already defends a rented stack. We take that invoice.
The CIO already defends a rented stack. We take that invoice.
The line they already take upstairs
The first dollars come from the line the CIO already walks into a board — seats, tokens, integration retainers, and the people who re-explain the company. The committee already pays OpenAI, Azure, Anthropic, or Bedrock. Those seats go blank when the tab closes. The token meter charges to replay last quarter. Staff paste the firm back in on Monday. Systems-integration retainers keep the window wired to the estate. That labor is already funded. What they have bought is a scratch pad with a meter on it.
The product we put on their floor is organizational memory. Write a fact once. Get the same fact back — the original — on hardware the buyer holds. The sale is a substitution: own the archive. When the contract ends, the memory stays in the building. When it continues, they stop paying a vendor to reconstruct who they already are.
Write a fact once. Get the same fact back. They own the record.
If the memory holds, the token meter is the expensive way to keep Monday. We take that invoice. We do not open a science budget beside it.
How large the signed bill already is
A typical 10,000-employee generative stack is modeled at $4–8 million a year. A 100,000-employee estate is modeled at $30–55 million a year. Those figures are incumbent spend, not Trinity revenue. They measure an invoice already signed — seats, tokens, integration, and the replay labor that never appears on the model invoice.
On a constructed Fortune 50 three-year comparison, the same substitution produces a $90.9 million takeaway: what the estate pays the incumbent layer minus what it would pay to keep the memory on machines it holds. The construction is a model. Diligence can reopen the arithmetic. The first dollars we intend to collect are a handful of those invoices, not a percentage of a national appropriation.
The Department of War’s $58.5 billion AI request is a policy line. It is weather around a later door. It is not our serviceable market. We sell a machine that sits inside the buyer’s perimeter and displaces a line item already upstairs. A government install, when it comes, is a task order against that same substitution — memory on their side of the wall — not a claim on the entire defense request.
A market that already writes this check
Palantir closed 73 deals of $10 million or more in the second quarter of 2026. That print is existence of a market that writes eight- and nine-figure software checks. It is not our pipeline. It is not our bookings. It tells an investor that the committee we want already signs in this band.
Seventy-three deals of $10 million or more is check-size existence. It is not a Trinity book.
A $2 million, $8 million, or $25 million software land is a live procurement motion for this class of buyer. Named-account work, an NDA, a proof on their hardware, and an install they keep — that is the motion we take to a different product: sovereign organizational memory on their floor. The ticket is the invoice already upstairs.
Three prices for the same store
What we put on the table is one product — sovereign organizational memory — priced as three SKUs. Same store. Wider land. None is live. They are the offer we take into the first proofs.
| SKU | Annual value | Who it is for | Proof |
|---|---|---|---|
| CORE | $2 million | First land, one division | 6 weeks |
| DIVISION | $8 million | A line of business | 10 weeks |
| ENTERPRISE | $25 million | Flagship or sovereign install | 16 weeks |
CORE is the door: first land, one division, six weeks, $2 million annual. DIVISION is the expand: a line of business, isolation, and a software air-gap option in ten weeks, $8 million. ENTERPRISE is the flagship: sixteen weeks, $25 million, air-gap as the default path when the data cannot leave the building.
As the story grows, commercial first customers plan in a $4–12 million band and sovereign work in a $15–40 million band. Those are later planning ranges. Year 1 opens on CORE, DIVISION, and ENTERPRISE. Year 1 is not two $40 million flags. The land is the memory. The expand is more of the same store, not a second product beside it.
The Year-1 mix is still a plan
The Year-1 base plan is $96 million booked and $48 million recognized, on a mix of six commercial and two government logos — none of them named today. Recognition is partial-year by design: first signatures around month five, a blended half of booked annual value recognized in the twelve months after close. The downside plan is $10 million recognized.
Proof-of-value conversion of 40 percent on CORE, 40 percent on DIVISION, and 35 percent on ENTERPRISE is a planning analog from early-pilot bands in this category. It is not a Trinity cohort we have already run. The motion is private: named-account development, an NDA, a proof on their hardware, and an install they keep. Public brand waits until logos exist.
Until a named row exists, the mix table is a construction. Diligence can reopen it. The sale does not change with the mix. First dollars still come from invoices the buyer already signs.
Twenty million stands the proofs
$20 million is an ask — seed to stand the factory that can run those proofs and put memory on the buyer’s floor. There are no named logos on this page. That is why the raise exists.
The factory is what a named committee sits on their hardware. Unused seed does not become a street campaign. Public brand waits until a logo exists. The sale stays the substitution: first dollars from the invoice already upstairs.
Sources: Trinity Sky, Investor business plan, August 2026, §11 “The budget we take first” (docs/gtms/09-raise/BUSINESS-PLAN-INVESTOR.md). First dollars from seats, tokens, integration retainers, and the staff who re-explain the company. 10,000-employee $4–8M/yr and 100,000-employee $30–55M/yr modeled incumbent spend. Constructed Fortune 50 three-year takeaway $90.9M. Palantir 73 deals ≥$10M is check-size existence. SKUs CORE $2M / DIVISION $8M / ENTERPRISE $25M proposed for first proofs. Y1 base $96M booked / $48M recognized is a plan. $20 million is an ask. No named logos.