Home / Decisions / Cloud Exit Assessment
Decisions · capital-grade

Exit the cloud or stay — a decision you can defend to the board

A reasoned answer for one rented workload: repatriate it or stay — with a number you can lean on. Sometimes the answer is "stay." We'll say so plainly.

Request an assessment See a sample verdict →
On the tableA Go / Conditional-Go / No-Go verdict, an honest break-even, the cost of being stuck both ways (Anchor), a demand check (The Bet) — one page for the board.
Who it's forThe CFO and CTO/CIO. This is a capital decision.
EngagementTimeline [timeline] · Price [price]

The moment you arrive here

The cloud bill crossed the line where "buy your own" is no longer theory but a question on the table. The board wants to understand why you pay this much. Vendors circle with decks where everything always pays off. FinOps shows what it costs now, but doesn't answer what happens if demand sags or capital gets more expensive.

And the one number nobody shows you is what it costs to be wrong. Exit for nothing — and you're stuck in hardware you can't resell on an illiquid market. Stay for nothing — and you overpay for years. The decision is irreversible, the cost of error is asymmetric, and it's you and the CFO who answer for it personally.

If this is about GPU rental, the question is sharper still. Stopping accelerator rental to stand up your own is the hottest capital decision of 2026: the bill grows fastest, lead times are long, and the board is asking. A dedicated read for GPU repatriation → [[FILL: link to the future segment variant, or drop this link until it exists]]

This isn't only our logic. Cloud repatriation is a public category with its own known exit stories and counter-analysis. [[FILL / can source on request: 2–3 real public links to cloud-repatriation stories and category analysis]]

What you get

A reasoned decision on one specific rented workload: repatriate it (to owned hardware, colo, neocloud or hybrid) or stay — and under what conditions. In your hands:

An answer that survives the boardNot a consultant's opinion, but a verdict with conditions — exit / exit-if / stay — that you put on the board's table and can answer for.Verdict: Go / Conditional-Go / No-Go
The point where owning is genuinely cheaper — and on what assumptionsAn honest break-even against today's cloud, costed not by vendor list price but by real, risk-adjusted cost. And not just "how much total," but "how much value per dollar."True Cost · Compute Yield
The real price of being stuck — both waysLeaving the cloud cuts vendor dependency but loads on capital dependency. A light rental anchor swaps for a heavy ownership anchor — and we show that trade openly, as a number, not a footnote.Anchor — your cost of exit
A check that you're betting on demand that arrivesMoving into hardware for a workload that can evaporate is the top way to get idle capital. We look at where your demand sits relative to break-even, and what happens on a dip and on a spike.The Bet
The real time and cost of getting into hardwareA server doesn't appear on signing day. How long you wait for capacity and what the "bridge" of that wait costs goes into the math — and often eats half the "owning is cheaper" win.Supply
An honest answer on whether your team can carry itThe most expensive failure isn't in the economics — it's in execution. We check this against verifiable facts, not feelings, and say it straight.Execution Readiness

The assessment collapses around the factors that actually decide the exit question; whatever we didn't assess, we flag honestly — nothing painted in. (In the standard: Partial Keel.)

And two lines we don't cross. If optimization without any move solves the problem, we'll say so and stop the calculation. If the target site can't carry your SLA, that's a "no" — however cheap it pencils out.

What lands on the table

A sample verdict — one page for the board. Sample — form, not real data.

Sample · illustrative form
Verdict: Conditional-Go (exit, with conditions)
Workload ‹name› · current posture: Hyperscale · recommended: Owned-Colo
True Cost — owning is cheaper from utilization ‹threshold›; your plausible corridor sits above it. By unit cost (Compute Yield) the gap is wider still.
Anchor — the cost to reverse rises from ‹rental› to ‹ownership›: you trade vendor lock-in for capital lock-in.
The Bet — pays off at demand no lower than ‹level›; your demand corridor lands inside it.
Rationale — one line on why it's exactly this.
Not assessed — Ground, Shelf Life (carried as flags).
Go conditions — ‹what must be true / done to exit›.
This is the form of the output, not the method of calculation: the page shows the decision and its supports, without revealing how it was computed.

And if the answer is "stay"?

Then you get exactly the same, in reverse: a board-defensible case to keep paying for cloud — with numbers that show why owning is worse here, and under what changes the question is worth re-opening. A "no" isn't an empty consulting invoice. It's a document that closes the question at the board and lifts the risk of an expensive "let's buy it anyway" off you. Our most frequent valuable result is precisely a clear "don't move yet."

The cost of not knowing the answer

You've already felt the cost of error: repatriate for nothing and it's seven figures of capital you can't get back; stay for nothing and it's years of overpayment. Against that, the assessment itself costs a fraction of one month of that overpayment, and a negligible fraction of the cost of idle hardware. The question isn't whether the assessment is expensive — it's how much it costs to make this decision blind.

Why you can trust us when we sell hardware

Let's name the conflict plainly: Haink supplies infrastructure, and on paper it suits us for you to leave the cloud and buy servers. So the assessment is built so it can't be tuned:

The math runs only on your real data — cloud bills and production telemetry. Vendor prices and decks, lab benchmarks, best-case slices are not admitted, even if they're put on the table. Data Admissibility
The method is an open standard — CCS (Compute Capital Standard): anyone, including your auditor, can read the factors, scale and logic. Only the calibrations and data are closed — on the rating-agency model.
And the clinching proof: we publish cases where the honest answer was "stay in the cloud."
A real "stay" case:
[[FILL: one anonymized but real case. Structure: (1) client situation and why they considered exit; (2) what was computed; (3) why the verdict was stay, with numbers; (4) what the client got / saved by heeding the "no." No invented numbers — real only.]]
Decisions, site-level — proof of the standard:
CCS is an open standard. [[FILL: link to CCS Public Reference, version number, date, change log. If available — names of early adopters / auditors who cite it.]]
Decisions, site-level — who's behind this:
[[FILL: team and track record. Where possible — "assessed $X of compute decisions" / number of projects / industries. The CFO buys a defensible verdict and should know its author.]]

Is this for you?

Five questions that show whether you're a candidate or not. We do honest work when we screen out the wrong fits early.

1.Is the bill for this workload large and steady, or small and choppy?

2.Have you already tried optimization (rightsizing, reserved/committed, spot) and hit the wall — or not yet?

3.Is the workload predictable 2–3 years out — or could it disappear or spike hard?

4.Are there hard latency / uptime requirements that can't be broken?

5.Do you have a team and experience to run your own hardware — or is it from scratch?

Answer the five above and we'll show, in plain terms, whether repatriation is likely your move — no thresholds revealed.

What this isn't

Not cloud optimization — in your current model we won't cut the bill. Not a from-scratch new-cluster decision — that's a separate product (Investment Review). Not a whole-portfolio review. Only the fate of one already-rented workload — and a clear answer on it.

Not sure you even have a problem? Start with the diagnosis — AI Infrastructure Audit → shows what you have and what state it's in, with no obligation to decide.

How the work runs

STEP 1
We take inAccess to your real cloud bills for a representative period and production telemetry (not a vendor deck), the workload profile, cost of capital and horizon, one or two target sites.
STEP 2
We computeBy the CCS standard, on admissible data. Real hardware prices, lead times and hosting cost are on our side.
STEP 3
We put on the tableA one-page verdict plus a factor-by-factor breakdown.
STEP 4
Timeline[timeline]

Frequently asked questions

Why not our own FinOps?

FinOps optimizes spend within a model you've already chosen and answers "what does this cost now." It doesn't answer "is it admissible to own this asset" and what happens under a shock to demand, rates or energy.

Why not Big-4 or a consultant with no stake in hardware?

They lack two things: real data on pricing, supply and the secondary market — and real execution. We don't just calculate, we deploy it, so our numbers aren't from decks.

Your calibrations and data are closed — why should I trust the number?

Because what matters is open: the methodology, factors and scale are published and reproducible, and the conflict of interest is disclosed. Only the data and calibrations are closed — exactly like a rating agency, whose grades the market trusts precisely because the model is transparent while the inputs are not. The verdict can be re-checked against the standard; it can't be tuned.

And if the verdict is "stay" — did I pay for nothing?

No. A board-defensible "don't move yet," with numbers, is the result you're buying. Paying for an honest "no" is cheaper than paying for a wrong "yes."

Ready, or just curious?

Ready — we'll request the data and start. Not ready to talk — take the artifacts and decide at your own pace.

Request a Cloud Exit Assessment

Timeline [timeline] · Price [price]