What It Costs and What It Returns

A proposal that does not price itself is a wish. This post states what the proposed system costs to run, what it would have to recover to pay for itself, and which of its inputs the answer actually depends on. The last of those is the only interesting one.

Priced from published rates, reported as an interval

[MODELED] The requirements list costs between $2.03 million and $2.46 million annually, with a midpoint of $2.25 million. Over three years that is $6.10 million to $7.37 million. Personnel, platform and enablement are the three categories, and personnel dominates all three scenarios.

Annual, low

$2.03M

Annual, midpoint

$2.25M

Annual, high

$2.46M

Three year total

$6.10M to $7.37M

Figures are rounded to three significant figures because the inputs do not support more. The loaded cost multiplier and the enablement percentage are conventions rather than measurements, and a total carried to the dollar would imply a precision those inputs do not have. The unrounded arithmetic is reproduced step by step in the plan’s appendix so that every line stays checkable.

[OBSERVED] For scale, the midpoint annual programme cost is 0.0097 percent of fiscal 2026 operating expenses of $23.076 billion on recomputation (Form 10-K, Item 7, p. 41). The cost is not the question.

Stated against the pool it is aimed at

The system is aimed at the $2.7 billion of fiscal 2026 provisions that no export control action explains. Break-even is therefore stated as a fraction of that pool rather than as a fraction of revenue, because revenue is not the thing the system acts on.

Cost case	Annual cost	Break-even share of the $2.7B pool
Low	$2.03M	0.0753%
Midpoint	$2.25M	0.0832%
High	$2.46M	0.0910%

Table 1. Break-even thresholds. Modeled. Computed as annual cost divided by the addressable pool. The full division is reproduced in the plan’s arithmetic appendix.

The system has to recover less than one tenth of one percent of the pool it is aimed at before it pays for itself. That is a low bar, and stating it plainly is more useful than dressing it up.

One input governs the result, and it is not the cost

[MODELED] Cost varies by 20.8 percent across its stated range and moves the break-even threshold by 1.57 basis points of the addressable pool. The assumed reduction in provisions varies by a factor of five across its range, from 1 percent to 5 percent, and moves the annual benefit by the same factor.

THE FINDING WORTH CARRYING

One input governs the result, and it is not the cost. Every hour spent refining the price of a business intelligence analyst is an hour not spent establishing whether the system reduces provisions by one percent or by five.

What these numbers are not

The benefit band of 1 to 5 percent is a stated assumption rather than a derived estimate. The two published benchmarks available at the time of writing are eight and eighteen years old respectively, and one of them was published by a vendor selling the product it benchmarks. Neither is load-bearing here and neither is presented as though it were.

The staffing model is also invented. No internal headcount, salary band or platform contract was available from outside the company, so the requirements list was sized from published wage data and published list prices and then labeled as modeled throughout. A reader who wants a different number should change the inputs rather than argue with the output.

What survives that honesty is the shape of the answer. The cost is small against the pool, the break-even is low, and the uncertainty lives almost entirely in one assumption that the client could measure in a single quarter and I cannot measure at all.

Provenance and marks

This post is part of the NVIDIA capstone for the M.S. Business Intelligence program at Full Sail University. It originates in BIN520-O, the second course of the program, and therefore predates the other capstone posts published here. Epistemic marks appear before the claim they govern: [OBSERVED] read from a primary source and verified against it, [MODELED] produced by a stated method from stated inputs, [INFERENCE] reconstructed from indirect evidence, [CONVENTION] a customary figure with no published benchmark behind it, [UNVERIFIED] asserted somewhere but not resolved to a locator.

Non-affiliation notice. The author is not employed by, affiliated with, sponsored by or endorsed by NVIDIA Corporation. NVIDIA is named here as the subject of an academic analysis prepared by an independent party, from public disclosure only. No confidential information of NVIDIA Corporation was used, and no representation is made on its behalf.