What the public record shows about exhibit costs

Every figure below is computed directly from the dataset and can be reproduced by filtering the browse page. Each finding is stated as the answer to a question a reader actually brings, with the caveats that matter — and the methodology page documents the sampling and composition effects you should read before citing any median. This is a descriptive record, not price advice.

1. What does an interactive exhibit cost? The record's price rungs

Here is what public awards actually run, by project scope. Medians alone mislead — the middle half of each class (25th–75th percentile) is the honest way to read it. Federal awards are the dataset's largest and most consistently classified source; all-US figures add grants and state contracts, which tend smaller.

Three things to know before quoting these:

These are records, not the market. The medians describe what this record captured, given its sources — and they move when the sources do. Adding 144 National Park Service design-only task orders in a 2026-07-22 sweep extension shifted the federal single-interactive median from roughly $68,000 to $45,000. The middle-half ranges are more stable than the medians and should be preferred in any citation.

Small task orders pull the medians down. A handful of high-volume government vendors contribute hundreds of small awards (the dataset's most frequent recipient averages about $117,000 per row across ~280 rows) — about a quarter of multi-station rows come from the five most frequent firms alone. A new-build, custom gallery program sits in the upper part of its class range, not at the median. See the sampling notes.

The kiosk class is a product market, not a bid market. 12 of the 31 kiosk rows are one vendor's (Ideum's) multi-touch tables, running $16,737 to $199,126 by size and customization — including three separate awards at exactly $36,629, which is schedule pricing, not competitive bidding. At kiosk scale, government buyers pay list price for a product; custom work starts above this class.

2. The biggest dollar figures in public data are not exhibit costs

The largest raw obligated amounts in the whole dataset — several in the $100M–$545M range — are not exhibits. They're whole-building construction-management and renovation contracts (e.g. the Smithsonian's National Air and Space Museum and National Museum of African American History and Culture revitalizations), classified project_class: other and excluded from every exhibit figure above by the scope-normalization standard.

This is worth stating plainly: if you sort public spending data by raw dollar amount without checking scope, you will conclude exhibits cost far more than they do — and this is exactly how "the new museum cost $500M" headlines get conflated with exhibit budgets. The classification system exists specifically to prevent that conflation.

3. Two published "exhibit cost" figures are not comparable until you know what's bundled

Published exhibit-cost figures differ less by market than by scope definition: whether a number covers interactive design, software, and AV integration alone, or bundles in exhibit fabrication — casework, graphics, physical build-out. Every row in this dataset carries a fabrication_inclusive flag so the two are never silently averaged together.

An honesty note on what the data does not show: pooled across all classes, fabrication-inclusive awards have a higher median than fabrication-excluded ones ($150,000 vs. $98,040) — but that gap is almost entirely class composition, not a measured fabrication premium. Within the multi-station class, where the data is deepest, the two medians are statistically indistinguishable ($165,728 vs. $160,222, n=412 vs. 56); the smaller classes are too thin to say. The flag exists so readers can compare like with like — not because this dataset has measured what bundling adds. See the composition-effects section of the methodology for the full computation.

4. Maintenance is its own market, not a rung on the ladder

Keeping existing exhibits and AV running is a distinct contract type — and it is not small. The dataset's maintenance class (n=86 US rows with usable amounts) has a median award of $72,307 ($59,825 federal-only), with a middle half of roughly $23,000 to $250,000 — overlapping the single-interactive class entirely. Many of these are multi-year support and service contracts for whole galleries or visitor centers.

The practical reading: institutions routinely commit single-interactive-sized money just to keep what they already built working. Anyone budgeting a new interactive without a maintenance line is reading only half the public record.

5. Major museums pay a standing exhibit vendor six — sometimes seven — figures a year

IRS Form 990 filings disclose a nonprofit's five highest-paid independent contractors over $100,000 — the only public window into nonprofit museum spending, which mostly never touches a procurement portal. Because the disclosure is capped at the top five and floored at $100K, these rows are by construction the top of each museum's vendor spending, not a market average — so this dataset does not publish a "median vendor payment" from them.

What the rows do establish: 13 of the 58 exhibit-relevant contractor disclosures exceed $1M in a single tax year, topping out at $7.1M (American Museum of Natural History to Design and Production Inc.), with the National Civil Rights Museum's $3.2M to Howard+Revis close behind. A museum's annual payment to a standing exhibit vendor can span several concurrent projects — these are relationship-level figures, not project prices — but they show that at major institutions, exhibit work is a continuous six-to-seven-figure vendor relationship, not an occasional purchase.

Only 58 of ~537 extracted contractor rows (across 156 researched museums) were kept as plausibly exhibit/AV/design related — see the methodology page for the keep/skip rule.

6. Whole-museum exhibit fit-outs: thin public data, and the big numbers are ceilings

The building-wide class — every exhibit in a facility, contracted at once — is the question buyers most want answered and the one public data answers worst: only about a dozen US rows. What they show is a range from $395,820 (a small national-park visitor center) through $1.4M (a NASA visitor center's complete exhibit rebuild) to $45.2M obligated (the National Museum of the U.S. Army's exhibit fabrication; $69.6M ceiling) and $7.1M obligated / $48.6M ceiling (the Smithsonian NMAAHC's inaugural exhibits).

The trap in this class is ceiling values. The largest figures in it — $57.4M and $13.5M, both one vendor's Texas Historical Commission awards — are flagged scope_confidence: low because they read as IDIQ contract-vehicle ceilings, not confirmed single-project spend. Obligated amounts and ceilings can differ by 7× on the same contract. Read any whole-museum number with the flag, the obligation, and the ceiling in view.

Sparse by design: the class is published with its real, still-small n rather than padded. See the methodology page for the 2026-07-22 reclassification pass that assembled it.

7. UK tenders land in the same range as mid-size US work

UK museum-sector tenders (non-US comparables, kept in their own file with a currency column, n=9) show gallery-scale exhibition fit-outs at roughly £380,000 to £900,000 — the same order of magnitude as the US multi-station middle half. The single UK building-wide row is a £2.2M whole-project exhibition-design-team appointment. Currency, VAT, and procurement-system differences make these directional comparisons only.

8. Most federal awards were never separately advertised — and that's not a red flag

A reasonable question when a row has a terse, site-code-only description and no findable public notice: was this competed at all? Every federal row now carries the answer, pulled directly from USAspending's own award-detail record (visible in the Competition column and filter on the browse page):

The 53% figure is most of the explanation for why solicitations are often unfindable: under FAR 16.505, placing a delivery order or BPA call against a vehicle that was already competed (a GSA Schedule, an agency-wide IDIQ) only requires the agency to give the vehicle's other holders "fair opportunity to be considered" — typically a quiet request for quote sent directly to a handful of eligible vendors, never posted publicly. That's exactly the mechanism behind this dataset's National Park Service and Smithsonian design-only task orders (see the methodology page's cross-reference section): the underlying vehicle was competed once, years earlier; the individual order for "Little Bighorn Battlefield" or "Lake Mead" wasn't.

Where a public notice did exist, SAM.gov generally purges it about a year after posting — so publicly_posted: YES confirms a notice existed at the time, not that it's still retrievable today. The CSV's new solicitation_id and parent_piid columns give a concrete number to search the Wayback Machine or a commercial contract-data aggregator for, when one is worth chasing.

This describes what USAspending's own record says about each award, not an independent assessment — accuracy depends on how the contracting officer coded it at the time, which federal data-quality audits have found is not always consistent. See the methodology page for the full field definitions and how this was pulled.