NIH research grants are getting fewer and nominally bigger, but real award size is roughly flat to declining

Summary

Federal research grants are becoming fewer in number but nominally larger — and once inflation is removed, the "bigger" part is much weaker than the nominal numbers suggest. At NIH, the average size of a Research Project Grant (RPG) rose from $566,744 (FY2020) to $675,426 (FY2025), a +19.2% nominal increase, while the number of new/renewal (competing) RPG awards fell from 11,052 (FY2023) to 8,161 (FY2025), a -26.2% drop in just two years, with the RPG application success rate collapsing from 21.3% to 13.0% over the same span. In inflation-adjusted (2025$, CPI-U) terms, however, the average award actually shrank from about $705,000 (2020) to $675,000 (2025) — roughly flat to slightly down in real purchasing power, because most of the nominal gain was eaten by the 2021-2023 inflation surge. NSF shows the same directional pattern independently: its average award size rose ~15% from FY2024 to FY2025 while new awards fell ~20%, though NSF's own FY2026 data complicates the "bigger" half of the story (see below).

Step 1 — What the literature and primary sources show (NIH)

NIH publishes its own year-by-year award-size and award-count tables in the annual "Extramural Nexus" post (grants.nih.gov). The FY2025 edition, fetched directly, gives:

A separate NIH release ("FY 2020 By the Numbers") gives the FY2020 average RPG size as $566,744. Total NIH extramural awards of all kinds (competing + non-competing) actually fell from 59,102 (2023) to 55,394 (2025), a -6.2% decline — the count contraction is broader than just the new-award figures above.

Mechanically, the pattern is: applications are rising (more researchers competing for the same roughly-flat total RPG budget), the number of awards NIH can make is shrinking, and average award cost is rising — the classic signature of "fewer, bigger" allocation under a budget that is not growing nearly as fast as demand. NIH's own FY2025 release explicitly attributes part of the rising average cost to more awards being "forward funded" (a larger share of a multi-year award's total budget obligated in the first year), which mechanically inflates the average award-size figure without necessarily reflecting bigger science.

Step 2 — Removing inflation changes the headline (this corpus's deflation check)

Per this corpus's own guidance on multi-year dollar comparisons, the nominal RPG-size trend was deflated to constant 2025 dollars using adjust_inflation (BLS CPI-U, CUUR0000SA0) before being reported as a finding:

Net 2020→2025: +19.2% nominal, but −4.2% in real dollars. The real figure actually bottomed in FY2024 and only partially recovered by FY2025. CPI-U is a general consumer-price index, not a research-cost index — NIH itself uses a Biomedical Research and Development Price Index (BRDPI) for internal budget planning, which historically runs at or above CPI-U because personnel, equipment, and clinical-trial costs inflate faster than the general basket. BRDPI is not available as a series in this corpus, so this is a CPI-U-based estimate; if BRDPI runs higher (its typical pattern), the real decline in average award size would be larger than −4.2%, not smaller. This is flagged as a genuine limitation, not resolved here.

Step 3 — NSF corroborates the direction, but its newest year complicates it

Independent, non-NIH corroboration comes from Grant Witness (grantwitness.org), a research-funding tracking project, fetched directly for this report. NSF's average award size rose roughly 15% from about $390,000 (FY2024) to $450,000 (FY2025), while new research-project-grant awards fell about 20% (roughly 11,000 → 8,800) — the same fewer/bigger signature as NIH.

However, NSF's FY2026 (partial-year, through May) data breaks that pattern: the mean award size is now about $100,000 smaller than the prior five-year average, because NSF made no $30M+ mega-awards this year, while the median award size is larger, because NSF cut small awards and fellowships (e.g., I-Corps). NSF is also forward-funding new awards — obligating a larger share of a grant's total budget up front rather than doling it out via multi-year Continuing Grants — which shrinks the visible count of "new awards" per year without giving researchers any more money per year. This is the same forward-funding mechanism NIH cites for part of its own average-size increase. Net read: at both agencies, part of the apparent "bigger awards" trend is a budget-accounting artifact of funding timing, not necessarily bigger grants in real terms.

Step 4 — Size distribution today: right-skewed, moderately concentrated

To characterize the current shape of the distribution (not a trend, since only one fiscal year of award-level microdata is loaded in this corpus — see limitation below), NIH RePORTER award microdata (research.nih_award_projects, FY2026, n=59,813 awards, $35.33B total) was queried directly. The distribution is strongly right-skewed: mean award size $590,685 exceeds the median of $442,688; the 10th percentile is $98,441 and the 99th percentile is $3,889,954, a roughly 40x spread. The Gini coefficient of award-dollar concentration is 0.476 (moderate concentration), with the top 10% of awards by size accounting for 37.8% of all dollars awarded. This snapshot is consistent with a grants system where most awards are modest R01-scale grants and a smaller number of large center/program awards carry a disproportionate dollar share — the same concentration dynamic that a shift toward "fewer, bigger" awards would tend to deepen over time, though this corpus cannot yet measure that shift directly (see below).

Data limitation

research.nih_award_projects, the AskAmerica table purpose-built to answer "is award size/count shifting toward fewer-bigger or more-smaller grants," currently holds only one fiscal year of data (FY2026, 59,813 rows) despite its declared 2022-2026 coverage window — confirmed via this session's own data_coverage call, which showed FY2022-2025 as "missing_vs_declared." This table therefore could not be used to compute the actual multi-year trend directly from warehouse microdata; it was used only for the single-year distribution/concentration analysis in Step 4. The multi-year trend in Steps 1-3 instead comes from NIH's and Grant Witness's own published aggregate tables, fetched directly via web_fetch and cited above. This is a genuine corpus coverage gap, not a data-quality issue with the years that are loaded.

Every query behind this report

3 warehouse calls ran in this session, in order. Each is reproducible against the same snapshot.

SELECT fiscal_year, COUNT(*) AS n_awards, SUM(award_amount) AS total_amt, AVG(award_amount) AS avg_amt, median(award_amount) AS median_amt FROM research.nih_award_projects WHERE award_amount > 0 GROUP BY fiscal_year ORDER BY fiscal_year

SELECT
  quantile_cont(award_amount, 0.10) AS p10,
  quantile_cont(award_amount, 0.25) AS p25,
  quantile_cont(award_amount, 0.50) AS p50,
  quantile_cont(award_amount, 0.75) AS p75,
  quantile_cont(award_amount, 0.90) AS p90,
  quantile_cont(award_amount, 0.99) AS p99,
  AVG(award_amount) AS mean_amt,
  stddev_samp(award_amount) AS sd_amt,
  COUNT(*) AS n
FROM research.nih_award_projects
WHERE award_amount > 0

SELECT award_amount AS value_col FROM research.nih_award_projects WHERE award_amount > 0