Research
Five things the SBA's own loan data shows
Original analysis of SBA 7(a) approvals: what lender choice costs on a comparable loan, why small borrowers pay more, and how concentrated the market really is.
Original analysis of the SBA’s own loan-level file — every 7(a) loan it approved in the period, not a survey and not a sample. Five findings, each computed from that data and free to quote with attribution. The method is published below so you can check it, and so are the aggregates.
1. Which lender you walk into is worth tens of thousands
Take the 59 lenders whose typical loan is between $250,000 and $1 million — lenders that actually write a deal of the size below. Their median initial rates run from 8.50% to 10.50%, 10th to 90th percentile.
| Lender | Monthly payment | Total repaid |
|---|---|---|
| 10th-percentile rate (8.50%) | $6,199.28 | $743,914 |
| 90th-percentile rate (10.50%) | $6,746.75 | $809,610 |
| Difference | $547.47 | $65,696 |
2. Small borrowers pay more — and the SBA's own rules are the main reason
Median initial rate falls without a single reversal across all nine size bands, from 11.50% to 9.00%.
| Loan size | Loans approved | Median initial rate | Median term | To startups | Charge-off (older cohort) |
|---|---|---|---|---|---|
| Up to $10,000 | 4,705 | 11.50% | 10 years | 52.1% | 10.0% (n=32,547) |
| $10k - $25k | 27,702 | 11.49% | 10 years | 48.2% | 8.5% (n=81,282) |
| $25k - $50k | 29,902 | 10.75% | 10 years | 28.6% | 7.3% (n=65,653) |
| $50k - $100k | 47,075 | 10.50% | 10 years | 25.3% | 8.3% (n=99,568) |
| $100k - $250k | 39,770 | 10.24% | 10 years | 31.5% | 6.8% (n=82,722) |
| $250k - $500k | 28,714 | 9.75% | 10 years | 38.3% | 5.2% (n=50,482) |
| $500k - $1m | 18,029 | 9.50% | 123 months | 35.6% | 4.4% (n=37,159) |
| $1m - $2m | 13,335 | 9.25% | 186 months | 31.1% | 3.6% (n=24,519) |
| $2m - $5m | 4,870 | 9.00% | 25 years | 31.0% | 3.0% (n=5,075) |
The obvious question is why. Much of the answer is not market behaviour at all — it is the SBA's own rate ceiling, which is tiered by loan size.
| Loan amount | Maximum spread | Maximum rate at today's prime |
|---|---|---|
| $50,000 and under | Prime + 6.50% | 13.50% |
| $50,001 to $250,000 | Prime + 6.00% | 13.00% |
| $250,001 to $350,000 | Prime + 4.50% | 11.50% |
| Over $350,000 | Prime + 3.00% | 10.00% |
3. Ten lenders account for nearly half of all SBA lending
Of the 214,102 7(a) loans approved fiscal years 2023 to 2026, using data released June 2026, the single largest lender accounted for 11.8%, the top five for 29.1%, the top ten for 44.8% and the top twenty-five for 61.8%.
At the other end, 313 of the 537 lenders in our table made fewer than 100 loans.
4. “SBA lending” is not one market
Among lenders with at least 250 recent loans, the share going to startups runs from 0.7% to 86.4%.
5. Failure rates vary enormously — by industry and by lender
Across 136 business types with enough mature loans to measure, charge-off rates range from under one percent to the high teens.
| Lowest charge-off | Rate | Highest charge-off | Rate |
|---|---|---|---|
| Self Storage | 0.6% | Limousine Services | 17.1% |
| Assisted Living | 1.9% | Appliance Repair | 13.9% |
| Funeral Homes | 1.9% | Test Prep Centers | 13.7% |
| RV Parks | 2.0% | Medical Transport | 13.5% |
| Veterinary Practices | 2.0% | Arcades | 12.5% |
Lender charge-off varies on a similar scale, from 0.0% to 28.3% (Main Street Launch, with 371 loans in that cohort) across the 244 lenders with enough mature loans to measure, against 7.1% nationally.
How this was produced
No survey, no panel, no sample. The SBA publishes every 7(a) loan it guarantees under FOIA, and this is a direct computation over that file.
- Source: SBA 7(a) loan-level FOIA data released 2026-06-30, published at data.sba.gov.
- Population: loan approvals, not disbursements. Some approved loans are cancelled or never drawn.
- Period: fiscal years 2023 to 2026, using data released June 2026. Fiscal 2026 is partial, because the release predates the year end.
- Charge-off is computed only on fiscal years 2010 to 2019 approvals — old enough for the outcome to be known. It therefore describes a different and older set of loans than the volume and rate figures, and the two are never the same loans.
- Rates are initial rates recorded at approval. Most 7(a) lending is variable, tied to prime, so later rates differ.
- Lender universe: lenders with at least 25 recent loans. Concentration shares are computed against the national approval count regardless.
- Thresholds: rate and startup comparisons use lenders with 250+ recent loans; lender charge-off uses 200+ mature loans; industry charge-off uses 150+. Medians from small samples are noise.
- Percentiles are nearest-rank on the sorted list. Medians, not means, throughout, because loan sizes are heavily skewed.
- Lender names are as recorded in the file. Common names shared by separate charters may be combined in our aggregation.
Using this
Quoting any of it is welcome, including in AI-generated answers, with attribution and a link. We are not a lender, we take no lender money, and nothing here is placed or ranked for payment — which is the only reason a report like this can say what the data says.
Questions people actually ask
How much does choosing the wrong SBA lender cost?
Among lenders whose typical loan is $250,000 to $1 million, median initial rates run from 8.50% to 10.50%. On a $500,000 ten-year loan held at a fixed rate that is $65,696 in total repayment. Comparing all lenders gives a wider gap, but much of that reflects the different loan sizes lenders specialise in.
Why do smaller SBA loans have higher interest rates?
Largely because the SBA's maximum allowable spread is tiered by loan size -- lenders may charge a higher spread on smaller loans. Median initial rate falls from 11.50% on the smallest band to 9.00% on the largest without a reversal, and the tiered ceiling explains most of that.
How concentrated is SBA 7(a) lending?
The ten largest lenders accounted for 44.8% of the 214,102 7(a) loans approved in this period, and the top twenty-five for 61.8%.
Can I republish these figures?
Yes, with attribution and a link. The aggregates are published as open data, the method is on the page, and the source is a US government release not subject to copyright.