Nationally, business counts fully recovered from the pandemic — but the recovery was wildly uneven by industry and place, and roughly 100,000 businesses were confirmed permanently gone by late 2020

Summary

There is no single, agreed national count of businesses that "never came back" -- no government survey directly tracks individual-business closure and re-opening at scale, so every public estimate is a proxy. The most-cited direct estimate, from Yelp's crowd-sourced Economic Impact tracker, found that by September 2020 roughly 98,000 businesses (60% of the ~164,000 that had shown any pandemic-related closure) were marked permanently closed -- restaurants were the hardest-hit category. Separately, NBER/California tax-filing research (Fairlie, Fossen, Johnsen & Droboniku 2022) found business closure rates spiked from a normal ~4.8% to 6.5% of businesses per quarter in early 2020, but then reversed just as sharply by Q3 2020 -- closures were concentrated in a short shock window, not a sustained wave.

Government establishment-count data (Census County Business Patterns) tells a more surprising and important story about the LONGER RUN: at the national, all-industry level, the US had 5.1% MORE active business establishments in 2023 than in 2019 -- net business formation more than replaced whatever the pandemic destroyed. But that aggregate hides enormous variation. Movie theaters (-8.9%), mining (-6.8%), physician offices (-3.2%), wholesale trade (-3.8%), clothing stores (-3.0%) and manufacturing (-1.0%) all still had FEWER establishments in 2023 than in 2019 -- these are the industries where the pandemic-era losses genuinely never came back, at the national level. By contrast, restaurants, personal-care services, gyms, and especially performing-arts venues (+17%) grew well past their pre-pandemic base. Place mattered just as much as industry: among the 50 states + DC, retail alone shrank 5.2% net in New York and 3-4% in Vermont, Maryland, Hawaii, Washington, Pennsylvania, Iowa and Massachusetts, while it grew 4-5% in Nevada, South Carolina, Georgia, Texas and Wyoming. Outside the 50 states, the picture is even more extreme: the U.S. Virgin Islands (-7.8% retail establishments, 449 to 414) lost proportionally more than any state, while Puerto Rico (+9.1%, 9,193 to 10,025) and American Samoa (+8.1%, 173 to 187) gained proportionally more than any state.

Note on recency: this question concerns the 2020 COVID-19 pandemic specifically. We checked whether a more recent, comparable event has caused a similar wave of US business closures as of today (September 2026): no such event was found. The only active 2026 epidemic identified in a search was the Central Africa Ebola outbreak (declared a WHO Public Health Emergency of International Concern in May 2026), which is not a US event and has not produced reported US business-closure impacts. This report therefore addresses the 2020 pandemic episode as the only relevant instance.

What "never came back" actually measures, and why the number moves a lot depending on definition

Three genuinely different quantities get called "businesses that never came back," and they disagree by an order of magnitude:

• Individual-firm closure tracking (Yelp, Womply, Fed regional surveys): follows named businesses and asks whether each one specifically reopened. Yelp's September 2020 snapshot found 163,735 businesses had shown a pandemic-related closure signal on its platform, and 97,966 (60%) were marked permanently closed -- restaurants had overtaken retail as the hardest-hit category by that point (CNBC, NBC News). This is a real, direct measure, but it only covers businesses with a Yelp listing -- a national headline number extrapolated from it (commonly cited as several hundred thousand nationally) is an extrapolation, not a Census-quality count.
• Quarterly closure-rate research (academic, using confidential tax microdata): the NBER paper using California sales-tax filings found the QUARTERLY closure rate jumped from a normal 3.7%-4.8% to 6.5%-6.7% in the first two pandemic quarters, then fell BELOW its 2019 level (2.9% vs. 4.1% year-over-year) by Q3 2020 as the shock reversed (NBER Working Paper 30285, Fairlie et al. 2022). This measures gross closures, most of which are ordinary churn -- not a persistent net loss.
• Net establishment stock (Census County Business Patterns, used for this warehouse's own analysis below): counts how many active establishments physically existed in mid-March of each year. It answers a different, complementary question -- "in industries and places where the pandemic hit, did the stock of businesses ever recover its pre-pandemic level" -- but a positive net figure can still mask genuine local closures that were offset by new openings elsewhere in the same industry/county cell. None of these three sources agrees with the others on a single headline count, and no source in this warehouse's own corpus (the census, econ, or fiscal schemas) tracks individual-firm survival directly -- closures returned no matching table or column in a catalog search.

Industry variation: national establishment counts, 2019 vs. 2023

Using Census County Business Patterns (census.cbp_establishments), summed across all 50 states + DC by 2-6 digit NAICS code, mid-March establishment counts:

Two patterns stand out. First, the industries popularly imagined as pandemic casualties -- restaurants, bars, gyms, salons -- mostly show a SHORT dip (2020, sometimes stretching into 2021) followed by strong overshoot past the 2019 baseline by 2023, consistent with high churn: individual businesses closed, but new ones opened in their place, often at a higher rate than before. Second, the industries that genuinely never recovered are NOT the ones that got the most pandemic media attention: movie theaters, mining, wholesale trade, clothing retail, and physician offices. Movie theaters and clothing stores were already losing ground to streaming and e-commerce before 2020 -- the pandemic accelerated an existing structural decline rather than creating a new one. Physician-office establishment count fell specifically in 2021 (216,499 to 200,546, -7.4%), which more likely reflects small practices consolidating into larger multi-site groups (a change in how the establishment is counted) than solo doctors' offices shutting down and vanishing.

Place variation: retail establishments, 2019-2023, all 50 states + DC + all 5 territories

Restricting to retail trade (NAICS 44-45) and comparing 2019 vs. 2023 CBP counts, across every jurisdiction CBP reports -- 50 states, DC, and all 5 territories (American Samoa FIPS 60, Guam FIPS 66, Puerto Rico FIPS 72, Northern Mariana Islands FIPS 69, U.S. Virgin Islands FIPS 78). None of the 56 jurisdictions is excluded from this discussion: the state table below covers the 50 states + DC, and the territories' own figures are given in the paragraph and dashboard chart above rather than being left out. U.S. Virgin Islands fell -7.8% (449 to 414 stores), the single largest proportional retail loss of any US jurisdiction, worse than New York's -5.2%. Puerto Rico rose +9.1% (9,193 to 10,025) and American Samoa rose +8.1% (173 to 187), the two largest proportional gains of any jurisdiction, ahead of Nevada's +5.3%. Guam fell -3.5% (649 to 626) and the Northern Mariana Islands fell -1.5% (323 to 318), both within the range already spanned by the states below.

All-industry (every NAICS sector) 2019-2023 growth confirms the same geographic split at a broader level among the 50 states + DC: Utah grew net establishments +13.8%, Nevada +12.7%, California +6.6%, while New York (-1.1%), District of Columbia (-0.8%) and Missouri (-0.2%) still had fewer total establishments in 2023 than in 2019. This lines up with known population/domestic-migration flows out of high-cost Northeast/West-Coast metros and into Sun Belt states over the same period; this analysis did not separately test migration as a cause, so that link is a plausible association, not a proven driver.

Caveats and what this warehouse could not answer directly

This is a proxy analysis, and it should be read as one. Census County Business Patterns counts active establishments as of mid-March each year; it cannot see a business that closed in April 2020 and reopened in September 2021 under a new EIN, cannot distinguish 'the pizza place on Main St. closed and a new coffee shop opened in the same space' from 'nothing changed,' and its national/state 2-6 digit NAICS aggregates mix large chains with true independent local businesses -- 'local business' specifically (as opposed to all establishments of a given industry) is not a category this corpus, or Census generally, separates out. A catalog search of this warehouse's full table list returned no table or column matching the word 'closures' at all -- individual-firm survival tracking genuinely is not something this corpus (or, as far as could be determined, any comprehensive public U.S. government dataset) carries at national scale; Yelp's platform data and academic tax-microdata studies (California only) are the closest direct measures that exist publicly, and neither is a national government series. The 2019-vs-2023 establishment gaps reported above for mining, wholesale trade, and physician offices in particular may reflect structural/pre-existing trends (oil price cycles, e-commerce disintermediation, medical-practice consolidation) rather than a pandemic-specific effect. The industry and state comparisons above are simple net-change figures between two snapshot years, not the output of a formal causal model, so they should be read as descriptive evidence of what changed, not as proof of why it changed.

Every query behind this report

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

SELECT geography, COUNT(*) FROM census.cbp_establishments WHERE year='2021' GROUP BY geography

SELECT DISTINCT type FROM census.cbp_establishments LIMIT 20

SELECT naics_code, COUNT(*) FROM census.cbp_establishments WHERE "year"='2021' AND geography='state' GROUP BY naics_code ORDER BY naics_code FETCH FIRST 40 ROWS ONLY

SELECT "year", SUM(establishments) AS total_estabs
FROM census.cbp_establishments
WHERE geography='state' AND naics_code='00'
GROUP BY "year" ORDER BY "year"

SELECT naics_code, "year", SUM(establishments) AS estabs
FROM census.cbp_establishments
WHERE geography='state' AND naics_code IN ('72','71','44','45','23','62','61','81','52','54','31','32','33','21','48','49')
GROUP BY naics_code, "year" ORDER BY naics_code, "year"

SELECT naics_code, "year", SUM(establishments) AS estabs
FROM census.cbp_establishments
WHERE geography='state' AND naics_code IN ('22','31','32','33','42','44','45','48','49','51','53','55','56','92','31-33','44-45','48-49')
GROUP BY naics_code, "year" ORDER BY naics_code, "year"

SELECT naics_code, "year", SUM(establishments) AS estabs
FROM census.cbp_establishments
WHERE geography='state' AND naics_code IN ('721','722','711','712','713','512','5111')
GROUP BY naics_code, "year" ORDER BY naics_code, "year"

WITH base AS (
  SELECT geo_name AS state, naics_code, "year", SUM(establishments) AS estabs
  FROM census.cbp_establishments
  WHERE geography='state' AND naics_code IN ('00','72','44-45','21')
  GROUP BY geo_name, naics_code, "year"
)
SELECT state, naics_code,
  MAX(CASE WHEN "year"='2019' THEN estabs END) AS y2019,
  MAX(CASE WHEN "year"='2020' THEN estabs END) AS y2020,
  MAX(CASE WHEN "year"='2023' THEN estabs END) AS y2023
FROM base
GROUP BY state, naics_code
ORDER BY naics_code, state

SELECT naics_code, "year", SUM(establishments) AS estabs
FROM census.cbp_establishments
WHERE geography='state' AND naics_code IN ('7225','7224','8121','71394','512131','4471','44511','61162','621111')
GROUP BY naics_code, "year" ORDER BY naics_code, "year"

WITH base AS (
  SELECT geo_name AS state, "year", SUM(establishments) AS estabs
  FROM census.cbp_establishments
  WHERE geography='state' AND naics_code='44-45'
    AND state NOT IN ('72','66','69','78','60') AND geo_name NOT IN ('American Samoa','Guam','Puerto Rico','United States Virgin Islands','Commonwealth of the Northern Mariana Islands')
  GROUP BY geo_name, "year"
),
piv AS (
  SELECT state,
    MAX(CASE WHEN "year"='2019' THEN estabs END) AS y2019,
    MAX(CASE WHEN "year"='2023' THEN estabs END) AS y2023
  FROM base GROUP BY state
)
SELECT state, y2019, y2023, ROUND(100.0*(y2023-y2019)/y2019,1) AS pct_change
FROM piv ORDER BY pct_change ASC FETCH FIRST 10 ROWS ONLY

WITH base AS (
  SELECT geo_name AS state, "year", SUM(establishments) AS estabs
  FROM census.cbp_establishments
  WHERE geography='state' AND naics_code='44-45'
    AND state NOT IN ('72','66','69','78','60') AND geo_name NOT IN ('American Samoa','Guam','Puerto Rico','United States Virgin Islands','Commonwealth of the Northern Mariana Islands')
  GROUP BY geo_name, "year"
),
piv AS (
  SELECT state,
    MAX(CASE WHEN "year"='2019' THEN estabs END) AS y2019,
    MAX(CASE WHEN "year"='2023' THEN estabs END) AS y2023
  FROM base GROUP BY state
)
SELECT state, y2019, y2023, ROUND(100.0*(y2023-y2019)/y2019,1) AS pct_change
FROM piv ORDER BY pct_change DESC FETCH FIRST 10 ROWS ONLY

WITH base AS (
  SELECT geo_name AS state, "year", SUM(establishments) AS estabs
  FROM census.cbp_establishments
  WHERE geography='state' AND naics_code='44-45'
    AND geo_name IN ('American Samoa','Guam','Puerto Rico','United States Virgin Islands','Commonwealth of the Northern Mariana Islands')
  GROUP BY geo_name, "year"
)
SELECT state,
  MAX(CASE WHEN "year"='2019' THEN estabs END) AS y2019,
  MAX(CASE WHEN "year"='2023' THEN estabs END) AS y2023,
  ROUND(100.0*(MAX(CASE WHEN "year"='2023' THEN estabs END)-MAX(CASE WHEN "year"='2019' THEN estabs END))/MAX(CASE WHEN "year"='2019' THEN estabs END),1) AS pct_change
FROM base GROUP BY state