A translucent green-blue boundary encloses five coloured miniature building groups while many grey buildings remain outside.
Conceptual bounded-sample illustration, not a company census. The five depicted groups do not represent either study count; leaf markers are not certification.
Diagram text description

A pale synthetic miniature urban and industrial landscape contains a translucent irregular green-blue boundary around five coloured building groups with leaf markers. Grey buildings and trees continue outside the boundary. There are no visible dates, company names, sample sizes, registry IDs or approval text. The outside landscape is uncounted, not established non-buying activity, and the leaves do not prove credit integrity or climate performance.

The research reviewed here does not establish a verified worldwide total of companies currently buying carbon credits. It does establish narrower findings: a May 2024 working paper identified 866 publicly listed corporations in historical retirement records, while Carbon Brief found credit use by 34 members of a selected 50-company cohort during 2020–2022. Neither figure is a global 2026 buyer count.

The distinction matters when a number enters a procurement presentation. A count of identifiable retirement users is not necessarily a count of contract payers, and a sample chosen for company size and climate targets does not describe all businesses. Undisclosed identities and inconsistent coverage also prevent a simple census from emerging by adding registry rows.

The useful answer is therefore conditional: which companies, which event, which observation window and which matching rule? The comparison below shows what these studies can support, where their evidence stops and how a buyer can use the findings without turning a bounded observation into a market-size claim.

The short answer: no verified global buyer count

A credit quantity, a retirement event and a company count are different measurements. One retirement can cover many units; one company can appear in many retirements. Another business may pay an intermediary while the registry names a beneficiary. Before counting organisations, decide which of those roles the question actually concerns.

Record or roleWhat it describesWhy it is not automatically a unique buyer
Credit unitsQuantity associated with an instrument or eventOne company may use very different quantities from another
Retirement eventA dated use of specified unitsThe same entity can have repeated events
Contract payerParty responsible for the purchase paymentIt may procure on behalf of someone else
BeneficiaryNamed recipient of the retirement's benefit or purposeIt need not be the payer or the registry account holder
Unique companyEntity counted under a stated identity ruleNames, subsidiaries and mergers need consistent treatment

Table 1, shown one row at a time. Every field is retained.

Row 1

Record or role
Credit units
What it describes
Quantity associated with an instrument or event
Why it is not automatically a unique buyer
One company may use very different quantities from another

Row 2

Record or role
Retirement event
What it describes
A dated use of specified units
Why it is not automatically a unique buyer
The same entity can have repeated events

Row 3

Record or role
Contract payer
What it describes
Party responsible for the purchase payment
Why it is not automatically a unique buyer
It may procure on behalf of someone else

Row 4

Record or role
Beneficiary
What it describes
Named recipient of the retirement's benefit or purpose
Why it is not automatically a unique buyer
It need not be the payer or the registry account holder

Row 5

Record or role
Unique company
What it describes
Entity counted under a stated identity rule
Why it is not automatically a unique buyer
Names, subsidiaries and mergers need consistent treatment

An analysis of retirements can reveal observed use even where it cannot reconstruct the original commercial transaction. Credits acquired now but held for later use may not yet appear as retirements at all. A future delivery commitment is a different event again.

This is why the headline question needs a definition of buying. For a retirement-based study, describe the companies identified through retirement evidence. Do not quietly broaden that result into all purchasing organisations or treat every absence from the records as proof of non-participation.

What one 866-company research sample actually measured

The May 2024 working paper by Sehoon Kim, Tao Li and Yanbin Wu starts from retirement records, then identifies listed firms. Its data section describes four registries: American Carbon Registry, Climate Action Reserve, Gold Standard and Verra's Verified Carbon Standard. The dataset assembled in December 2022 contained nearly 279,000 retirement entries covering 2005–2021.

Researchers manually extracted firm names from the entries and matched them to GVKEY identifiers in Compustat North America and Global. That process yielded 866 unique public corporations from 46 countries. The diagram shows the filtering logic, not the relative sizes of every intermediate group or an independently recreated dataset.

Four registry sources lead to identifiable retirement records, then matching yields 866 listed firms in a bounded historical sample.
Kim, Li and Wu, May 2024 draft: 2005–2021 retirements, dataset assembled in December 2022, matched to Compustat identifiers. The 866 listed firms are not a worldwide 2026 buyer count. Box sizes do not encode quantities.
Diagram text description

The first box names the four-registry source frame, not four corporate buyers. Named records pass through manual firm-name and Compustat GVKEY matching to the 866 unique listed corporations from 46 countries reported by the study. No intermediate group size is inferred. Unnamed identities and private firms are outside the listed-firm match; purchasing, beneficiary and claim roles are not collapsed into a universal payer identity.

The result is a count of identifiable listed firms within those sources and dates. It is not a headcount of every private business, intermediary, undisclosed beneficiary or purchaser holding unretired credits. The paper's footnote specifically explains that undisclosed retirement identities cannot be linked to listed firms through its matching procedure.

Dates belong to different layers. The research draft is dated May 2024, the underlying retirement window ends in 2021, and the data assembly point is December 2022. Reading the paper in 2026 does not extend that window. Nor does matching to a corporate identifier prove that every contract payer or all subsidiaries were independently audited here.

The paper provides evidence about a defined observed corporate sample, not a sampling fraction from which this article can estimate a worldwide total. Its data-method pages were consulted on 31 October 2026 (UTC).

Why the 34-of-50 example is not a market share

Carbon Brief's September 2023 analysis asks a different question about a deliberately selected group. Its methodology describes 50 large publicly listed companies with net-zero or similar pledges, selected using Net Zero Tracker and ranked by revenue. It identifies credit use by 34 of those companies during 2020–2022.

There is a source-description wrinkle: an introductory sentence refers instead to market capitalisation. The detailed methodology repeatedly specifies revenue. The comparison here uses that methodological description while retaining the discrepancy; reconstructing the exact original ranking would require the dated underlying cohort list, not a guess about which sentence was intended.

The arithmetic 34 divided by 50 gives 68% within this selected cohort. It is not the share of all companies buying credits, a share of global credit volume or a current participation rate. The group is intentionally weighted towards large listed businesses with announced climate goals, rather than selected as a representative sample of every business.

The remaining 16 must not all be labelled confirmed non-buyers. Incomplete disclosure and anonymity can prevent observed use from being attributed. Carbon Brief also distinguishes retirement in its study period from commitments whose credits would be used later. These limitations do not make the observed 34 meaningless; they specify the narrower question that number answers.

Why retirement volume cannot be converted into a company count

There is no fixed conversion between credit volume and the number of unique companies. A high volume can reflect a few large users or many small ones, while a stable company count can accompany changes in the quantities used. The two measurements need separate denominators and cannot be substituted for each other.

Consider a fictional illustration, not a study result. Two retirement entries contain 100 and 200 units, both attributable to the same identified company under the chosen matching rule. The visible records then show two events, 300 units and one identified company. Adding another 50-unit retirement with no disclosed identity increases visible volume to 350 units, but does not establish a second company.

That undisclosed record might concern the existing company, another company or a different kind of organisation. Guessing one new buyer per anonymous row would invent identities. Conversely, assigning all anonymous units to known companies would invent a different answer from the same incomplete evidence.

Even a correctly calculated average quantity per observed firm cannot safely fill the gap. Named and unnamed activity need not have the same distribution, and inclusion in the named group may depend on company size or disclosure practices. Report volume and identifiable entities separately, and leave the unobserved identity component unestimated unless a defensible method and supporting evidence exist.

Are more companies buying, or are repeat buyers doing more?

Repeat participation and broader participation can occur together. Existing buyers may retire more units while additional identities first appear in a tracking dataset. A rise in volume concentration does not logically mean the company count fell; a longer list of visible names does not necessarily mean every name represents a genuinely new entrant.

AlliedOffsets' 2026 commentary reports recurring large buyers in its tracked retirements and offtakes, alongside an increase in distinct buyers over the periods it discusses. That is a useful observation about its tracked activity. The article does not supply a complete worldwide buyer denominator or a reproducible global census that could replace the two historical studies here.

A ranked list answers who accounts for the most observed volume, not how many organisations participated outside the ranking. Its event mix matters too: an offtake commitment and a completed retirement need not occur in the same year or represent the same stage of a procurement decision.

When discussing change, separate first observed in a dataset from first ever purchased. Expanded registry coverage, better name matching, a merger or newly disclosed information can alter the visible population. A genuine new-entry claim requires a stable observation rule and sufficient history, not simply a name absent from last year's top-buyer list.

How to compare two buyer-count claims

Compare the observation contracts before comparing the totals. In the table, “observed” means supported by the source's stated procedure, not independently re-audited raw records. The two studies have different routes into their populations and do not provide the common denominator needed to pool their results.

DimensionKim, Li and Wu working paper, May 2024Carbon Brief analysis, September 2023
Finding and population866 unique listed corporations from 46 countries matched from registry records34 observed users within a selected cohort of 50 large listed companies with net-zero or similar targets
Observation periodRetirements in 2005–2021; dataset assembled in December 2022Retirements from 1 January 2020 to 31 December 2022; cohort entries mostly updated around late 2022 or early 2023
Event and source frameRetirement entries from ACR, CAR, Gold Standard and VCSAlliedOffsets retirement matching across 20 voluntary registries, supplemented by CDP and selected academic findings
Identity ruleManual firm-name extraction and matching to Compustat North America/Global GVKEYsMatching to the named 50-company cohort; name changes and mergers are acknowledged in the methodology
Deduplication informationUnique corporate identifiers anchor the reported firm count; this article has not independently audited the underlying crosswalkCDP adds only retirement quantities exceeding the corresponding provider values; both names may be listed after changes or mergers. This is not a disclosed universal payer-level deduplication audit
Missingness and exclusionsUnnamed entities cannot be matched by that procedure; private firms fall outside its listed-firm frameCompletely anonymous buyers remain unidentifiable; gaps in company/provider reporting and unavailable 2022 CDP data limit coverage
Supported interpretationIdentifiable listed-firm use in that historical registry windowObserved use within that selected climate-target cohort and three-year window

Table 2, shown one row at a time. Every field is retained.

Row 1

Dimension
Finding and population
Kim, Li and Wu working paper, May 2024
866 unique listed corporations from 46 countries matched from registry records
Carbon Brief analysis, September 2023
34 observed users within a selected cohort of 50 large listed companies with net-zero or similar targets

Row 2

Dimension
Observation period
Kim, Li and Wu working paper, May 2024
Retirements in 2005–2021; dataset assembled in December 2022
Carbon Brief analysis, September 2023
Retirements from 1 January 2020 to 31 December 2022; cohort entries mostly updated around late 2022 or early 2023

Row 3

Dimension
Event and source frame
Kim, Li and Wu working paper, May 2024
Retirement entries from ACR, CAR, Gold Standard and VCS
Carbon Brief analysis, September 2023
AlliedOffsets retirement matching across 20 voluntary registries, supplemented by CDP and selected academic findings

Row 4

Dimension
Identity rule
Kim, Li and Wu working paper, May 2024
Manual firm-name extraction and matching to Compustat North America/Global GVKEYs
Carbon Brief analysis, September 2023
Matching to the named 50-company cohort; name changes and mergers are acknowledged in the methodology

Row 5

Dimension
Deduplication information
Kim, Li and Wu working paper, May 2024
Unique corporate identifiers anchor the reported firm count; this article has not independently audited the underlying crosswalk
Carbon Brief analysis, September 2023
CDP adds only retirement quantities exceeding the corresponding provider values; both names may be listed after changes or mergers. This is not a disclosed universal payer-level deduplication audit

Row 6

Dimension
Missingness and exclusions
Kim, Li and Wu working paper, May 2024
Unnamed entities cannot be matched by that procedure; private firms fall outside its listed-firm frame
Carbon Brief analysis, September 2023
Completely anonymous buyers remain unidentifiable; gaps in company/provider reporting and unavailable 2022 CDP data limit coverage

Row 7

Dimension
Supported interpretation
Kim, Li and Wu working paper, May 2024
Identifiable listed-firm use in that historical registry window
Carbon Brief analysis, September 2023
Observed use within that selected climate-target cohort and three-year window

Deduplicating credit quantities across sources and deduplicating companies are separate tasks. Carbon Brief's rule for adding only excess CDP-reported quantities helps avoid simply summing overlapping volume reports; it does not establish that an arbitrary list of corporate name variants can be counted without a crosswalk. The working paper's GVKEY matching supplies a stated company-identity anchor, but not a guarantee about every possible purchasing role.

For cohort selection, Carbon Brief's methods say revenue while an introductory sentence says market capitalisation. The exact ranking reconstruction remains unresolved here. If a decision depends on reproducing which firms entered the 50, obtain the dated underlying list first; do not resolve that question by silently widening the denominator to all large companies.

Do not add the 34 to the 866: the populations may overlap, their periods overlap only partly, and their inclusion rules differ. Do not infer a growth rate by placing the findings in publication-date order. Neither study supplies the complete global population, unknown identities or inclusion probabilities needed to turn these two observations into a worldwide estimate.

A reusable comparison therefore asks five things: which event, which sample and dates, which identity rule, which missing records, and which deduplication method? If any differs, explain the effect before quoting a difference in totals. Some methodological details remain unverified without access to the underlying matched datasets; that limit belongs beside the figures, not hidden after a market-size headline.

What a buyer can use these figures for

Use the 866 listed-firm registry match from 2005–2021 differently from the 34 observed users in a selected 50-company climate-target cohort during 2020–2022. The former depends on identifiable registry names matched to corporate identifiers and excludes private firms from its frame; the latter starts with a chosen company list and combines several retirement-evidence sources. Neither resolves completely anonymous identities. These findings can help evaluate evidence, not decide that a business must buy because its peers do. They establish neither the suitability nor integrity of any credit offered today.

For example, a procurement team receiving a pitch that “most companies already buy” can ask which cohort and years support it. If the answer is 34 of 50 selected climate-target companies in 2020–2022, the team has a specific observation to discuss—not a general adoption rate for its own sector, country or company size.

Likewise, a demand forecast needs its own assumptions about eligible buyers, future participation and quantities per buyer. A list of recurring high-volume names cannot supply all three. Retaining the sample limit prevents a historical retirement result from becoming an unsupported sales forecast or a substitute for a procurement policy.

Use the comparison to improve the questions asked of a dataset or adviser: does it observe contracts or use, can identities be resolved, and what activity remains invisible? The answer may support cautious benchmarking within a comparable group. It does not by itself approve a climate claim or demonstrate that purchasing credits reduced the buyer's own operational emissions.

Keep payer, beneficiary and claim evidence distinct

Keep the evidence for your own purchase traceable without claiming to enumerate the entire market. A useful record distinguishes the contract payer, registry account or retiring party, named beneficiary and owner of any public claim. These may coincide, but they should not be assumed to be the same merely because a single organisation's name appears on a certificate.

In a fictional procurement handoff, an agency pays on behalf of a manufacturing client, the retirement names that client as beneficiary, and a separate marketing team proposes a public statement. Connect the invoice, instructions, registry and serial references, date, quantity, retirement evidence and proposed claim to their actual roles. Do not count the agency and beneficiary as two end-user companies solely because two names occur in the file.

Keep company emissions records distinct from the credit transaction and claim evidence. Storage in a platform or an internally complete folder does not independently verify the units, certify climate performance or establish the acceptability of the proposed claim. Generic evidence organisation is not approval of a named supplier's capabilities.

Where the payer-beneficiary relationship is unclear, leave that relationship open and obtain the relevant instruction or contract. This is a narrower, achievable task than producing a universal buyer census, and it makes subsequent review of the actual procurement decision more useful.

Conclusion: ask which companies, in which sample and year

The defensible answer is not one convenient worldwide number. The reviewed evidence identifies 866 listed firms through a historical registry-matching procedure and 34 observed users within a selected 50-company cohort. Their dates, populations and identity rules are part of those findings, not optional footnotes.

Keep purchase, commitment, retirement, beneficiary and unique company distinct. Unknown identities cannot be converted into buyers by averaging volumes, and overlapping studies cannot be added without a compatible population and entity-level reconciliation.

Before using a buyer count, ask which event was counted, which companies were eligible, which years were observed and how names and missing records were handled. Then quote the bounded conclusion the evidence supports. For your own procurement, preserve the roles and records behind the decision; a market statistic is context, not evidence that a particular credit or public claim is suitable.