AI training data funding tracker (81 deals)

Our funding tracker covering the AI training data market

Below is the table listing all the deals. You may find our methodology at the end of this page.

And, if you want to gain a deeper understanding of the market and its current dynamics, get our report covering the AI training data market.

Is the AI training data market actually well-funded, or are a few giant deals making it look bigger?

The AI training data market is a genuinely substantial venture market, with at least $2.04 billion in usable disclosed funding across 81 deals, but a small number of exceptionally large financings make it look much richer than the typical startup experiences.

We identified 66 companies receiving funding between January 1, 2024 and October 8, 2026. The median transaction was $10 million, compared with an average of $27.9 million. A few unusually large investments pull the average well above what most startups actually raise.

The concentration becomes clearer when we remove the biggest rounds. Eight financings above $50 million accounted for 56.5% of disclosed capital. Excluding them leaves $886.3 million, spread across a much broader population of startups.

So there is real depth beneath the headline figures. Dozens of companies are receiving relatively modest amounts, while a small group can attract hundreds of millions.

Our $2.04 billion figure is also conservative. Eight transactions lack a usable numeric amount, including Scale AI's $1 billion financing announced in May 2024. Scale described that transaction as containing both primary and secondary shares, so we cannot count the entire amount as new capital entering the company.

How many AI training data startups are actually getting funded?

We found 66 distinct funded AI training data companies, but most have appeared in the financing market only once during the period.

Of those companies, 54 raised a single qualifying round, nine appeared twice, and three recorded at least three financings. Only 18.2% of companies therefore raised more than once within the window.

That distinction is worth making because 81 financing announcements can suggest a larger ecosystem of independently financed businesses. Several rounds actually belong to the same companies.

Snorkel AI, Encord, micro1, David AI and Protege illustrate the repeat-financing population. They have secured multiple investments while dozens of newer competitors have only one recorded financing event.

We should not interpret the absence of a second round as business failure. Many companies were founded in 2025 or 2026 and have not had enough time to return to investors. Still, the evidence of sustained financing is currently concentrated in a relatively small group.

Who actually holds most of the funding in AI training data?

A small group of AI training data companies controls most of the sector's disclosed capital: the ten largest recipients account for 71.6% of the entire usable financing pool.

Snorkel AI leads with $450 million raised within the period, followed by Mercor with $350 million and micro1 with at least $135 million. Together, these three companies account for almost half of the known capital.

The next group includes Encord, AfterQuery, David AI and Toloka, which have accumulated considerably more funding than the typical entrant.

These figures represent cumulative qualifying financing during our study period, not lifetime funding since incorporation.

Company	Qualifying rounds	Capital raised in window
Snorkel AI	2	$450M
Mercor	1	$350M
micro1	2	≥$135M
Encord	2	$90M
AfterQuery	2	$80.27M
David AI	3	$80M
Toloka	1	$72M
XDOF	1	$70M
Mecka AI	3	$68M
Protege	3	$65M

Are repeat fundraisers taking most of the money?

Yes. Repeat-funded AI training data companies represent only 18.2% of the funded businesses, yet they capture approximately 55.9% of usable disclosed capital.

We find roughly $1.14 billion allocated to companies with multiple qualifying financings, compared with $899 million for companies appearing only once.

Investors are not distributing capital proportionately across the startup population. Larger commitments are going to a narrower group that has already demonstrated an ability to raise institutional funding repeatedly.

Snorkel's $100 million and subsequent $350 million financings illustrate this pattern at the high end. David AI's three rounds and Protege's three financing events show that younger businesses are also attracting repeat backing.

The sector keeps producing new entrants, but far fewer companies are accumulating the capital needed to build large operating platforms.

What does a normal AI training data startup actually raise?

A typical AI training data financing is around $10 million, and most companies raise considerably less than the amounts associated with headline-making industry leaders.

The most common financing range is $5 million to less than $20 million, containing 32 deals. Another 15 transactions were below $5 million. Together, these ranges account for 47 of the 81 financing events.

Only nine disclosed transactions reached $50 million or more, including four at or above $100 million.

For founders, a $350 million financing shows what the biggest businesses can attract. It is hardly a useful benchmark for an ordinary training-data startup.

The company-level figures reinforce the point: among businesses with fully quantifiable financing during the period, median cumulative capital was $9 million. For 27.1% of those companies, total recorded funding did not exceed $5 million.

Deal size	Deals	Usable capital	Share of capital
Below $5M	15	$35.4M	1.7%
$5M to <$20M	32	$303.9M	14.9%
$20M to <$50M	17	$497.0M	24.4%
$50M to <$100M	5	$302.3M	14.8%
$100M and above	4	$900M	44.1%
Unavailable or unusable size	8	Not quantified	-

Which parts of the AI training data market attract the most investment?

Expert feedback and data annotation dominate AI training data funding today, collecting almost two-thirds of usable disclosed capital despite accounting for just over one-quarter of financing events.

RLHF and Expert Feedback leads with at least $670.8 million, followed closely by Data Annotation and Labeling at $632.2 million.

These categories contain some of the sector's largest financing recipients. Mercor and micro1 operate in expert-data markets, while Snorkel and Encord represent the more established annotation and data-development infrastructure segment.

The surprise is Data Collection Services. It produces more deals than either of the leading categories, yet attracts less than half the capital received by RLHF and Expert Feedback.

Synthetic Training Data shows an even sharper contrast. It accounts for 13 investments but only $63.3 million in usable disclosed funding.

The categories generating the most new financing activity are clearly not always the ones attracting the biggest checks.

Category	Deals	Capital raised	Share of capital
RLHF & Expert Feedback	9	$670.8M+	32.9%
Data Annotation & Labeling	13	$632.2M	31.0%
Data Collection Services	16	$276.1M+	13.5%
Domain-Specific Training Data	14	$208.8M	10.2%
Licensed Proprietary Datasets	8	$113.5M	5.6%
Data Curation & Cleaning	8	$73.9M+	3.6%
Synthetic Training Data	13	$63.3M+	3.1%

Why are investors putting so much money into human expert feedback?

Human expert feedback is attracting unusually large AI training data investments because frontier AI developers increasingly need specialized professional judgment, not simply large volumes of annotated information.

The funding figures are striking. Nine RLHF and Expert Feedback transactions account for almost 33% of disclosed market capital, making this the most capital-intensive category relative to its deal count.

Mercor provides a clear commercial explanation. Its October 2025 $350 million Series C, led by Felicis, valued the business at $10 billion. The company described a model in which doctors, bankers, lawyers and other specialists help train AI systems to perform complex professional tasks.

That financing was intended to expand the expert network, improve specialist matching and accelerate delivery to customers.

Micro1 demonstrates that Mercor is not an isolated case. In August 2026, TechCrunch reported that micro1 had reached approximately $500 million in gross annualized run rate, driven by demand for specialized training data.

These revenue figures help explain the financing scale. Expert-data businesses can command large budgets from model developers that need difficult, high-value tasks completed reliably.

But gross revenue or gross run rate should not be confused with net revenue or profit. Human-expert businesses can have substantial delivery costs, and their economics depend on how much value they retain after paying contributors.

Even allowing for that distinction, specialized human expertise has clearly developed into a serious commercial market.

Why is traditional data annotation still attracting huge rounds?

Data annotation remains one of the best-funded AI training data categories because its leading companies have evolved from labeling providers into broader data-development and model-improvement platforms.

The category received $632.2 million in usable disclosed capital across 13 deals. Unlike most newer segments, it has eight later-stage financing events.

Snorkel AI is the clearest example. On September 22, 2026, the company announced a $350 million Series E at a $3.5 billion valuation, co-led by Insight Partners and S32.

Snorkel also disclosed that its newer data-as-a-service offering had grown more than eighteenfold and reached a $375 million annualized revenue run rate. Its business now spans complex datasets, specialist tasks and environments used to improve frontier AI systems.

Encord, which received $90 million across two qualifying financings, represents another approach: software infrastructure for curating, annotating and managing multimodal datasets.

These are different businesses, but both have moved well beyond the old model of cheaply labeling images. Customers need systems that organize difficult data, measure model weaknesses, produce corrective examples and continually improve training pipelines.

This also helps explain why annotation companies are considerably older than most new entrants. Their median age at financing is six years, compared with approximately one year for several emerging data categories.

Is robotics becoming the next big AI training data market?

Robotics training data is one of the most promising emerging investment clusters, but it is not yet a mature financing market.

Data Collection Services has 16 qualifying deals, more than any other category, and a median company age of just one year. Yet none of its classified transactions has reached Series B or later.

The technical problem is substantial. Language models can learn from enormous digital text collections, whereas robots need physical-world demonstrations involving movement, object manipulation, sensor readings and interactions with changing environments.

Collecting those demonstrations may require robot fleets, teleoperation systems, sensors, human operators and dedicated facilities. It's an expensive business to get off the ground.

XDOF illustrates the funding consequences. The company announced a $70 million Series A in June 2026. In September, TechCrunch reported that it was already discussing a Series B at a valuation of approximately $1.2 billion, with annualized revenue approaching $50 million. Those discussions were not equivalent to a completed new financing.

Mecka AI provides another example of rapid capital formation, with $68 million recorded across three financings. Newer businesses entering physical-data collection are also attracting early-stage backing.

The absence of later-stage transactions is understandable given the youth of the category. Investor appetite has emerged much faster than a conventional financing ladder.

The harder question is whether expensive physical-data collection can become a repeatable business with attractive margins.

Why is synthetic training data getting so little funding?

Synthetic training data is attracting plenty of startup activity but surprisingly little venture capital compared with expert feedback, annotation and physical-data collection.

We identified 13 synthetic-data financings, representing 16% of all deals, but only about 3.1% of usable disclosed funding.

No disclosed financing in the category exceeds approximately $10.6 million, and no classified company has raised a Series B-or-later round during the period.

That funding gap does not necessarily reflect weak commercial demand. Companies can generate useful training examples through simulations, models and software without needing the same operating infrastructure as physical-data businesses.

Synthetic generation is also present inside better-funded categories. Snorkel, for example, combines synthetic-data methods with specialist human input in its broader data offering.

Funding for standalone synthetic-data startups therefore captures only part of the industry's investment in the technology.

For now, investors appear far more willing to write large checks for integrated data platforms and scarce expertise than for standalone synthetic-data businesses.

How young are AI training data startups when investors fund them?

AI training data startups are being funded remarkably early: the median company is only one year old when it receives a qualifying financing.

Across records with reliable founding dates, average company age is approximately 2.3 years. The funded population is strongly concentrated among businesses established recently, including 16 founded in 2024 and 20 founded in 2025.

That youth is visible in the types of rounds being raised. Seed financings typically go to companies around one year old, while Series A companies have a median age of two years.

Larger rounds generally involve older businesses. Companies raising more than $50 million have a median age of approximately four years, and Series D+ financings involve companies with a median age of seven years.

There are exceptions. Some expert-data companies have expanded rapidly enough to attract substantial funding at unusually young ages.

But generally, investors are prepared to finance new training-data approaches almost immediately. The largest checks tend to come later, once commercial traction or operating infrastructure is more established.

How far do AI training data startups actually get before funding dries up?

The AI training data funding ladder narrows sharply after Series A, with 22 Series A transactions falling to only seven Series Bs and two Series Cs.

Seed financings dominate deal count, accounting for 40 of the 81 transactions. Series A contributes another 22, meaning the great majority of classified financings sit at the beginning of the venture lifecycle.

The contrast in capital is considerable. Early-stage transactions represent 82.7% of classified deal count but only 37.4% of capital allocated to rounds with known stages.

Later-stage companies are much less numerous, but their individual financings are substantially larger.

These figures are not literal progression or failure rates. The financing window is too short, and many companies have not had time to graduate. The ratios compare financing events, not a single startup cohort followed over its lifetime.

Still, the scarcity of middle-stage rounds is hard to ignore. There are plenty of initial financings and a few very large winners, but relatively little observable activity in between.

Stage	Deals	Usable capital	Median financing
Seed	40	$275.6M+	$6M
Series A	22	$458.4M+	$23M
Series B	7	$295.8M+	$36M
Series C	2	$410M	$205M
Series D+	3	$450M*	$225M*
Growth Equity	1	$72M	$72M
Unknown	6	$76.8M+	$18M

*Series D+ capital and median exclude Scale AI's $1 billion mixed primary-secondary transaction.

Why do so few AI training data companies reach Series C?

Very few AI training data startups have demonstrated the scale needed to attract Series C capital, and the market's middle-stage financing history is still thin.

The sharpest conventional contraction occurs between Series A and Series B. There are more than three Series A events for every Series B event, followed by seven Series Bs against just two Series Cs.

The average Series B is already worth about $42 million, considerably above the $22 million Series A average. The two Series C financings average $205 million, although that figure is heavily influenced by Mercor and should not be interpreted as a normal Series C check.

Commercial requirements also change as financing sizes grow.

A young data collection company may be funded to demonstrate that it can obtain difficult robot-training examples. A larger platform needs to show that customers will purchase those datasets repeatedly, that production can scale, and that margins justify additional investment.

An expert-feedback provider faces comparable questions around specialist recruitment, customer concentration, delivery economics and the durability of demand.

Investors appear comfortable financing plenty of early experiments. Far fewer companies have yet built the commercial record needed to secure substantial growth capital.

The data cannot tell us how many Seed and Series A companies will eventually succeed. It does show that a broad, proven middle market has not emerged.

Is most funding going to new startups or companies investors already know?

Most AI training data funding is going to previously financed companies rather than first-time fundraisers.

We identified 27 rounds classified as first financings, compared with 53 follow-on rounds. First financings represent one-third of deal events but only approximately 12% of usable disclosed capital.

The median first financing is $7 million. For follow-on rounds, that figure rises to $22 million.

Venture investors are simultaneously supporting the creation of new companies in physical AI, specialized datasets and synthetic data, while making substantially larger commitments to businesses that have already secured financing and begun demonstrating their ability to operate.

That difference helps explain the apparent contradiction between the sector's very young median company age and the large amount of capital concentrated in mature platforms.

New ventures are being created quickly. Most of the money, though, is still flowing to businesses further along in development.

Is AI training data funding accelerating or just becoming more volatile?

AI training data financing activity has become denser toward the end of the period, but dollar volumes remain extremely volatile because individual megadeals can dominate entire months.

Across the 33 completed months from January 2024 through September 2026, we find an average of 2.45 transactions per month and a median of two.

Activity was often limited to one or two financings in individual active months during 2024. By May–September 2026, monthly deal counts more regularly reached five to seven.

September 2026 was the busiest month, with seven transactions, and also the largest by usable disclosed capital at approximately $531.3 million.

However, Snorkel AI and micro1 together contributed roughly $450 million of that month's total. Around 85% of the capital therefore came from just two events.

Announcement timing also complicates the monthly picture. Some deals first announced in 2026 had economically closed during 2025. Worldmodeldata and Proximal are examples, while Mecka's financing disclosures also complicate the apparent monthly sequence.

The increase in publicly announced deal activity is visible. The fluctuations in invested dollars are much less useful as an indication of the underlying market's momentum.

Is the AI training data funding market dominated by the United States?

The United States overwhelmingly dominates disclosed AI training data venture capital, although the market has a meaningful population of European and Asian companies.

North America accounts for 54 transactions and $1.765 billion in usable capital, representing 86.6% of the market's disclosed financing dollars.

The United States alone contributes 51 transactions across 40 companies, with approximately $1.733 billion in usable disclosed capital.

Europe represents the second-largest funding region, with $216.5 million across 13 transactions. Toloka's $72 million financing is one of its most important contributions.

Asia-Pacific contains 14 financing events across 13 companies, nearly matching Europe's activity by deal count. Yet only $57 million in usable financing amounts is disclosed.

That gap is partly a reporting problem. Several Chinese transactions report vague amounts such as tens or hundreds of millions of renminbi without a usable exact value.

Asia-Pacific's measured capital is therefore especially incomplete.

Even allowing for those missing amounts, the largest fully quantified financings are concentrated in North America. Startup formation is international, but the biggest disclosed venture checks are overwhelmingly American.

Which investors keep funding AI training data startups?

No single venture capital firm consistently controls AI training data financing. Investor leadership is fragmented, with relatively few firms repeatedly appearing as disclosed leads.

Across the database, 71 distinct lead or co-lead investor names appear, but only six participate as leads in more than one transaction.

Framework Ventures is the most frequent, appearing three times. General Catalyst, Wing Venture Capital, Lightspeed Venture Partners and Amplify Partners each appear twice.

This fragmentation contrasts with the high company-level concentration of capital.

Large companies can accumulate substantial funding without the industry depending on one dominant investor network. Snorkel's 2026 financing involved Insight Partners and S32, while Mercor's Series C was led by Felicis. Different investor groups are backing different parts of the market.

Disclosed lead-investor coverage is approximately 79%, so the table cannot capture every repeat relationship. Nevertheless, the observed market has a surprisingly dispersed group of lead investors relative to how concentrated its funding recipients have become.

Lead investor	Led deals	Size of associated rounds
Framework Ventures	3	$69M
General Catalyst	2	$46.2M
Wing Venture Capital	2	$39.95M
Lightspeed Venture Partners	2	$39.2M
Amplify Partners	2	$36.65M
Oriental Fortune Capital	2	Undisclosed

Are strategic investors making a serious bet on AI training data?

Strategic investors are meaningful participants in AI training data funding, particularly in larger rounds, but they are not involved across the entire market.

We identified corporate or strategic investment participation in 17 of 81 financings, equivalent to 21%.

The median strategically backed financing was $21.5 million, versus $9 million for transactions without an identified strategic participant.

Corporate investors are more frequently present in larger transactions, especially where the technology is relevant to industrial AI development or the business has moved beyond initial experimentation.

Strategic participation is most common in annotation, with six deals, followed by synthetic data with four and data collection with three.

NVIDIA appears strategically in three financings, while LG Technology Ventures participates in two.

For data-infrastructure startups, corporate investors can bring potential customers, technical relationships and industry access alongside capital. Model developers, computing-platform providers and industrial technology businesses may have direct uses for the datasets or evaluation systems being developed.

Of course, an investment does not automatically imply a purchasing agreement, preferential data access or commercial deployment. Those arrangements require separate evidence.

What the financing data shows is selective strategic involvement, generally associated with larger rounds and commercially relevant AI infrastructure.

What is missing from the AI training data funding market today?

The most conspicuous absence is a broad population of mid-stage AI training data companies capable of raising substantial growth capital repeatedly.

Several categories are producing new startups without developing a visible late-stage financing ladder.

Data Collection Services has 16 transactions but no classified Series B-or-later rounds. Synthetic Training Data has 13 transactions and likewise no classified Series B+. Licensed Proprietary Datasets has eight financings, all at Seed or Series A.

Data Curation and Cleaning presents a different problem: concentration. DatologyAI's $46 million Series A represents more than 62% of usable category capital.

These gaps separate commercially promising technologies from financing markets with demonstrated scaling histories.

There is also limited evidence of repeat fundraising among newer companies. Most of the 66 businesses appear only once, leaving little observable history of company progression.

And while strategic investors are clearly present, we have not found evidence that a handful of industrial backers systematically support the entire ecosystem.

Much of the funded population is still very young, so it would be premature to read these absences as evidence of commercial failure.

For now, the sector has demonstrated that it can generate new ventures and finance a few very large companies. The diversified middle tier is still missing.

What are investors really buying when they fund AI training data companies?

Investors are increasingly funding the ability to produce difficult, commercially valuable training experiences for AI systems, rather than merely the ability to supply large quantities of data.

Three economic models stand out in the financing data.

Mercor and micro1 organize specialist human knowledge into training and evaluation work for AI developers. Their challenge is delivering expertise at scale while maintaining quality and workable economics.

Snorkel and Encord provide the infrastructure to prepare, curate and improve datasets and model-training workflows. Their value comes partly from software, operational systems and integration with customers' AI development processes.

XDOF and Mecka represent a newer model built around collecting real-world experience for robotics and physical AI. Their operations can require equipment, teleoperators, facilities and physical data-production capacity.

These businesses have very different cost structures, competitive advantages and scaling constraints. Treating them all as conventional data-labeling companies misses much of what investors are actually financing.

One commercially important shift is toward increasingly complex tasks and interactive environments. Snorkel's September 2026 announcement explicitly emphasized expert agentic tasks, environments and detailed evaluation criteria rather than simple labeling.

The biggest checks are going to companies positioned around hard-to-obtain expertise, integrated training workflows and complex data-production systems.

Whether those businesses generate the strongest long-term returns remains to be seen. But the financing decisions already show which capabilities investors are willing to pay a premium to build.

What should we actually take away from the funding data?

The AI training data market is now a substantial but highly uneven venture ecosystem. Its defining characteristic is the gap between the number of companies receiving early funding and the small group capable of attracting genuinely large amounts of capital.

The market has meaningful breadth. We identified 66 funded companies and 81 financing events, with $886 million in usable capital remaining even after excluding rounds above $50 million. This is far more than a sector sustained by a few exceptional transactions.

Capital concentration is nevertheless severe. Ten companies account for nearly 72% of usable disclosed funding, and repeat fundraisers absorb more than half the money despite representing fewer than one in five funded businesses.

The most financially developed categories are expert feedback and annotation. Mercor, micro1, Snorkel and Encord demonstrate that these businesses can attract large institutional commitments, supported in several cases by substantial commercial run rates.

Physical-AI data collection is the most interesting emerging category by startup activity. It has the largest number of deals but no classified Series B-or-later financing. Investors are backing its potential before the market has developed a mature funding history.

The funding ladder is the central weakness. Forty Seed financings and 22 Series As contrast with just seven Series Bs and two Series Cs. That is not a measured startup failure rate, but it reveals how little middle-stage depth currently exists.

Strategic capital matters selectively rather than universally. Corporate investors participate in roughly one-fifth of transactions, disproportionately in larger rounds, while traditional venture leadership remains fragmented across many firms.

Ultimately, the evidence points to a market where raising initial capital is relatively common, but building a company that deserves $50 million, $100 million or more remains exceptional. The next test for AI training data will be whether today's large population of young companies develops into a diversified group of durable businesses, or whether commercial demand and venture capital continue concentrating around a small number of dominant data platforms.