Papaya Global's case for AI workforce infrastructure rests on its own country counts

The operations lead who has to stand up a labelling operation across eight countries gets a demo and a quote, then hits a calendar problem ^[21]. In Germany an employer-of-record arrangement can run 18 months; in France, 36 ^[8]. The contract signed in week three needs an answer for month 19.

Take the size of the population first. Dataintelo and Market.us put the people who label data and review outputs at roughly 3 million now, heading toward 35 million as LLM training, reinforcement learning from human feedback and computer-vision crowd work scale ^[2]. That is a twelvefold increase ^[16]. Papaya's own figure for the wider gig workforce, presented to analysts, runs from about 100 million to 435 million ^[4], a 4.35 times increase ^[17]. The Next Web, which reported the presentation, wrote that the claims "still need checking" ^[15].

The counts that decide whether a multi-country rollout is buildable at all are Papaya's. By its tally, 48 countries require a staffing licence to lease labour ^[7]. Sixty-five apply strict contractor-classification tests and 28 actively enforce them ^[9], which leaves 37 where the test exists and enforcement is not active ^[18]. A buyer planning annotation capacity cares about the 28. In those countries the bill for misclassifying annotators is back pay and benefits, plus a criminal referral in some jurisdictions ^[10].

The exposure starts before payroll. TransUnion surveyed 1,012 US adults for its 2026 Gig Economy Worker Report, published in January, and found that one in four gig workers have rented out their verified accounts and one in five have sold them, rising to 31 percent for renting among Gen Z and millennial earners ^[11]. Only 45 percent said the platforms they use have effective identity verification ^[12]. The sample is US adults; whether account-sharing runs at similar rates among annotators hired offshore is not in the survey.

Mostly it is not the labs doing those checks. SOMO counted almost 500 companies active in AI data collection and labelling in March 2026, and found Amazon, Google, Meta, Microsoft and Nvidia using at least 30 intermediaries between them ^[3], an average of six per buyer ^[19]. The employer-of-record category grew up on a simpler promise, a dozen well-funded vendors offering foreign hires without opening an entity, with employment handled through local partners and payments through third-party rails ^[13]. Some now emphasise owned legal entities in the countries they serve ^[14]; others bundle global payroll into all-in-one HR, IT and finance platforms for buyers who want one system of record ^[20]. Owned entities and local partners are different answers to the licence requirement in those 48 countries ^[7][14].

Two facts per country set the shape of the deal, and both have to come from somewhere other than the deck. The first is whether the law caps the duration of the arrangement ^[8]. The second is whether classification is actively enforced ^[9]. A capped, enforcing country like Germany needs a conversion date and a named entity to hold the workers before the first annotator starts ^[8]. Where there is no cap but enforcement is real, ask which local entity signs the contract and who carries the back pay if a tribunal reclassifies the workers ^[10]. Where a cap exists without active enforcement, it is a scheduling question. And in the countries that simply require a licence to lease labour, the question is whether the vendor holds one ^[7].