Website concept by Stoa · Not the official TechEquity website

Update

The hidden workforce behind every AI tool you use: Announcing the data work landscape 2.0

“Beheadings, child abuse, people making love—how many times can you watch these things before you stop feeling anything at all? . . . We lose our ability to be human. We lose our emotions. We can’t be happy. We can’t be sad. We become robots.”
Content moderator, Türkiye, from The People Behind the Screens: Why Tech Companies Need New Protocols for Safe Content Moderationby UNI Global Union

That quote didn’t come from a war correspondent or an emergency room doctor. It came from someone doing a job that most of us rarely consider—a job that makes the internet (and increasingly the AI tools) you use every day possible.

The AI tools you use feel seamless. Instant answers, filtered feeds, recommendations that know you better than you know yourself. What’s hidden behind that seamlessness is a massive human workforce—annotators, content moderators, and AI trainers (what we call “data work”) who form the invisible backbone of the artificial intelligence economy. There are millions of them around the world. Most people have no idea they exist. And the companies profiting most from their labor want to keep it that way.

Today, TechEquity is launching Version 2.0 of our Data Work Landscape—a map of the companies, investors, and client relationships powering the data portion of the AI supply chain. This update goes deeper: we’ve added detailed profiles of ten of the most prominent players in the industry, what they do, where they sit in the ecosystem, and what makes them notable—alongside a new interactive visualization that maps 1,924 relationships between the 116 data work companies in our database and their customers, partners, subcontractors, and investors,  surfacing dependencies and concentrations of influence in the landscape. 

Here’s what we’ve learned, and why it should matter to all of us.

Three truths about this industry

The industry is booming—but not for the workers. Data annotation is one of the fastest-growing jobs in the United States, according to LinkedIn, Upwork, and Deel. While many people fear the potential of job loss as AI is deployed across industries, tech executives often discuss the opportunities created by new job categories. Data work is a key example of the new jobs created by AI, but is this quality employment? Companies like Handshake, Mercor, Surge, and Scale AI are each generating close to $1 billion in gross revenue—and the profits are getting higher. Since the initial launch of the Data Work Landscape last June, Meta took a 49% stake in Scale AI for over $14 billion—valuing Scale at over $29 billion and marking Meta’s second-largest acquisition deal ever. The deal reshuffled the competitive landscape, with Scale’s rivals reportedly moving quickly to poach its customers. Meanwhile, the workers powering all of this growth remain in precarious, easily-terminated contract positions with few protections and little recourse.

Big tech controls this industry from the shadows—and won’t say so publicly. Amazon, Google, and Meta all refused in 2025 to disclose which data work vendors they use. Their contract decisions have immediate, life-altering consequences for workers around the world. That opacity is not accidental—it is a business strategy.

The risks don’t stay contained to data workers. Cybersecurity vulnerabilities, intellectual property concerns, and the rapid expansion of gig work into new sectors affect everyone. This is not a niche labor story. It is a story about how AI gets built, who pays the price, and what the hidden costs of the AI economy mean for all of us.

The human cost of “flexible work”

What does precarious labor actually look like in this industry? It looks like a big tech firm is canceling a contract overnight, and thousands of workers are losing their income—with limited warning, severance, and explanation.

When Meta recently shifted its vendor contracts, Sama laid off over 1,000 Kenyan workers. Workers at Covalen in Ireland experienced the same. These are not isolated incidents. They are the predictable, repeatable result of a system deliberately designed to keep labor costs flexible and accountability nonexistent.

The precarity is also reaching the United States. TechEquity’s research with AWU-CWA highlighted how many U.S. workers get involved in data work seeking flexibility due to mental or physical disabilities, or care responsibilities. And yet, while 34% of our survey respondents last year reported having a disability, only 7% (three out of 42) reported having an accommodation at their job. Data work companies are also increasingly recruiting from the ranks of American universities, where young workers face a challenging job market alongside experienced workers who have been laid off, including federal government employees and academics.

And the harm goes beyond job loss. Investigations have found gig workers in Africa unknowingly contributing to U.S. military AI projects. Others have reportedly been asked to review intimate content generated by consumer products such as Meta’s Ray-Ban glasses. 

Watch the recording on YouTube

Big Tech’s invisible hand

The companies at the top of this supply chain present themselves as AI innovators. What they rarely acknowledge is their role as the architects of the conditions described above.

In a competitive landscape where data work companies battle to maintain contracts with AI firms, Big Tech shapes the entire industry. As a recent report by Amsterdam-based organization SOMO noted: “Even when not directly employing data workers, major tech companies influence working conditions through pricing pressures and tight deadlines and by shifting contracts between vendors. Their decisions have immediate consequences for workers’ pay, stability, and job security.” 

SOMO’s report, and the additional relationships revealed in our new network graph, also show how big tech companies like Alphabet, Amazon, and Microsoft exert influence in the data work industry as customers, investors, and by controlling marketplaces, since these companies also serve as cloud providers through which data work companies access additional clients. As SOMO says: “By embedding data work into their platforms, cloud providers both enable AI development and shape working conditions for its largely hidden workforce.”

This secrecy makes it more challenging for workers to build the necessary leverage to hold accountable the companies that most significantly shape their working conditions in the supply chain. That is not an oversight. That is the point.

Trends worth watching

“White-collar” workers are the new target

This is no longer a story only about workers in the Global South. AI training companies are capitalizing on mass layoffs across academia, government, law, and tech in the U.S. as well. Laid-off scientists, lawyers, and academics—and college graduates who can’t find work in their fields—are actively recruited to train the models threatening their professions. The cruel reality at the center of this trend: these workers are building their own replacements.

But the impact extends beyond the workers doing the training. Our research reveals that data work companies are increasingly selling their services to enterprises across a remarkably broad range of industries—from tech to manufacturing, agriculture, health care, retail, education, entertainment, insurance, real estate, transportation, construction, and to governments around the world. As AI gets embedded into more sectors of the economy, the workforce disruption that data workers are already experiencing today is a preview of what’s coming for workers across the board. The companies building the AI supply chain aren’t just processing data. They are fueling the expansion of AI into every corner of the economy.

Your favorite gig apps are getting into AI training

Uber, DoorDash, and even LinkedIn are all testing or launching AI data work programs using their existing workforces. Drivers and couriers are being asked to complete digital tasks for AI and robotics training between gigs, with little visibility into what they are building or who it is for. The gig economy just got a new product line. Workers didn’t get a vote.

Your work product may already be involved

Data work firms have been asking contractors to submit professional documents, code, and analyses from prior jobs—work that may legally belong to former employers. Workers are also being paid to scrape social media feeds for training data, which raises complicated intellectual property implications. The companies soliciting this work are counting on this ambiguity. A new cottage industry has even popped up to sell the digital footprint of work products from shuttered companies, everything from Slack messages to emails and Jira tickets, to tech companies for AI training. Even for those of us who are not directly in “data work” roles, our work product may be used for this purpose: Meta announced that it will closely monitor its employees’ work processes on its computers in order to train its AI tools.

Cybersecurity risks are everyone’s problem

Data workers are routinely required to install extensive monitoring software on their personal devices, creating vulnerabilities at scale. The consequences are already here: a breach at AI training firm Mercor exposed information subsequently used to access a sensitive, unreleased Anthropic AI model, Mythos. And it may be that cybercriminals are now in possession of identification information, computer screenshots, facial biometric data, and studio-grade voice recordings of tens of thousands of contractors: “a nightmare scenario.” Scale AI separately reportedly exposed both sensitive details about its contractor and client data, including information about Meta and xAI, in publicly accessible Google Docs. When the people building AI systems operate under precarious conditions with inadequate security infrastructure, those vulnerabilities don’t stay contained. They travel up the supply chain. Eventually, they could reach all of us.

The data work industry’s growing political influence

Just before Michael Kratsios was confirmed as the Trump Administration’s director of the Office of Science and Technology Policy (OSTP), he was a managing director at Scale AI. In his current role within the White House, Kratsios shapes all AI policy for the country (and exerts a global influence). Scale AI has spent $420,000 on lobbying in the US so far in 2026. 

Some companies in the data work landscape, including data work companies like Cognizant and popular customers like Box, Salesforce, and ServiceNow, who are involved in “developing, refining, integrating, and deploying AI technologies,” launched the AI Integrators Council to be the “voice for companies operationalizing AI across sectors.” (Companies like ServiceNow sometimes also serve as an investor in data work companies, for example, through its venture fund, which has invested in Scale AI.) The AI Integrators Council spent $160,000 lobbying the US federal government in 2025 and released a 2026 white paper outlining a more limited scope of responsibility for their companies in AI policy while expanding their access to public dollars. 

The workers training AI face being replaced by it

Data work companies are increasingly deploying automated systems to handle annotation tasks that were previously done by humans—repositioning human workers as a review layer on top of AI output rather than as the primary workforce. This is a significant structural change. It means fewer hours, less pay, and less leverage for workers—while allowing companies to market the continued involvement of human oversight to clients who demand it. It is, in other words, another way the industry extracts value from workers while reducing its obligations to them.

Pulling back the curtain

Even with dedicated research, basic information about data work companies remains remarkably hard to find. Corporate structures are opaque, vendor relationships are undisclosed, and worker counts are unverified. 

What does emerge clearly is how uniformly these companies market themselves: as providers of elite, expert labor from PhDs to Olympiad medalists, clinicians, and engineers. Many data work companies, including HumanSignal, Labelbox, Prolific, Scale, Snorkel, and Welocalize, among others, now also offer AI evaluation and benchmarking tools that align with their efforts to market access to elite knowledge. The marketing tells one story. The lawsuits, the data breaches, and the overnight layoffs tell another story.

The Data Work Landscape exists to document the gap between the two. On the updated microsite, you’ll find three tools for exploring it: a database of 116 companies involved in sourcing and managing AI data work globally, categorized by product type and business model; a network graph mapping over 1,900 customer, partner, and investor relationships across the ecosystem; and detailed profiles of ten of the most prominent players—their business relationships, recent developments, and media coverage. Together, they offer a picture of this industry and its broad impact.

Why this matters and what you can do

The AI tools reshaping our economy are built on a hidden workforce. That workforce deserves visibility, accountability, and protection. And the companies profiting from their invisibility deserve scrutiny. Transparency requires pressure. 

The Data Work Landscape is one tool for creating that pressure.

Explore the updated landscape Read our report on U.S. data workers 

If this story matters to you, share it. The workers who make AI possible deserve to be seen. Are you a data worker yourself, or do you have information about the industry’s inner workings? Sharing your story can help elevate this issue and get those in power to act.

For Further Reading:

The issue