There's a peculiar irony in artificial intelligence right now. The technology promising to automate human labor has created a booming market for, well, human labor. Specifically: the kind of expertise-driven data annotation work that can't yet be automated away.
Turing, a Palo Alto company that supplies that specialized training data to OpenAI, Anthropic, and the rest of Silicon Valley's AI establishment, just raised $111 million at a $2.2 billion valuation. The Series E, which closed March 6 with Malaysia's sovereign wealth fund Khazanah Nasional leading, brings total funding to roughly $225 million. But the financing itself isn't the story—or at least not the most interesting part of it.
What catches the eye: Turing says it hit $300 million in annual recurring revenue while turning profitable in 2024, numbers the company disclosed January 28. That's revenue nearly tripled from the prior year, achieved while most AI infrastructure companies are still lighting money on fire in the name of market share.
The trajectory reflects something bigger than one startup's financials. As frontier labs run up against what insiders now casually term the "data wall," the scramble for high-quality training examples has become expensive, frantic, and essential.
When Internet Data Runs Dry
The constraint is straightforward, if inconvenient. AI labs have largely exhausted the easily accessible text and images scraped from the open internet. What remains—the good stuff needed to push models toward more sophisticated reasoning—requires human experts to create from scratch or annotate in painstaking detail.
A single complex annotation can run hundreds of dollars. Models need millions of them. Do the math.
Turing built its business model around that appetite, assembling a network of more than 4 million developers and STEM specialists across its platform. According to Reuters, customers include OpenAI, Google, Anthropic, and Meta—essentially everyone racing toward artificial general intelligence. The company provides coding expertise, advanced reasoning inputs, and what's known in the trade as multimodal training data (think: teaching AI systems to understand images, text, and other formats simultaneously).
Its ALAN platform handles the technical infrastructure: model evaluations, supervised fine-tuning, reinforcement learning from human feedback (RLHF, for the acronym collectors), and agent development. These are the unsexy pipes beneath the consumer-facing chatbots everyone's now arguing about at dinner parties.
The Dual Revenue Engine
Turing organizes its work into two divisions, though both draw from the same talent pool. Turing AGI Advancement is the prestige operation, partnering with frontier labs to improve coding, reasoning, and multimodal capabilities in cutting-edge models. Turing Intelligence, meanwhile, builds applied AI systems for corporate clients: healthcare supply chain optimization, finance chatbots, the kind of enterprise implementations that Fortune 500 companies in tech, banking, retail, and healthcare actually purchase.
It's a sensible hedge. Training data for AGI research is lucrative but concentrated among a handful of well-funded labs. Enterprise AI deployments offer volume and diversification, if less of the intellectual cachet.
The company's origin story follows a familiar Bay Area pattern. Founders Jonathan Siddharth and Vijay Krishnan, both Stanford computer science graduates, previously sold a content recommendation startup called Rover to Revcontent for north of $30 million in 2017. They launched Turing in 2018 as an AI-powered vetting platform for remote software developers—basically a hiring marketplace with algorithmic matching.
The pivot toward training data infrastructure came as model development grew increasingly specialized. Whether the founders anticipated that shift or simply reacted quickly is the kind of question that gets rewritten in retrospect. Either way, timing worked in their favor.
A Crowded Cap Table, A Different Strategy
The Series E adds Khazanah Nasional to a cap table already thick with names. WestBridge Capital, which led an $87 million Series D in December 2021 at a $1.1 billion valuation, participated again. So did Sozo Ventures, UpHonest Capital, AltaIR Capital, Amino Capital, Plug and Play, MVP Ventures, Fortius Ventures, Gaingels, and Mastodon Capital Management.
Earlier rounds included a $14 million seed in 2020 led by Foundation Capital, featuring Quora CEO Adam D'Angelo and former Facebook executive Gokul Rajaram as angel investors, followed quickly by a $32 million Series B.
The new capital will fund R&D and sales expansion across both business lines, along with continued development of the ALAN platform. Standard stuff.
What's less standard: the company's approach to growth economics. At $2.2 billion, Turing's valuation sits well below competitor Scale AI, which reportedly reached $14 billion in recent funding rounds. But Turing's profitability at a $300 million revenue run rate—implying about a 7.3x revenue multiple—suggests management chose a path less traveled in the venture-backed world. Sustainable unit economics over blitz-scaling, perhaps.
The distinction may matter more than usual. Regulatory pressure is mounting on the industry's labor practices. The Department of Labor is currently investigating Scale AI's contractor compensation, a reminder that rapid expansion built on gig-economy-style workforces can attract unwanted attention. Defensibility increasingly means more than just technological moats.
Reading the Revenue Acceleration
Here's a detail worth pausing on: when Turing priced its Series E, the company was reportedly running at $167 million in annual recurring revenue. By the time it announced the round, that figure had jumped to $300 million. Revenue nearly doubled between deal closure and public disclosure, which suggests either exceptionally aggressive accounting practices or genuine momentum in customer demand.
The latter seems more plausible. As the data wall becomes undeniable, labs are paying up for quality training data. The alternative—inferior models, slower progress toward AGI benchmarks, competitive disadvantage—is simply too costly in an arms race this expensive.
Whether that demand sustains as models improve their own data synthesis capabilities remains an open question. But for now, at least, being the intermediary between AI labs hungry for training data and millions of subject matter experts willing to provide it isn't a bad place to sit. Even if the work involves less glamour than building the models themselves.
The company declined to comment on specific revenue projections or customer contracts. A spokesperson confirmed the funding amount and valuation but offered no additional details beyond the January 28 announcement. In an industry prone to hyperbole, the relative restraint feels almost quaint.
