Surge AI: The $24B Data Engine Teaching AGI Real Human Taste

In partnership with

(5 minutes)

Hi 👋, this is the Today in AI Newsletter: The weekly newsletter bringing you one step closer to building your own startup.

We analyze a cool, industry-shaping AI startup every week, with a full breakdown of what they do, how they make money, how much they’ve raised, and the opportunity ahead.

Let’s get to the good stuff in this email:

• 💡 This startup is the human-data engine behind frontier labs like OpenAI, Anthropic, Mistral AI & more.

• 📈 They’ve bootstrapped to $1B+ in revenue, stayed profitable from day one, and now manage a core team of about 250 people.

• 🚀 They are now preparing their first-ever external raise of up to $1B, targeting a ~$24B valuation - founder Edwin Chen owns ~75% of the company and sits on an estimated $18B net worth as the youngest member of the Forbes 400 (#55).

So what’s the startup and who are the founders behind it? Here’s the story of Surge AI 📈

Surge AI was founded by Edwin Chen in 2020 to build high-quality human data that trains and evaluates the most advanced AI models.

Instead of cheap crowd work that tags cat photos, Surge runs a curated marketplace of elite annotators who handle things like RLHF, safety, reasoning, taste, and multi-step agent behavior. 🤖

Their mission is to raise AGI like you would raise a child so it grows up curious, imaginative, and aligned with rich human values, not just optimized for leaderboard scores & clickbait.

On the product side, that looks like:

Humans and models working together, with models generating and self-labeling, then Surgers reviewing, correcting, and adversarially testing them.

Complex RL environments that feel like video game worlds filled with tools like Slack, Gmail, GitHub, spreadsheets, and CRM systems, where agents learn to handle real productivity tasks across long timelines.

Surge’s key idea: the ceiling of your model is set by the quality of your human data.

You can’t train a master chef by giving them a million recipes for boiling water.

You train them on rare, expert recipes, subtle flavor calls, and detailed feedback.

Surge gives that level of training signal to AI.

Backstory 👀

Edwin Chen grew up in Crystal River, Florida, in the back of his parents’ Chinese-Thai-American restaurant, Peking Garden.

He did homework near the cash register while his parents worked shifts, and he fell in love with math in the eighth grade, when calculus made the world feel legible.

That curiosity took him to Choate, then to MIT at 17, where he studied math, linguistics, and computer science and even worked on decoding an Incan knot language.

He was obsessed with how humans encode meaning into patterns, whether it was language, symbols, or data.

After MIT, Edwin went deep into applied math and ML.

He did algorithmic work at Peter Thiel’s Clarium Capital, then held ML roles at Google, Facebook, Twitter, and Dropbox, shipped recommendation systems, search, and content moderation.

This is where the frustration started.

At Facebook and Twitter, he watched outside vendors deliver datasets that were slow, and off-target.

Basic tasks like distinguishing a coffee shop from a hospital would come back with ridiculous errors.

The teams were spending millions on data that made models worse. Edwin saw a huge waste of human potential.

Smart people inside big tech were building powerful models, while outsourced annotators with weak tooling and low pay were tagging data with almost no context.

Around 2020, right after GPT-3 landed, Edwin snapped.

He realized the entire industry treated data labeling as a commodity cost center, when it should be a core R&D function.

So he quit, moved to Manhattan, and decided to build the opposite of the standard playbook.

High-end human data, top talent, premium pricing, and no VCs telling him to chase vanity metrics.

The Hustle 🤑

Edwin wrote the first version of Surge AI’s MVP in about one month, mostly solo, in a small San Francisco apartment.

The product was simple at first. A platform where ML teams could define rich labeling tasks, and route them to a hand-picked pool of annotators.

He used savings from his decade in big tech, a "couple million" by his own count, and set three strict rules for the company:

1. No external financing.

2. No burning cash.

3. No low-value, simple orders.

If a task looked like commodity tagging for pennies, Surge would decline it.

Edwin then used one of his most underrated assets. His blog.

He had built a reputation in the data science community through long technical posts, like the famous "Pop vs. Soda" map that went viral from his Twitter days.

He turned that distribution into a GTM channel.

Early Surge customers came through deep content aimed at serious data scientists.

He wrote about data quality, annotation philosophy, and AGI alignment.

From there, the team focused on building the highest-quality data annotation workforce on the planet.

To join Surge Force, annotators had to pass a brutal funnel.

1. Strong professional background.

2. Five trial writing tasks.

3. Review and approval by senior annotators.

They pulled in professors from Stanford, Princeton, and Harvard, lawyers, finance pros, and native speakers across 50+ languages and dialects.

For many roles, the acceptance bar is described as harder than getting into an Ivy League. Surge paid accordingly.

Contractors earn $20–$40+ per hour, with some tasks paying 30–40 cents per working minute, far above other gig platforms.

Inside the platform, Surge’s own ML models track keystrokes, speed, accuracy, and agreement patterns, constantly updating per-annotator trust scores.

Bad or lazy work is detected and removed.

Surge also leaned into adversarial work.

They run AI red teams, including projects with groups like Redwood Research, where experts try to trick models into sneaky behavior, then label and fix the failure modes.

Quality is the product.

That contrarian take turned into a huge advantage when Meta invested heavily in rival Scale AI and took a huge stake.

Labs like Google, OpenAI, Anthropic, and xAI did not want their most sensitive data and roadmaps flowing through a vendor effectively tied to a competitor.

They needed a neutral, independent supplier.

Surge was sitting there, profitable, battle-tested, and fully independent.

Stats 📊

From a cold start in 2020, Surge AI scaled to about $1.2B in revenue in 2024, beating Scale AI’s ~$870M over the same period.

The company has roughly 250 people including part-time and consultants, which means close to $9M+ in revenue per person.

On top of that, Surge coordinates a global workforce of over 1M contractors, with about 50,000 considered expert, high-skill Surgers.

Surge has been profitable almost from day one, growing into what many consider the most capital-efficient AI infrastructure company in Silicon Valley.

It has done this with $0 in outside funding so far. Edwin still owns roughly 75% of the company.

At a current estimated valuation of ~$24B, that gives him a personal net worth of ~$18B. In 2025, he entered the Forbes 400 at #55, as the youngest person on the list.

Customer spend shows how central Surge has become.

Meta reportedly spends over $150M per year with Surge.

Google spends over $100M per year.

Other major labs, including OpenAI, Anthropic, Microsoft, and xAI, rely on Surge for RLHF, model evaluation, and safety-critical pipelines.

Surge often charges 50% to 10x more than other providers.

They’re so good because their data improves model performance in ways you cannot reach with cheap labeling.

Before you go 👋

Thank you for reading. Know someone who'd love this newsletter? Share the love. 🌟

🚀 Building something in AI? Get featured in front of 100,000+ founders, builders, and investors. 👉 [email protected]