This Is How AI Data Firms Make Money

🍿 Quick Snack

• 🍽️ AI companies are hungry for more data and are willing to pay up to $300 per hour of video. Some people are going out of their way to acquire some of photo assets directly from the police and medical students.

• 🍁 Big win for Canada. Clio, a legal practice management platform, just raised $900M and reached a $3B valuation.

• 🍟 Extra Fries: OpenAI’s reaches $3.4 billion in revenue, and Wiz rejects Google’s $23 billion offer.

🍔 The Full Meal

The Economics of AI Data Firms

The market for datasets to train AI is forecasted to reach $4.8B in 2024. By 2032, it is expected to reach $27B.

This has given rise to “AI data firms”, whose sole purpose is to secure rights to real-world content and license the data to companies like Google, Meta, Apple, Amazon, and Microsoft. One of these companies is Defined.ai, a dataset marketplace specialized in curating and providing high-quality and ethical data for AI applications.

But exactly how did we get here? and much money can these companies make?

Let me explain.

What Led Companies to Want to Pay for Data

AI companies are incentivized to leverage as much as data as possible to train their models. After all, the more training data, the more powerful the model is.

But, there is a problem. When AI companies resort to scraping data in efforts to keep their costs down, they’re prone to copyright infringement. Just last year, the New York Times sued Open AI and Microsoft for this exact reason. To avoid any further troubles, AI companies are now securing deals with content owners to establish a proper data-supply chain.

This rise in demand has also created an appealing opportunity for a special type of broker — data marketplaces.

The Economics of Licensed Data

In 2023, Shutterstock’s AI-Licensing Business generated $104 million in revenue.

Freepik has also struck agreements with two large tech companies to license the majority of its archive of 200 million images at 2 to 4 cents per image. That is a $200 million business at the very minimum. And it gets juicier.

According to Daniela Braga, CEO of Defined.ai, buyers are generally willing to pay:

• $0.001 per word

• $1 to $2 per image

• $2 to $4 per short-form video

• $5 to $7 per image of nudity

• $100 to $300 per hour of video

This has created a push for AI data firms to secure rights to content like podcasts, custom visuals, voice samples, and sort-form videos. But also, it has incentivized some companies to put in the effort and monetize the higher-priced tiers.

For example, one supplier of Defined.ai, who wishes to remain unnamed, sells images used to train AI systems that block content like graphic violence. These would be images from crime scenes, conflict violence, and surgeries. To obtain these images, they source them directly from police, freelance photojournalists, and medical students.

Then, the company hires nurses (who are accustomed to seeing violent injuries) to anonymize and annotate these graphic images.

Wild 🤯, isn’t it?

This Is the $3B Company Powering Law Firms

Clio, a Canadian startup providing law firms with management software just raised a $900M round, part of which will be used to allow existing investors and employees to cash out some of their shares.

In only 2 years, the company added another $100M in ARR to its revenue base. Total ARR currently sits at $200M.

This is what their cap table looks like after the deal:

🍟 Extra Fries

• Kennet Partners closes €266M fund to invest up to €50M in European SaaS startups (link)

• OpenAI’s annualized revenue doubles to $3.4 billion (link)

• Wiz walks away from Google’s $23B acquisition offer (link)

PS. This is the e-mail the CEO of WIZ sent to the entire company after rejecting the deal 👇