From Cloud Hosting To Stock Photography: Revealing Big Tech's Fierce Underground Generative AI Race

While tech companies can certainly scout and scrape the web, there are limits into what they can do.

First, there is the capacity, and that companies have limited resources. There is no way for any single entity to entirely consume all of the web in a short burst.

The web grows, and is still growing fast. Even Google Search, the largest internet search engine of all, only covers the surface web.

Second, there are boundaries that are protected by the law, meaning that tech companies cannot do anything they want, because there are things that are legal, and things that are illegal.

To ensure a healthy competition that doesn't annoy regulators, and in order to do things legally, and in their capacity, one of the things tech companies can do, is buy the data they want, at their own pace.

AI needs a lot of images to train.

And among the places where lots of images exist, include stock photography platforms, and cloud hosting companies.

One example, is Photobucket.

In the early 2000s, it was the world's top image-hosting site, that hosted users' images for services like MySpace and Friendster. It was also the default photo sharing platform for Twitter.

At its peak, it boasted 70 million users, hosting images accounting to about half of all photos uploaded by users in the U.S.. At that time, it employed 120 people and accounted for 2% of American internet traffic.

Fast forward, only 2 million people still use Photobucket, and that it only employ about 50 employees.

And the AI revolution that was kickstarted by OpenAI when it released ChatGPT, was soon followed by other big tech companies that wish to piggyback the trends.

This is giving Photobucket a new lease of life.

CEO Ted Leonard said that tech companies are asking to license Photobucket's 13 billion photos and videos to be used to train generative AI models that can produce new content in response to text prompts.

He has discussed rates of between 5 cents and $1 dollar per photo and more than $1 per video, he said, with prices varying widely both by the buyer and the types of imagery sought.

"We've spoken to companies that have said, 'we need way more,' Leonard added, with one buyer telling him they wanted over a billion videos, more than his platform has.

"You scratch your head and say, where do you get that?"

Tech giants like OpenAI, Google, Meta, Microsoft and others initially used reams of data scraped from the internet for free to train generative AI models.

They have said that doing so is both legal and ethical. But sooner than later, they were facing lawsuits from a string of copyright holders.

Because they cannot freely take anything they like and claim them as if they own them, these tech companies began quietly resource to third-parties.

They started paying other companies to obtain content that are locked behind paywalls and login screens.

The generative AI data gold rush urged AI model makers to start hedging risks and secure data-supply chains, both through deals with content owners and via a burgeoning industry of data brokers that has popped up to satisfy demand.

Months after ChatGPT debuted, for instance, companies including Meta, Google, Amazon and Apple all struck agreements with stock image provider Shutterstock to use hundreds of millions of images, videos and music files in its library for training, according to a person familiar with the arrangements.

It's said that the deals between cloud hosting companies and stock photography platforms with Big Tech firms initially ranged from $25 million to $50 million each, though most were later expanded

A Shutterstock competitor, Freepik, also said that has agreements it had struck agreements, where it's licensing the majority of its archive of 200 million images at 2 to 4 cents per image.

Rates vary by buyer and content type, but generally, tech companies are willing to pay $1 to $2 per image, $2 to $4 per short-form video and $100 to $300 per hour of longer films. The market rate for text is $0.001 per word.

Images of nudity, which require the most sensitive handling, are worth around $5 to $7.

The most expensive of all images, are images that are used to train AI systems that block content like graphic violence barred by the tech companies.

These images can include crime scenes, conflict violence and surgeries, all of which were obtained by the police, freelance photojournalists and medical students.

Sources of these images are accustomed to seeing violent injuries to anonymize and annotate the images, which are disturbing to untrained eyes.

However, the rush is shifting the trend.

Before, companies were after publicly-available data they can find on the web. This time, they're more into the private collections of the copyright holders that is not available for scraping.

What this means, these tech companies are also getting access to content locked behind paywalls and login screens, which can include everything from chat logs to long forgotten personal photos from faded social media apps.

OpenAI, Google, Meta, Microsoft, Apple and Amazon all declined to comment on specific data deals and discussions.

But what is certain, the market for the AI data market is worth around $2 billion.

Forecast suggests that it could grow to $30 billion in just a decade.

According to Photobucket CEO Leonard, licensing data is an alternative to selling ads. Because of that, he had updated his company's terms of service in October that grants it the "unrestricted right" to sell any uploaded content for the purpose of training AI systems.

"We need to pay our bills, and this could give us the ability to continue to support free accounts," he said.

Photobucket is not alone among platforms in embracing licensing.

Tumblr's parent company Automattic said last month it was sharing content with "select AI companies.

" In February, Reuters reported Reddit struck a deal with Google to make its content available for training the latter's AI models.

Even Reddit is opening itself towards data-licensing business.

But it's worth noting that the sources said that their datasets are "ethically sourced," as they are obtained from the consent from people, and that the data also have any personally identifying information removed from them.

Read: Generative AI Is Turning 'Will Never Happen' Into 'Will Happen Faster Than You Think'