The Billionaire Whose Army Fuels the AI Boom

[00:00:00] [SPEAKER_01]: With Ecolab Science Certified, we take cleaning off your plate so you can focus on what's most

[00:00:04] [SPEAKER_01]: important to your restaurant, your guests, and having them switch from giving your restaurant

[00:00:08] [SPEAKER_01]: a go to making it a go-to spot. Ecolab Science Certified, count on the scientific clean.

[00:00:13] [SPEAKER_02]: Learn more at sciencecertified.com. Welcome to Tech News Briefing. It's

[00:00:20] [SPEAKER_02]: Tuesday, September 24th. I'm Zoe Thomas for The Wall Street Journal.

[00:00:26] [SPEAKER_02]: Pledges from tech giants to cut their carbon footprints are being upended by efforts to build

[00:00:31] [SPEAKER_02]: up artificial intelligence capacity. It's spurring companies like Google, Amazon, and Microsoft to

[00:00:38] [SPEAKER_02]: team up with power producers to find creative ways to speed the development of new clean energy

[00:00:44] [SPEAKER_02]: sources. We'll tell you about them. And then, a startup called Scale AI has built an army of

[00:00:50] [SPEAKER_02]: data labelers who are doing the work needed to power the modern AI boom. Demand for scale services

[00:00:58] [SPEAKER_02]: has made its young founder and CEO a billionaire. Our reporter Burber Jinn will join us to explain

[00:01:04] [SPEAKER_02]: how Scale's business model works. But first, AI data centers require a massive amount of energy.

[00:01:15] [SPEAKER_02]: Consider this. A search on a generative AI platform like ChatGPT uses at least 10 times the energy as

[00:01:23] [SPEAKER_02]: a standard one on Google. So, as tech giants build out their AI capabilities, they'll need more power.

[00:01:29] [SPEAKER_02]: And to meet the emissions targets many of those companies made just a few years ago,

[00:01:35] [SPEAKER_02]: they'll need that power to be clean. Our reporter Jennifer Hiller is here to tell us about those

[00:01:40] [SPEAKER_02]: efforts. So, Jennifer, what are some of the new clean energy sources tech companies are looking into?

[00:01:46] [SPEAKER_03]: So, the tech companies for many years have been backing renewable energy projects and they

[00:01:51] [SPEAKER_03]: continue to do that. They're big backers of wind, solar, and now more battery projects. But they are

[00:01:59] [SPEAKER_03]: kind of in a desperate hunt for clean energy that can go around the clock. And that is just really

[00:02:06] [SPEAKER_03]: difficult to find out there. So, what we are seeing is that some of the companies are trying

[00:02:11] [SPEAKER_03]: to come up with ways to get new technologies onto the grid. So, they are doing things like backing

[00:02:18] [SPEAKER_03]: geothermal startups and trying to help them with first-of-a-kind projects. They're working with

[00:02:24] [SPEAKER_03]: utilities in different states to try to come up with maybe new rate structures where they would

[00:02:30] [SPEAKER_03]: pay essentially a higher rate that would help underpin some of the investment needed in something

[00:02:37] [SPEAKER_03]: like long-duration battery storage. So, they're trying to basically find a way to get those kinds

[00:02:44] [SPEAKER_03]: of power generation technologies out there without having regular rate payers have to foot the bill.

[00:02:50] [SPEAKER_02]: Can you give us some specific examples of these projects?

[00:02:53] [SPEAKER_03]: There's one from Google in Nevada where they are partnered with a geothermal company called

[00:02:58] [SPEAKER_03]: Fervo and they're working with Envy Energy, which is the utility there. And they've basically been

[00:03:05] [SPEAKER_03]: working on pilot projects that have been successful with Fervo and have helped back those

[00:03:11] [SPEAKER_03]: and prove it up and have started buying some geothermal power. And now what they're doing is

[00:03:16] [SPEAKER_03]: going to the state regulators to try to get the okay for a rate structure that would basically

[00:03:23] [SPEAKER_03]: have Google buy geothermal power from Envy Energy. And then for the data centers that Google builds

[00:03:30] [SPEAKER_03]: in the Nevada area, they will be able to say that they're being powered at least in part by geothermal.

[00:03:38] [SPEAKER_02]: What's the timeline for when some of these new power sources are expected to be up and running?

[00:03:42] [SPEAKER_03]: That's one of the difficult things that's unanswered in this industry because

[00:03:47] [SPEAKER_03]: the need for power is there now and it takes a few years to get new projects online and especially

[00:03:54] [SPEAKER_03]: newer first of a kind projects. So, we're seeing some new clean energy projects that will be

[00:04:00] [SPEAKER_03]: delivered this decade. But for the most part, a lot of these in bulk are not going to be coming

[00:04:07] [SPEAKER_02]: onto the grid into the 2030s. What is tech's demand for energy meant for nuclear power?

[00:04:13] [SPEAKER_03]: We are seeing just a whole turnaround in the sentiment around nuclear. There's a lot of people

[00:04:18] [SPEAKER_03]: very excited about the possibilities there and talking about maybe being able to finance new

[00:04:25] [SPEAKER_03]: projects. And then we're also seeing deals like the one announced last week between Microsoft and

[00:04:31] [SPEAKER_03]: Constellation. Constellation Energy owns a large number of nuclear reactors across the country,

[00:04:39] [SPEAKER_03]: including Three Mile Island, the undamaged reactor from Three Mile Island. And it closed down in 2019

[00:04:46] [SPEAKER_03]: basically because it was uneconomic and it couldn't compete in the electricity market there

[00:04:51] [SPEAKER_03]: and it was just losing money. And now there's a power purchase agreement between Microsoft

[00:04:58] [SPEAKER_03]: and Constellation that's going to cover the cost essentially of bringing that reactor back online.

[00:05:03] [SPEAKER_03]: That's the kind of thing that just wouldn't have been imaginable a handful of years ago.

[00:05:09] [SPEAKER_02]: All right, that was our reporter Jennifer Hiller. Coming up, we'll tell you about the 27-year-old

[00:05:14] [SPEAKER_02]: billionaire whose army of gig workers is doing the grunt work behind the AI boom.

[00:05:20] [SPEAKER_02]: That's after the break.

[00:05:28] [SPEAKER_01]: With Ecolab Science Certified, we take cleaning off your plate so you can focus on what's most

[00:05:32] [SPEAKER_01]: important to your restaurant, your guests, and having them switch from giving your restaurant a go

[00:05:36] [SPEAKER_01]: to making it a go-to spot. Ecolab Science Certified. Count on the scientific clean.

[00:05:41] [SPEAKER_01]: Learn more at sciencecertified.com.

[00:05:47] [SPEAKER_02]: Getting a response from an AI chatbot can feel as easy as typing in a prompt and hitting enter.

[00:05:53] [SPEAKER_02]: But there is a lot more that goes into it. And Alexander Wang has become one of the youngest

[00:05:59] [SPEAKER_02]: self-made billionaires by building a company that shapes the way AI models behave. His business,

[00:06:05] [SPEAKER_02]: Scale.AI, has built an army of more than 100,000 contractors who perform the grunt work that powers

[00:06:12] [SPEAKER_02]: the modern AI boom. There's such a demand for Scale.AI services that its revenue pace tripled

[00:06:19] [SPEAKER_02]: last year, boosting its valuation to $14 billion. Here to tell us more is our reporter Berber Jin.

[00:06:27] [SPEAKER_02]: So Berber, let's start with Alexander Wang. What's his background?

[00:06:31] [SPEAKER_00]: So Alexander Wang is one of the youngest founders to have built a multi-billion dollar

[00:06:38] [SPEAKER_00]: tech company in the past 10 years or so. He dropped out of MIT as a freshman to start Scale.AI.

[00:06:48] [SPEAKER_00]: And he kind of ran with this idea of not coding up a buzzy software app like a lot of founders

[00:06:55] [SPEAKER_00]: dream of doing when they're young, but taking on this unglamorous task of building out the sort of

[00:07:02] [SPEAKER_00]: labor workforce that he predicted a lot of tech companies would continue to need

[00:07:09] [SPEAKER_00]: to help with everything from customer service to content moderation. More recently, he landed on

[00:07:16] [SPEAKER_00]: what so far has been the biggest revenue goldmine for Scale, which is this huge generative AI

[00:07:23] [SPEAKER_00]: industry where a big part of training these AI models is actually having thousands and thousands

[00:07:31] [SPEAKER_00]: of humans manually write up the question and answer pairings that these models need to

[00:07:37] [SPEAKER_02]: learn how to sound more like humans. So Scale.AI is effectively having contractors label data

[00:07:45] [SPEAKER_02]: and write these prompts for generative AI. Who are some of Scale.AI's customers now?

[00:07:50] [SPEAKER_00]: Their big customers include Meta and Google. They've had a longstanding relationship with OpenAI.

[00:07:58] [SPEAKER_00]: You know, Alex Key likes to compare Scale to NVIDIA in the sense that it's one of these

[00:08:03] [SPEAKER_00]: companies that's providing a service that all these different companies in the AI race sort of need.

[00:08:10] [SPEAKER_02]: Tell us a little bit more about how Scale's business model works.

[00:08:13] [SPEAKER_00]: If you boil it down, it's basically arbitrage on human labor. So on the one end,

[00:08:20] [SPEAKER_00]: you have a bunch of contractors that Scale has recruited over the years. Alex has said

[00:08:27] [SPEAKER_00]: the numbers at around 100,000 contractors that Scale manages based all over the world.

[00:08:34] [SPEAKER_00]: And then on the other side of the equation, you have these big tech companies that

[00:08:38] [SPEAKER_00]: want this sort of invisible workforce to perform a lot of the grunt work needed to train AI models.

[00:08:46] [SPEAKER_00]: And these tech companies don't want to hire these workers themselves. And so Scale is kind of like,

[00:08:51] [SPEAKER_00]: we'll handle the back end of that. And in exchange, they basically charge a premium

[00:08:58] [SPEAKER_02]: on the service that they offer. Where does it find all those contractors?

[00:09:02] [SPEAKER_00]: Scale has workers everywhere from Kenya and the Philippines to Australia, New Zealand,

[00:09:09] [SPEAKER_00]: and the United States. They mostly advertise for these gig work jobs just online. So through Reddit

[00:09:15] [SPEAKER_00]: forums and Discord servers, and sometimes through direct LinkedIn outreach. Do the contractors know

[00:09:23] [SPEAKER_00]: who they're working for? Most of them do not because the two websites that these contractors

[00:09:29] [SPEAKER_00]: use to complete assignments for Scale, they're called RemoTasks. And more recently,

[00:09:35] [SPEAKER_00]: they rebranded it to Outlier. And neither of those two websites publicize their affiliation with

[00:09:42] [SPEAKER_02]: Scale. This sounds like it could get pretty complex. How has the company managed all of this?

[00:09:48] [SPEAKER_00]: That's one of the interesting things about Scale because on the face of it,

[00:09:52] [SPEAKER_00]: it looks like any other hot startup, but their jobs are not like any other tech company. They're

[00:09:57] [SPEAKER_00]: managing this really complex, sprawling network of contractors that are based all across the world.

[00:10:04] [SPEAKER_00]: So those types of employees at Scale are called queue managers because a lot of the work there is

[00:10:09] [SPEAKER_00]: making sure the queue is proceeding according to plan. One big thing that Scale has struggled with

[00:10:15] [SPEAKER_00]: is fraud on the platform. Last year, for example, there was a big blow up at Scale where some of

[00:10:21] [SPEAKER_00]: the data that they sent to Meta to help train Meta's AI chatbots on Facebook and Instagram,

[00:10:27] [SPEAKER_00]: they were basically created by chatbots. The contractors who were on that project had found

[00:10:33] [SPEAKER_00]: a way to make money off Scale's services without actually providing the human inputs that Meta

[00:10:42] [SPEAKER_00]: wanted. We know about Scale's stressful experience with Meta from a bunch of former employees who

[00:10:48] [SPEAKER_00]: spoke with us. What is competition like for Scale AI? Competition is very intense. You have

[00:10:55] [SPEAKER_00]: companies like OpenAI themselves that are building their own in-house labeling platforms.

[00:11:01] [SPEAKER_00]: If these tech companies are developing AI models, think it's going to be more and more important,

[00:11:07] [SPEAKER_00]: they might end up just investing their own resources to building out this platform themselves.

[00:11:12] [SPEAKER_00]: And so there are all these sorts of uncertainties around how sustainable Scale's growth

[00:11:18] [SPEAKER_00]: and business model is. The other thing to keep in mind as well is that it's unclear how long this

[00:11:24] [SPEAKER_00]: AI spending boom will last. If those concerns materialize into something more tangible,

[00:11:31] [SPEAKER_00]: the money spigot could really slow down and that would be very detrimental to Scale's business.

[00:11:36] [SPEAKER_02]: That was our reporter, Barbara Jinn. And that's it for Tech News Briefing.

[00:11:41] [SPEAKER_02]: Today's show was produced by Julie Chang with supervising producer, Catherine Millsop.

[00:11:45] [SPEAKER_02]: We had additional support from Trina Menino. I'm Zoe Thomas for The Wall Street Journal.

[00:11:51] [SPEAKER_02]: We'll be back this afternoon with TNB Tech Minute. Thanks for listening.