- Start-up Scale helps companies label data for use in artificial intelligence models.
- It has been trying to keep up with one of its customers, DoorDash, which saw revenue accelerate as the coronavirus took hold.
- Scale is now worth $3.5 billion and breaking even.
AI has been around for decades, but building an AI model that does what it's supposed to do isn't always fast or easy. Part of the drudgery is augmenting raw data so that a model can learn to make accurate predictions.
Scale AI, founded in 2016, specializes in performing that work for its clients. Customers include Airbnb, General Motors, Nvidia, OpenAI, Pinterest and DoorDash. Now Scale is is breaking even, founder and CEO Alex Wang told CNBC on Monday, even as it more than doubled its annualized revenue run rate year-over-year in the third quarter. In a major validation for its business model, Scale announced Tuesday that it has raised $155 million, led by Tiger Global, at a valuation over $3.5 billion.
The AI market isn't as large and easy to understand as the market for, say, social media. It amounts to an enabling technology, like database software, in the sense that AI can be embedded into existing products. Still, it's a real, and growing, market.
In revealing its plan to go public last month, C3.ai, a company looking to trade under the ticker symbol "AI," said IDC estimates the enterprise artificial intelligence software market will be worth $44 billion in 2024, up from $18 billion in 2020. And software makers have continued to pick up AI tools through acquisition. ServiceNow, for example, announced the acquisition of burgeoning industry-oriented start-up Element AI on Monday.
"In 2014 or 2015, a lot of this stuff started kicking into high gear," said Wang. "It's such a fertile new industry for us to operate in, and it's been growing so quickly in past few years."
Scale can handle a variety of data that customers send to it. Contractors review the data and annotate it as needed. For example, the contractors can look at photos and count the number of cars in each picture. A model that's been trained on that refined data can then evaluate a new photo and estimate how many cars are present.
There are other options for farming out these tasks, including Australian company Appen, which acquired start-up Figure Eight for $300 million in 2019. There's also a 15-year-old service from Amazon called Mechanical Turk, which farms out small tasks to many workers. Wang said Scale strives to stand out by focusing on doling out very high-quality data and solving problems with technology at scale.
Scale, which charges based on the amount of data it processes for customers, has seen major growth by working with DoorDash. The delivery-service has experienced accelerated growth this year as people order food deliveries during the coronavirus pandemic. "We're frankly just trying to keep up," Wang said.
San Francisco-based Scale sought to be more conservative with its money as the pandemic created a U.S. recession.
"I think we were fortunate," Wang said. "We were able to mostly kind of be a bit less aggressive than we were going to be." The company slowed the pace of hiring, he said. The idea is to remain helpful to its customers — Wang doesn't want the company to just disappear one day.
The company's financial position is enabling it to make its own investments. For example, Scale said Tuesday it's acquiring Helia AI, a start-up that was building software to help companies run AI models on video streams in real time. The team includes people who worked on Tesla's Autopilot system.
"There's a certain level of vision and experience with working with very large-scale machine learning problems that we're able to gain through the acquisition," Wang said.