The Fuzzy Math Behind Scale AI’s Valuation

And yet the most recent fundraising values Scale at more than 25 times this year’s expected gross profits, measured before operating costs. That’s slightly higher than for privately held enterprise software firm Databricks, which helps companies store and manage large amounts of data.

Even giving Scale some benefit of the doubt that it will remain central to an ongoing generative AI boom, it probably shouldn’t be valued higher than $10 billion today.

To be fair, valuing Scale isn’t a straightforward exercise. I’ve tried to get closer to the answer after obtaining its financial forecasts and investor presentations, and talking to investors that either bought in or turned it down, as well as former employees, customers and competitors. First you have to get your head wrapped around the question: What exactly is Scale?

What Scale is not is a cloud software company like Databricks. Scale expects to have a 53% gross profit margin this year, far below the median 76% margin for publicly traded cloud software companies and 78% for Databricks. While Scale builds software tools for customers, its core service is supplying human contractors for AI training. And that is expensive.

Scale is growing faster—its revenues doubled to $330 million last year and are expected to triple to over $1 billion this year, whereas Databricks grew topline more than 50% last year and is expected to increase revenue 60% this year.

But Scale’s growth has, in part, been supercharged by a shift over the past year from low-wage workers in developing countries to higher-paid experts training LLMs. “Revenues are high because the human salaries are high, and you’re making money on top of it,” one executive at a Scale competitor said.

A lot of Scale’s revenues are also concentrated in only a handful of large customers, like Alphabet and Microsoft, according to investors I spoke to. Cloud software companies like Databricks are usually less focused, with hundreds or thousands of customers, reducing the risk that revenues will drop if a big customer flees.

Scale’s revenue also isn’t as certain to recur in the years to come, as a subscription software firm’s would. Wall Street tends to prefer the latter business model. As if to remind people of that, the “ARR” Scale cites in its investor presentations stands for “annualized run-rate revenue” rather than “annual recurring revenue.”

Instead, many of its customers require separate, bespoke projects. Scale’s customers have been spending more with it each year recently, to be sure. But if AI firms’ needs change, it’s unclear how Scale’s business will evolve.

Yet on a multiple of next year’s expected gross profits—which the company expects to fatten over time as it automates more work—Scale is still valued slightly higher than other top software firms like Atlassian, Confluent and MongoDB. (For public companies, future gross profit estimates are based on S&P Capital IQ. Databricks’ and Scale’s estimates are based on last year’s forecasts, which I obtained.)

Sum of the Parts

So what is Scale worth? Maybe the best way to figure that out is to put a separate valuation on its two business lines—what the company calls infrastructure and applications.

The larger infrastructure work of data annotation and LLM fine-tuning that it does for Alphabet, Meta and OpenAI has a similar look and shape to the work of TaskUs, a little-known firm that handles data label and customer service outsourcing.

I could argue that Scale deserves twice the gross profit multiple of TaskUs, which trades at 3 times forward gross profits but has grown much more slowly and spends much less on research and development than Scale does. At 6 times next year’s gross profits, this business line would notch a $3.3 billion valuation for Scale.

Scale has also been building an applications business to “power every enterprise’s AI capabilities,” starting with companies like Morgan Stanley, Chegg and Accenture, as well as military agencies including the U.S. Army and the Department of Defense. Scale expects applications to have higher gross margins of about 65% this year, and anticipates that it will generate 22% of revenue, or more than $220 million.

This business line looks similar to the consulting-style tech and AI work handled by Palantir, which gets a 20 times multiple on next year’s gross profits. Because of the uncertainty Scale still faces in building up that business, investors will probably apply a slightly lower multiple on Scale’s expected gross profits from it—call it 17 times—to value the revenue at $5.4 billion.

Even that may be generous: Two investors who considered investing in Scale told me the startup lacked the strong sales staff required to sell its application services to more big companies. Wang himself was taking a lot of that on.

Combine the two business lines, and you get a valuation that’s a hair above $10 billion.

Silicon Valley Logic

Of course, this analysis doesn’t take much into account Silicon Valley logic: put a heavy weight on the high-upside scenario or look like a fool.

In a May interview, I asked Dan Levine, a venture capitalist at Accel who sits on Scale’s board and led the startup’s first and most recent round, whether Scale was more of a low-margin services business or a high-margin software company. “I think time tells,” he said. “People have always discounted all the great companies, from the Airbnbs to the Stripes to the Dropboxes.”

He felt confident that Scale’s “high-quality data at the lowest price point will always be very valuable.”

The startup’s bull case, laid out in its investor presentation, is that AI development will need constant training and retraining. “As models get bigger, they need more data,” the company said in its pitch deck. And as Scale reduces its labor expenses through more automation and wage cuts, its margins should improve. It says it will already be Ebitda break-even, on an adjusted basis, this year.

Wang himself may be the strongest positive point for investors. He has cultivated a reputation as an AI prodigy who gets in the room with the right people and steers the company in the right direction for every role it plays, from data labeler for self-driving cars, to government contractor, to LLM whisperer.

One Scale investor I talked to admitted to “bumpiness and uncertainty” in the startup’s business over time. But he said Wang was “someone we trust to figure it out.”

Notably, Scale didn’t allow any mutual fund investors to put money into the round, we reported. Such investors could publicly disclose Scale’s valuation each month.

Given the valuation’s fuzziness, leaving those investors out of the picture was likely a savvy move.