It scored a $249 million contract last year to provide a range of AI tech to the Department of Defense. Scale also counts the Army, Air Force, the Marine Corps University and military truck maker Oshkosh among its individual customers.
In May, Scale became the first AI company to have a “large language model” — the tech behind chatbots such as ChatGPT — deployed on a classified network after it signed a deal with the Army’s XVIII Airborne Corps. Scale’s chatbot, known as “Donovan,” is meant to summarize intelligence and help commanders make decisions faster.
To Wang, who describes himself as a “China hawk,” the stakes are high: Without AI developed by private tech companies, the United States won’t be able to maintain its technological edge over the rising military power of China.
“Data is ultimately the ammunition of AI warfare,” he said in a recent interview, repeating a line he has used in conferences and during a congressional hearing in July. And the United States is already behind in stockpiling that ammunition, Wang said.
The U.S. military has made AI a key part of its strategy for the coming decades, laying out plans to field autonomous ships and planes to back up human-piloted machines, use algorithms to improve logistics by predicting when certain parts should be replaced, and scanning drone footage with image-recognition tech to free up human analysts.
Scale has benefited from being ahead of the latest AI boom, triggered last year when OpenAI released ChatGPT to the public. Scale has raised hundreds of millions of dollars and was valued by its investors at over $7 billion in 2021, making Wang the world’s youngest self-made billionaire at the time, according to Forbes.
But competition for military contracts is fierce. Big Tech companies such as Google, Microsoft and Amazon are all aggressively trying to court the Pentagon. In December 2022, those three firms, plus business software company Oracle, were awarded exclusive rights to bid for $9 billion worth of cloud computing contracts across the Defense Department, which could eat up some of what Scale is hoping to compete for. A growing group of other start-ups including Shield AI and Helsing are raising significant amounts of money and working to sell their tech to the military too.
(Amazon founder Jeff Bezos owns The Washington Post, and the newspaper’s interim CEO, Patty Stonesifer, sits on Amazon’s board.)
Scale has taken hits to its reputation because of its work in the Global South, where it employs thousands of moderators in what some call “digital sweatshops.” In January, the company laid off 20 percent of its staff, part of a wave of layoffs in the tech industry, and a Washington Post investigation found that dozens of Scale contractors in the Philippines hadn’t been paid on time or at all for work they did. A Scale spokesperson at the time said delays and interruptions to payments are “exceedingly rare.”
Arms control advocates have protested the use of AI by militaries out of concern that it will gradually remove humans from key decisions, including who or what to target on the battlefield. Some weapons have had autonomous capabilities for years, and drones that can automatically recognize targets and divebomb them without final permission from humans are already in military arsenals around the world. The U.S. military says a human will always be “in the loop,” but studies have shown that people are likely to follow the advice of authoritative-sounding machines rather than trust their own judgment.
Wang didn’t set out to become a military contractor. He founded Scale after dropping out of MIT at age 19 with Lucy Guo, a fellow intern at the question-and-answer site Quora. At the time, breakthroughs in AI research were leading to algorithms that could detect images and seamlessly translate language. But the reams of images and videos AI labs were pulling off the web to train their data needed to be labeled to help teach the algorithms what they were looking at. Wang and Guo built Scale to help solve that problem, hiring contractors around the world to painstakingly add labels to images and then charging for the service.
In 2018, two years after founding the company, Wang traveled to China for a learning tour of the country’s booming AI scene. As he walked into the office of a facial recognition start-up he was visiting, a giant screen played a video feed of the lobby, identifying the demographic information of everyone who entered, and in the case of Chinese citizens, displaying their names and biographical details alongside their faces.
“It was meant to be a demo of their technology, but it’s also just terrifying,” Wang said of seeing such a blatant display of surveillance. The casual presentation of how commercial tech could be used for government purposes underlined to Wang how the close relationship between China’s tech companies and its military was giving the country an edge that might allow it to pass the United States’ decades-long dominance in high technology.
At the same time, the gap between American companies and the government only seemed to be growing. Wang’s trip happened around the same time that Google said that it wouldn’t renew a contract to provide AI to help the military parse through drone footage after many of the company’s employees objected to the program, known as Project Maven.
“It felt like this noticeable rift in tone that if it continued was going to result in China having AI military dominance over the United States,” Wang said.
Some tech executives and investors have long argued that navigating the military’s bureaucratic procurement process is too complicated and time-consuming for companies that also serve regular commercial customers. Many AI researchers are deeply skeptical of the U.S. government’s motives and fear putting AI in the hands of powerful militaries could lead to increased surveillance, and even the potential for AI getting out of human control and causing real-world damage.
“As with each arms race, each side’s arming of itself is justified by the other’s arming of themselves,” said Lucy Suchman, a retired professor from Lancaster University in the United Kingdom who researches how AI is used by militaries. “It’s a self-perpetuating and self-intensifying cycle.”
Wang acknowledges the risks of rushing advanced AI into the battlefield and said the new tech will always need rigorous testing before being deployed.
A spokesperson for Microsoft declined to comment. Spokespeople for Google and Oracle did not respond to a request for comment.
Concerns that China will invade Taiwan and Russian President Vladimir Putin’s invasion of Ukraine have pushed the tech industry back toward government, said Michael Brown, a partner at defense-focused venture capital firm Shield Capital and the former director of the Pentagon’s Defense Innovation Unit — which serves as the military’s embassy to the tech industry.
In 2022, Google, Facebook, Microsoft, Apple and Amazon spent a combined $198.9 billion on research and development, while the Pentagon spent $132 billion, including on testing and evaluating new weapons and tools.
“Defense is ripe for disruption from both Big Tech and newer entrants,” said Nathan Benaich, founder of Air Street Capital, a venture capital firm that has invested in a handful of military tech companies.
Scale is also pushing forward into providing tools of its own.
Its military-focused ChatGPT competitor Donovan is already being tested by military units and by students at the Marine Corps University. Scale is pitching the bot as an “AI-powered decision platform” that can crunch intelligence data from different sources and provide recommendations to human officers.
A demo of Donovan shows the chatbot identifying a suspicious Chinese vessel near Taiwan, then providing options for an officer to find more information, such as dispatching a plane to do a flyover or pulling recent satellite imagery. Once the images are available, image-recognition algorithms identify high amounts of radiation, prompting the officer to pass the information up the chain of command and dispatch a drone to investigate the ship.
But selling to the military is difficult, even without intense competition, Benaich said.
“Defense is a customer like no other, and it requires deep institutional expertise to succeed. Selling to the Army isn’t like selling to Uber,” he said.
Wang said Scale still has a wide range of customers and doesn’t need military money to be successful. He said working with the Pentagon comes from the company’s desire to help the United States maintain its power in the world even as the 21st century throws more conflicts and complications at the world order.
“In this extremely abrupt and sharp technology transition, if Scale can be the company that helps ensure that the United States maintains this leadership position, that is a very real and very tangible impact,” he said.
