Geniuses never play by the script, especially in Silicon Valley.

The legendary story of Bill Gates dropping out of school to start a business has long been known to everyone, and with Meta spending $14.3 billion to acquire Scale AI founded by Alexandr Wang, the 28-year-old genius once again verified this iron rule with his own experience.

▲Alexandr Wang

According to the official announcement, Alexandr will continue to serve as a member of the board of directors of Scale AI on the one hand, and will pack up his own team to join the super intelligent team personally formed by Meta CEO Zuckerberg on the other.

After Llama 4 suffered a Waterloo, Meta and Zuckerberg urgently need to fight a turnaround battle. This urgency is real, but the question is, is it really worthwhile to bet so much on Alexandr, who does not come from a technical background?

An MIT dropout has become the king of outsourcing AI training

In 1997, Alexandr was born in Los Alamos, New Mexico, USA.

This place, which sounds a little strange, was the main battlefield of the "Manhattan Project" in the United States during World War II, and the atomic bomb was born here. In Alexandr's memory, this small town is surrounded by national laboratories, and the daily programs are classical music concerts, Halloween science classes on low-temperature physics, and scientific heritage everywhere.

His parents are also Chinese immigrants who worked as nuclear physicists at Los Alamos National Laboratory.

Since childhood, Alexandr has shown a strong interest in mathematics. In his childhood, he won the MATHCOUNTS mathematics competition and got a chance to go to Disney. It was the first time he went far away in his life. It was also at that moment that he fell in love with the fun of solving problems.

Interest brings driving force. He has participated in more competitions. In 2013, he was selected for the US Mathematical Olympiad project and in 2014, he was selected for the US Physics Olympiad national team. These experiences may have laid the foundation for his later success.

At the age of 7, he was admitted to the Massachusetts Institute of Technology (MIT) and took a graduate-level machine learning course in the first semester.

At the same time, he was not idle. He first worked as a software engineer at Addepar and joined Quora a few months later. At Quora, he was quickly promoted to technical director and led the team to promote the implementation of various indicators of the infrastructure team.

In 2016, he resigned from Quora and moved to Hudson River Trading as an algorithm developer.

But what he really wanted in his heart was to start a business, so Alexandr made a far-reaching decision to drop out of MIT and join the well-known startup incubator Y Combinator. Little-known fact is that the CEO of Y Combinator at that time was Sam Altman, the current CEO of OpenAI.

▲ Sam Altman on the left, Alexandr Wang on the right

It is rumored that Alexandr was roommates with Altman for several months.

He later mentioned in an interview that he told his parents at the time: "I told my parents that this was just a summer project, and I never went back to school."

His experience at Quora gave him a deep understanding of the difficulties of infrastructure and data management in machine learning systems, and it was these difficulties that became the breakthrough for Scale AI. In 2016, he co-founded Scale AI with Lucy Guo, also a former Quora employee, focusing on the most basic but little-known key work behind the development of AI: providing large-scale, high-quality human-annotated data.

There is also a long-standing story about the opportunity to create Scale AI.

It is said that Alexandr realized very early that AI and machine learning would change the world. In his words, "We initially built machines that could do arithmetic, but it was an exciting technological breakthrough to enable them to perform more complex tasks that require human-like understanding."

One day, he tried to install a camera in his home refrigerator to determine whether the milk was about to run out. A few weeks later, he found that he could not get enough data to train the system to accurately identify the contents of the refrigerator. This made him realize that data would be one of the key obstacles to AI breakthroughs in the next 20 years.

As a result, he founded Scale, aiming to become "the data infrastructure that drives AI transformation."

In 2016, there was still a long time before ChatGPT became popular. In contrast, the popularity of autonomous driving was the hottest topic in Silicon Valley at the time. Scale AI focused on serving the autonomous driving track in the early stage, providing image recognition data for the vehicle system to solve the "data shortage" of AI visual training.

With its refined labeling services, Scale AI gradually established a reputation and won the trust of early customers.

As a 19-year-old entrepreneur, it is not easy to gain a foothold in this track. But Alexandr took a very pragmatic strategy, bringing his notebook and product demo to the top computer vision conference CVPR, and promoted the product from booth to booth.

In 2019, Scale AI received a $100 million investment from Founders Fund of PayPal co-founder Peter Thiel, and officially became a "unicorn". A few years later, Scale received another $580 million in financing, with a valuation of $7.3 billion.

▲ Peter Thiel, a famous Silicon Valley investor

During this period, Alexandr and Guo both made the list of enterprise technology in the "30 Under 30" list of Forbes. Not long after, Guo left the company because of differences in product vision and development path, but there were also rumors that he was expelled.

Cold knowledge, after Meta announced the acquisition of Scale AI, Lucy Guo also surpassed the 35-year-old pop singer Taylor Swift and became the youngest self-made female billionaire because of her holding of Scale AI shares.

After Guo left, Alexandr continued to take the lead.

The three elements of AI development are inseparable from algorithms, data and computing power. Large language models (LLMs) require huge data sets for training. The larger the large language model, the more valuable the data. Scale AI's data outsourcing factory has become increasingly important.

By hiring thousands of contract workers to screen, label and clean data, and then providing these organized data sets to technology giants for model training, Scale AI's customer list spans technology companies and traditional companies, including Waymo, Toyota, Honda, Alphabet, Accenture, OpenAI, etc.

Of course, along the way, Scale AI has not been all bright. After ChatGPT became popular, Scale AI frequently appeared on international headlines. In addition to the myth of the genius boy's wealth creation, more exposure was his outrageous remarks and negative information such as exploitation of labor.

According to The Washington Post, in the Philippines, one of the world's largest digital outsourcing centers, at least 10,000 workers provide data annotation services for Scale AI through the Remotasks platform.

However, through testimonies from dozens of current and former employees, as well as platform screenshots, payment records, internal notices and other materials, it was found that these workers were paid very low wages, delayed payments, and even cancelled without reason, which has become the norm. And the complaint channels are almost non-existent.

In sharp contrast, in 2024, Scale AI's revenue was about US$870 million, and before being acquired, it was expected that revenue would double to US$2 billion in 2025, and its valuation was expected to hit US$25 billion.

$14.3 billion, a sky-high talent "merger and acquisition"

Before 2025, Meta had always been the leader in the field of open source models, until the emergence of DeepSeek at the beginning of the year, which disrupted Meta's rhythm. It was even reported that Meta employees broke the news that the company's executives' salaries were higher than DeepSeek's training costs.

The Llama 4 model, which was hastily prepared for the battle, was criticized in the public opinion storm because of suspected cheating, and the reasoning class and the maximum parameter version model were also nowhere to be seen. Technology lags, talent loss, product difficulties, Turing Award winner Yann LeCun is still in charge of the AI ​​research route, and it is difficult to reverse the decline.

It is hard not to say that Meta fell into the darkest moment in April.

Zuckerberg, who chose AIl In AI, will naturally not restrain his ambitions. His goal is to integrate AI into all the company's products, including Ray-Ban smart glasses and social matrices such as Facebook, Instagram and WhatsApp.

AI is the most important part of this, and it is also the part that cannot be held back.

Based on this, we can see that Meta has been active in poaching people recently. Zuckerberg personally called, sent text messages, and sent emails to researchers from companies such as OpenAI and Google, and even offered a nine-digit price to try to poach them.

According to The Information, when he was far behind his competitors, Zuckerberg increasingly consulted an atypical technical figure. Yes, it was the protagonist of this article: Alexandr Wang.

The 28-year-old young man gave Zuckerberg a lot of practical suggestions.

Even Zuckerberg began to quote Alexandr's feedback on AI issues in internal meetings. He believed that Alexandr had first-hand experience working with multiple AI research labs and could accurately grasp what kind of data these labs were pursuing and how to optimize models.

More importantly, it was Zuckerberg himself who proposed to hire Alexandr to lead Meta's super intelligence team.

Over the past year, Zuckerberg has also approached other candidates, such as Google's chief AI scientist Koray Kavukcuoglu and former OpenAI CTO Mira Murati, but in the end, he returned to Alexandr.

One of the reasons is that Alexandr has a good relationship with Meta's chief product officer Chris Cox and other executives. Although Scale AI is not directly involved in the development of the most cutting-edge AI models, Alexandr's understanding of the industry's development path and his control of basic capabilities have won Zuckerberg's trust.

What's more, Scale AI's revenue of $870 million last year also fully proves his outstanding business capabilities.

There is nothing new under the sun. This acquisition is essentially a standard Silicon Valley-style talent acquisition. In the name of acquisition, large companies actually hire the founders and core employees of a startup, and the acquired company often stops its original business after the transaction.

Silicon Valley is no stranger to this acquisition model, but the AI ​​wave has driven the booming development of this model.

▲ Mustafa Suleyman

Microsoft once acquired AI startup Inflection for $650 million in "licensing fees", essentially to hire founder Mustafa Suleyman and his team; Google acquired Character.AI's "cooperation license" for $2.7 billion, and the focus was also on its founder Noam Shazeer and key technical personnel.

Meta's acquisition of Scale this time is essentially the same script.

This acquisition is Meta's second largest acquisition in history, second only to WhatsApp's $22 billion that year. For Meta, whose cash reserves exceeded $70 billion, using money to exchange for talent and achieve the effect of buying horse bones with a thousand gold coins does not seem to be a loss-making deal.

There is no doubt that the news of Meta's acquisition of Scale AI has shocked the entire industry.

On the one hand, this move will bring rich returns to Scale AI shareholders - early investors including Accel, Index Ventures, Founders Fund, etc. will retain the remaining shares while partially cashing out.

On the other hand, the involvement of such a large company has also caused concerns among other customers of Scale AI, especially whether Scale AI will lose the neutrality of the platform after being acquired by Meta, and there is a risk of customer data being leaked to Meta.

According to the latest news, Google is considering completely terminating its cooperation with Scale AI, and the data services originally planned to be used for the training of its next-generation Gemini model, which are worth about 150-200 million US dollars, will be transferred to other suppliers. OpenAI executives have also publicly stated that they do not want such acquisitions by giants to destroy the AI ​​ecosystem. If the parties exclude each other, it will slow down the pace of innovation in the industry.

At the same time, competitors are taking advantage of the situation.

The CEO of data labeling company Labelbox publicly stated that it is expected to take away a large number of customer contracts from Scale this year; the CEO of Handshake even said that after the news of Meta's acquisition was made public, customer demand soared by two times overnight.

Although Scale's valuation has hit a new high with the support of Meta, how to appease customers and maintain the neutrality of the data service platform will be the major challenges that the company will face next.

As the acquirer, it is also unknown whether Meta can really turn things around with this acquisition.

The Llama series has not yet completely turned around, and rivals such as OpenAI, Google, and Anthropic are still strong. Meta must run AI in the social matrix and run reasoning capabilities on new hardware such as smart glasses, and at the same time take into account product experience, talent recruitment, and ecological discourse power.

Alexandr's joining could be a turning point, or just another expensive and disappointing gamble.

For now, everything is just a bet.