In today's mergers and acquisitions, it's popular to directly poach talent.

On December 24, 2025, on Christmas Eve in Silicon Valley, a brief announcement declared the demise of a potential giant in the AI chip industry.

The founder, president, and core team joined NVIDIA, but the company "continues to operate independently," and the CFO took over as CEO.

This was a company that was valued at $6.9 billion three months ago and had just completed a $750 million financing round. The money recently invested by giants like Blackstone, Samsung, and Cisco might not even have had time to "warm up."

Fortunately, the transaction price was fair: $20 billion - this is NVIDIA's largest deal in history, three times higher than its acquisition of Mellanox ($6.9 billion) in 2019. The cost was that about 90% of Groq's employees were taken away in a package and all settled in cash. The remaining 10% stayed in the "continuing to operate" shell.

The founder is the designer of Google TPU, known as the "father of TPU." This is the only player that can threaten NVIDIA in the AI inference market. Groq's LPU chip has an inference speed 10 times that of NVIDIA's GPU under specific workloads, with only one-tenth of the energy consumption and can process 500 tokens per second.

Now, this threat is gone.

NVIDIA didn't issue a press release or submit regulatory documents. The CFO said "no comment" when questioned by the media. The official statement is "non-exclusive technology licensing + talent recruitment," not an acquisition.

But everyone understands what this really is - spending $20 billion to buy the founder, the core team, and all the technology, leaving an empty shell with the name Groq, and telling the regulatory authorities "look, the competition is still there."

Silicon Valley has given this kind of operation a nice name: acqui - hire (talent acquisition).

But in 2024, acqui - hire is no longer the win - win exit it was ten years ago. It has become a new weapon for large companies to eliminate competitors - cheaper than acquisitions, more efficient than price wars, and more concealed than anti - monopoly reviews.

It's like a gentle knife.

Back then, acquisitions were still a decent thing

On April 9, 2012, Facebook announced that it would acquire Instagram for $1 billion.

At that time, Instagram only had 13 people, had been established for less than two years, and had no advertising revenue. Zuckerberg spent $1 billion on a filter app and 13 young people.

The entire Silicon Valley was shocked at that time. Some said Zuckerberg was crazy, and some said this was the Internet bubble 2.0. But three months later, everyone shut up.

Because all 13 people at Instagram joined Facebook and all got paid.

The founders, Kevin Systrom and Mike Krieger, continued to lead Instagram. The product operated independently, and the brand continued to exist. Two years later, Instagram's user base exceeded 200 million, and its valuation exceeded $35 billion. When Systrom left in 2018, Instagram's valuation had exceeded $100 billion.

Everyone won in this deal.

The founders made money and continued to work on the products they liked; all 13 employees received generous stocks, and no one was abandoned; Facebook got the future of the mobile end; and the investors, Benchmark and Sequoia, made dozens of times their investment.

In Silicon Valley in 2012, this kind of deal was called acqui - hire (talent acquisition). At that time, this term was not derogatory.

One year later, Google spent $1.1 billion to acquire Waze.

Waze is an Israeli navigation app company with more than 100 employees. After the acquisition, Waze continued to operate independently in Israel, retaining its brand, team, and product roadmap. The founder, Noam Bardin, continued to serve as CEO until 2021.

All more than 100 employees stayed with the company and received benefits.

In 2013, Apple CEO Tim Cook explained in an interview why Apple acquires small companies: "We acquire teams to add value. We look for people who understand entrepreneurship and can adapt to the scale of a large company."

This sounds high - sounding, but in the early 2010s, Silicon Valley really did things this way.

Acqui - hires in that era had a common feature: the company remained intact after the acquisition.

It's not just Instagram and Waze. YouTube continued to operate independently after being acquired by Google, and the founders used the money to continue working on the product. Tumblr's founder continued to serve as CEO after being acquired by Yahoo. WhatsApp's founders joined the board after being acquired by Facebook.

These acquisitions, of course, had premiums and bubbles, but there was a bottom line: the people were intact, the company was intact, and the product continued to develop.

The founders got the money not to leave but to make the product better on a larger platform. The employees received stocks not as severance pay but as recognition of their past contributions.

Silicon Valley at that time believed in a logic: acquisitions were not to kill competitors but to make good things grow bigger.

More importantly, there was a "cycle after exit" in that era.

In 2002, PayPal was acquired by eBay for $1.5 billion. Logically, the story should have ended there - a group of young people in the payment industry took the money and retired.

But it didn't.

This group of people from PayPal later founded or invested in: Tesla, LinkedIn, YouTube, Yelp (Jeremy Stoppelman), and Palantir.

Silicon Valley calls them the "PayPal Mafia." The value they created later far exceeded the $1.5 billion acquisition price back then.

They learned how to scale, how to manage a large company, and how to handle complex regulatory and compliance issues at eBay. Then, with this experience, they went out to start new businesses, and their success rate was much higher than the first time.

In 2013, an investor in Silicon Valley, Peter Relan, proposed the theory of a virtuous cycle of "start - up - acquisition - re - start - up":

"Entrepreneurs start their first business, make a good product and get acquired; learn how to scale in a large company; then go out and start a new business, and this time the success rate will be much higher. Acqui - hire is not the end but a stage in the growth of entrepreneurs."

This sounded very reasonable in 2013. Because the stories of Instagram, YouTube, and PayPal all confirmed this logic.

Acquisitions in that era also had a characteristic: people knew how the money was distributed.

Although the distribution ratio varied from company to company, the general rule was:

The founder took 40 - 50%, the core team took 30 - 40%, and the investors took 20 - 30%. When WhatsApp was acquired by Facebook for $19 billion, the founder Jan Koum took $6.8 billion, and the co - founder Brian Acton took $3 billion. But the 55 employees of the company also each received an average of tens of millions of dollars.

In 2014, The New York Times interviewed several employees of acquired start - ups. One of the engineers said: "I know the founder made much more money than me, but I also got money that changed my life. More importantly, the product we made is still there, and we're still working together."

2024: What happened in nine months

But this era gradually came to an end.

From March to August 2024, three deals were completed one after another. Microsoft spent $650 million to poach the founder Mustafa Suleyman and most of the team from Inflection AI. Google spent $2.7 billion to recruit Noam Shazeer and 30 core engineers back from Character.AI. Amazon poached the founder David Luan and 66% of the employees from Adept.

All three companies were valued at over $1 billion, had raised hundreds of millions of dollars in financing, and were led by star founders. But none of them were formal acquisitions. They were all "technology licensing + talent recruitment." The large companies paid a licensing fee, took the people away, and left an empty - shell company.

The flow of money was clear. The Character.AI deal was the most transparent: Shazeer took $7.5 - 10 billion, the 30 engineers who went with him received an average of tens of millions of dollars each, and the investors got a 2.5 - fold return. What about the remaining 220 employees? Their stock options continued to vest, but the company only had 18 months of cash left and had abandoned self - developed models and switched to Meta's open - source Llama. The interim CEO - the former general counsel - said: "We've basically given up competing with OpenAI, Google, and Amazon."

The investors in Inflection got a 1 - 1.5 - fold return - in the Instagram era, this was called barely breaking even. The investors in Adept got back $415 million, exactly a 1:1 return, with no profit. The company retained $25 million and 20 employees. Amazon's statement was very cautious: "This is not an acquisition. We're not interested in owning Adept's business."

Compare this with 2012. In the Instagram acquisition, all 13 people got paid, a 100% benefit rate. In Character.AI, only 30 out of 250 people really benefited, 12%.

This path was paved ten years ago. In 2013, the San Francisco start - up Hooked wanted to sell itself to Apple. Apple said it only wanted the engineers, not the technology. Hooked refused, and couldn't find a buyer after a few months. When it was running out of money, the CEO voluntarily proposed: sell 3 engineers to you and you pay the headhunter fee. Apple said it would consider it and then directly contacted the engineers and hired them without paying a single cent to the company.

Hooked had to sue. As a result, in 2020, the California Court of Appeal ruled that it's not illegal for a company to poach employees from a competitor unless there are independent illegal acts. This ruling gave the green light to large companies: they can contact employees during negotiations and then directly poach them without paying the company.

By 2024, this had become the norm. The Federal Trade Commission has been keeping a close eye on tech giants in recent years. If Google wanted to formally acquire Character.AI, the approval process would take one or two years, and it might not even be approved. But "technology licensing + poaching" is not considered an acquisition legally. Google's statement is: we pay a licensing fee and hire a few people by the way. The company is still there, the brand is still there, and the competition is still there. The UK Competition and Markets Authority later determined that the Inflection deal "constituted a merger" but did not hinder competition.

The characteristics of AI companies also determine that the model of poaching core personnel is more cost - effective. Instagram's value lies in its 100 million users and social network, so it had to be bought as a whole. The core of an AI company is technology and a few key engineers. Character.AI has 250 people, but Google only wants 30 who master the core technology. Google doesn't need the others who are in product, operation, and customer service.

This is not necessarily a bad thing for investors. Character.AI was valued at $1 billion, but its product burned a lot of money, and commercialization was far off. Should they wait for an IPO or accept Google's $2.7 billion offer and exit immediately? The investors chose the latter. Although $2.7 billion is not much, a 2.5 - fold return can be secured. They don't care how much the employees' equity is worth.

The impact is already evident. The proportion of 2024 graduates from Stanford's Computer Science Department choosing large companies is 15 percentage points higher than last year. The number of applications for Y Combinator's 2024 fall session is 20% less than last year. It's getting harder and harder for start - ups to recruit core engineers.

The logic is simple: you get cash at Google and Microsoft, but stock options at a start - up. In the past, stock options could turn into tens of millions of dollars, but now they're probably worthless.

Over the past 20 years, there has been an implicit contract in Silicon Valley: take risks with the founder, share the money if successful, and get a decent amount of money even if the company is acquired - a "soft landing." In 2024, this contract was broken. The founder can still get the money, but the employees can't.

A partner at the investment institution Khosla Ventures summarized: "Early - stage employees are the biggest losers in this kind of deal." Hundreds of engineers have witnessed the same thing: you thought you were starting a business, but in fact, you were just a bystander. The founder takes the money and leaves, and you're left holding worthless stocks.

China: A different ecosystem

Also in October 2024, OPPO announced the acquisition of the AI writing assistant company Waveform Intelligence. The founder Jiang Yuchen, the COO, and the CTO joined OPPO.

Waveform Intelligence was established in May 2023 and completed its Pre - A round of financing in January 2024, led by Lanpao Venture Capital, with the personal participation of Jing Xiandong, the chairman of Ant Group. The product, Wawa Writing, has a cumulative user base of 300,000 and has generated 20 billion characters.

It took 19 months from its establishment to the acquisition. It took 9 months from the financing to the acquisition.

The acquisition price was not made public. Waveform Intelligence's registered capital is 1.34 million RMB, with a paid - in capital of 240,000 RMB. It raised tens of millions in the Pre - A round, and its valuation was between 50 million and 100 million RMB. Compared with Groq's $20 billion, this difference in magnitude is huge. Including the recent acquisition of Manus, which only offered an acquisition price of 30 million RMB in China, the premium for acquisitions in China has been greatly compressed.

But this doesn't mean that Chinese AI start - ups are not valuable. It's just that the rules of the game in the two markets are completely different.

When ByteDance wanted AI talent, it directly poached Zhou Chang from Alibaba with an eight - figure annual salary and took away more than a dozen people. When Xiaomi wanted experts in large models, it directly poached Luo Fuli from DeepSeek with an annual salary in the tens of millions.

The cost of directly poaching people is much lower than acquiring a company. If Jiang Yuchen's team only had a few core members, giving each of them a few million in annual salary, the total cost could be solved with tens of millions. There's no need to bear the company's debts, no need to resettle other employees, and no need to go through the acquisition process.

The implementation of non - compete agreements in China is relatively lenient, and the legal risk of job - hopping is smaller than in the United States. Although Zhou Chang had a non - compete dispute with Alibaba, he still went to ByteDance in the end.

So Chinese large companies don't need to pay billions of dollars for "technology licensing + poaching" like Microsoft and Google to avoid anti - monopoly regulations. If they want people, they just poach them directly. If they want technology, they buy it or develop it themselves. The market environment determines the transaction structure.

Waveform Intelligence's product positioning also exposes the dilemma of vertical models. Wawa Writing targets online novel writers and content creators. The users' payment ability is limited, but the computing power cost for long - text generation is extremely high. The product has users, but there's no clear path to profitability.

In the second half of 2024, it became very difficult for AI applications in vertical scenarios to raise funds. Investors are more willing to invest in general large models or applications that already have revenue. Waveform Intelligence's 19 - month journey sent a signal to the market: the commercialization window period for vertical models is very short.

This is in contrast to the situation in the United States. Groq was working on AI inference chips, which directly threatened NVIDIA's core business, so NVIDIA was willing to pay $20 billion. Waveform Intelligence was working on a vertical application, which didn't pose a threat to any large company and didn't have core technology that must be owned. Naturally, the acquisition price couldn't go up.

But this doesn't mean that the AI start - up environment in China is worse. The characteristic of China is not that large companies acquire start - ups, but that large companies "purchase" the products of start - ups. In this model, if a start - up's product is really competitive, it can survive and grow through orders. If the product is not good,