news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Author: Jessica
Email: [email protected]

Character.AI has finally sold itself, or more accurately, the founder of Character.AI has finally sold himself.

On August 2, AI chatbot unicorn company Character.AI suddenly announced a major "acquisition" agreement with Google. Google will pay a licensing fee to use Character.AI's technology and models, and directly recruit its two co-founders and several researchers.

In terms of the specific transaction details, the shares held by the original investors will be bought out at $88 per share, bringing Character.AI's valuation to $2.5 billion, about 2.5 times its Series A financing in 2023, but nearly half of the $5 billion disclosed last year. Unvested employee options will continue to be granted at $88 per share before the end of July 2026, paid through Google's agreement payment.

What has caused even greater controversy is that Character's former CEO and one of the authors of the Transformer paper, Noam Shazeer, and President Daniel De Freitas will return to their former employer, joining the DeepMind research team, nearly three years after leaving Google. General Counsel Dom Perella will take over as interim CEO.

Not only that, 30 of the company's approximately 130 employees who are engaged in model training and voice artificial intelligence will also be incorporated into Google's Gemini AI department. At the same time, Character will no longer use internal models, but will instead use open source model-driven products such as Meta platform's Llama 3.1.

“We’re especially excited to have Noam back, a leading researcher in the field of machine learning,” Google said in a statement. “This agreement will provide Character.AI with more funding to continue to grow and focus on building personalized AI products for users around the world.”

Google DeepMind Chief Scientist Jeff Dean welcomes old colleague back

In the past few years, Character.AI has rapidly risen to become one of the leaders in the field of AI chatbots with its innovative technology and personalized services. However, challenges such as high training costs and insufficient paying users have gradually led it into a commercial dilemma.

As early as July last year, Character.AI began discussing possible cooperation with giants such as Google, xAI and Meta, and has been in-depth discussions on potential acquisition plans in recent months. Earlier, The Information reported that it was facing financing difficulties and was likely to be acquired by xAI, which was publicly denied by Musk.

Now, with the final "flower falling into the Valley family", Character.AI has almost repeated the script of two other AI unicorn companies, Inflection and Adept. This is about to become a new trend created by this wave of large-scale model startups in Silicon Valley:

The founder sold himself and distributed the money to employees and shareholders, and everyone involved was happy. As for the company that once pinned the "AGI dream" under various names? Just leave a skeleton.

1

Character.AI's life starts high and ends low

Character.AI was co-founded by Noam Shazeer and Daniel De Freitas in 2021. Shazeer is an OG-level figure in the big model world. He was the chief software engineer at Google and participated in many important projects including spelling correction and advertising algorithm development. He was also one of the main authors of the groundbreaking paper "Attention Is All You Need" in 2017. While at Google, Daniel De Freitas co-developed a chatbot called Meena with Shazeer, which later evolved into the conversational language model LaMDA.

Image source: New York Times

With their rich experience in natural language processing and model development, the two led the team to launch the Character.AI platform in 2022. The core technology is AI chatbots that can simulate various characters. These robots can generate realistic conversations and provide highly personalized interactive experiences.

Character.AI allows anyone to freely create and customize AI characters, which greatly stimulates users' creativity and sense of participation, and quickly became popular in the community, attracting countless fans. In a short period of time, it raised a total of US$193 million in financing from a16z, SV Angel, former GitHub CEO Nat Friedman, and angel investor Elad Gil, with a valuation of US$1 billion, successfully joining the ranks of AI unicorns.

According to statistics, people have created millions of chatbots on the platform, and the psychological counseling character "Psychologist" has received 78 million messages in total. The most popular characters are from games and animations, such as General Raiden and Shadow in "Genshin Impact", which have received 371 million messages from users.

Unfortunately, despite significant user growth, Character.AI has not yet overcome the hurdle of monetization capabilities.

The training and reasoning of large language models requires a lot of computing resources, resulting in very high operating costs. Character.AI has always built products based on self-built models. Despite the launch of a $9.99 monthly subscription service, only 100,000 of its 6 million monthly active users in July were paying users. The estimated revenue for the whole of 2024 is only $16.7 million, which is obviously not enough to cover the high expenses.

In particular, Character.AI user profiles are concentrated in the young Z generation aged 18-24, with a narrow range and difficulty in breaking out of the circle. Although the stickiness is high, the economic foundation is inevitably relatively weak. In addition, when facing free heavy users, the token consumption is very large, which further aggravates the financial pressure.

In addition, since many of Character.AI's popular characters are based on game and anime IPs, it is easy to cause copyright disputes. The aforementioned "Genshin Impact" Raiden Shogun and Shadow, and "Naruto" Kakashi were forced to be removed from the shelves for this reason. The platform also occasionally omits content supervision of some emotional companion robots, and some developers have "jailbroken" the model, generating inappropriate content and causing user loss.

In short, in order to survive under all kinds of difficulties, Character.AI can only forget the glory of its debut, lower its profile, and find a good family for itself. Noam Shazeer and Daniel De Freitas, who chose to leave Google to start their own businesses because they were "dissatisfied with the company's bureaucracy", also returned to the embrace of Big Brother.

1

Silicon Valley is playing a very new "acquisition"

Google's operation this time is a typical "reverse acquisition". As an emerging M&A model, it has appeared many times in recent years when technology giants have deployed in the field of artificial intelligence.

Unlike traditional direct acquisitions, reverse acquisitions achieve control over a startup’s technology and talent by hiring its core team and signing technology licensing agreements, while avoiding antitrust reviews that may be triggered by direct acquisitions. For startups, it can also quickly recover funds and continue to advance the development process.

Before Google "implicitly bought" Character.AI, Microsoft had adopted a similar strategy, paying $650 million to obtain the technology license for Inflection AI. At the same time, the two co-founders Mustafa Suleyman and Karen Simonyan brought most of the company's technical backbones to join Microsoft's newly established AI department.

As for the AI ​​star product Pi, no one ever mentioned it again.

Another giant, Amazon, also paid a high fee for a non-exclusive license to use some of Adept's technologies to strengthen its capabilities in AI and automation, especially to enhance the competitiveness of its voice assistant Alexa. Adept co-founder and CEO David Luan will lead Amazon's AGI Autonomy team and report directly to Rohit Prasad, director of artificial intelligence. Several other founding members and some employees have also joined Amazon since then.

However, even if these transactions are not technically considered acquisitions, they have begun to attract the attention of antitrust regulators. For example, last month, the UK Competition and Markets Authority (CMA) issued a notice that they were investigating Microsoft's hiring of key personnel from Inflection AI to see if large technology companies are trying to circumvent regulatory scrutiny. The FTC also launched a similar investigation into Microsoft's $650 million transaction in June.

As for Google's acquisition of Character.AI, the vast majority of netizens' comments were: You really made a fortune!

Notion engineer Linus Lee believes that Google has brought back a strong player: "Noam Shazeer trained a 5 billion parameter language model six months before the release of GPT-2, predicting the future possibilities of MoE expert mixture models, model parallelism, and ultra-large-scale models and datasets. He has been working in the same field almost since the original Transformer model."

Ying Xiao, a Character machine learning researcher who worked with him, said, "Everyone knows that Noam Shazeer is a genius, but many people don't know how kind and honest he is. Today, Noam treats all employees very fairly and generously, and I will continue to work for him 100%."

Perplexity founder Aravind Srinivas also commented, “This is a major win for Google in the competition with OpenAI’s scale and infrastructure.”

However, as the next ideal candidate for "reverse acquisition" in the eyes of the public, Srinivas was immediately reminded:

“Make sure you sell to Satya Nadella on time. No need to struggle for too long. Then after the 4-year vesting period (make sure to have all the kids during this time), start a second startup. You will fail, but you will be in the news soon.”

1

“I get acquisition requests from ten startups every week”

In fact, in addition to Adept, Inflection, Character and other well-known AI unicorns that have successfully "landed" in large companies, there are many companies that are also struggling and seem to be ready to join this wave at any time.

Perplexity, the popular AI search engine, has been unable to shake Google's solid market share despite its impressive financing and valuation. Its management has reportedly discussed the possibility of a merger with at least four companies, but no deal has been reached yet. Writer, an enterprise-level AI writing platform, is also actively seeking acquisition opportunities and has approached at least three companies for negotiations, despite completing a $100 million Series B financing round last year. Stability AI is still in a state of anxiety, however.

On the other hand, Hugging Face CEO Clément Delangue once revealed that more and more AI startup founders have expressed their willingness to sell their companies to him, and Hugging Face receives acquisition requests from about ten startups every week.

In 2023, the field of generative AI experienced a significant investment boom, which was called the "generative AI bubble" by some industry insiders. Major breakthroughs in fields such as natural language processing and image generation attracted widespread market attention, huge amounts of capital were injected, and the valuations of AI startups rose rapidly.

However, despite the obvious technological progress, high computing costs, copyright privacy, content supervision and user experience still make it difficult for many companies to make sustained profits. In 2024, the market gradually calmed down and entered the "de-bubble" process, and the generative AI industry began to face reality. The commercialization dilemma made it difficult for many AI unicorns to survive independently, and small companies were washed out in batches.

Artificial intelligence expert Gary Marcus believes that the fact that three well-known AI company CEOs joined large companies and Microsoft took 15 years to recoup its investment is a sign that the generative AI bubble is about to burst.

At this time, cooperation with technology giants or acquisitions naturally become a more feasible way out. In addition to sufficient financial support, they will also obtain advanced resources and infrastructure, mature market channels and brand recognition, reducing market competition pressure. If you can't beat them, join the other side of the scale - holding a stable job is more secure after all.

However, this trend of capital market integration of AI startups may also hollow out a small entity, leading to limited innovation diversity.

For example, some Character users have begun to complain that they cannot log in normally and no one responds from the backend.

So did Character.AI fall down or stand up?

Were these AI unicorns suppressed or reborn?

Perhaps only time can give the answer.