With $200 million, Musk has "bet on" the man behind him.

As a titan in the tech world, every move of Elon Musk draws attention.

Recently, Igor Babuschkin, Musk's close partner and co-founder of XAI, announced his departure from XAI. However, their "close relationship" didn't end there. Musk promised to invest $200 million to become a limited partner (LP) of Babuschkin's newly founded venture capital firm, Babuschkin Ventures.

The firm will support research on AI safety and invest in startups that "drive human progress and uncover the mysteries of the universe." This allows Babuschkin, who was originally a business partner, to support Musk's exploration in the tech field in a new capacity as an "investor."

Babuschkin's Stunning Transformation

For Babuschkin, this transformation of identity is undoubtedly a brand - new beginning. During his time at xAI, Babuschkin achieved remarkable results.

Four months after founding xAI, Babuschkin led the team to build the "Memphis Supercluster" with 100,000 H100 GPUs, with a peak computing power of 2.6 exaFLOPS (FP16), and its training speed was 33% faster than that of GPT - 4 at the same stage.

In March 2024, xAI released Grok - 1 (with 314 billion parameters), and in November of the same year, it was iterated to Grok - 4, achieving a high score of 87.4 on the MMLU benchmark, second only to GPT - 4 Turbo's 87.6.

During the process, Babuschkin personally wrote the scheduling framework, reducing the interruption recovery time of thousand - card parallel training from an average of 30 minutes to 90 seconds, greatly improving the training efficiency. Later, he introduced heterogeneous hybrid training of TPU - v5 and GPU, successfully reducing the training cost per trillion tokens by 27%, giving xAI a cost advantage in the fierce competition and making him a "key figure" in xAI's development process.

These advancements made xAI's financing go smoothly. In May 2024, xAI completed a $6 billion Series B financing in one go, with a post - investment valuation of $24 billion. The list of investors was very impressive, including industry giants such as A16Z, Sequoia Capital, and Saudi PIF.

In November 2024, xAI completed a new $5 billion financing, doubling its valuation to $50 billion. One month later, xAI announced that a $6 billion Series C financing was reached, with major investors including A16Z, BlackRock, Fidelity Management & Research Company, Saudi Kingdom Holding, Lightspeed Venture Partners, Sequoia Capital, etc.

Then, under such rapid development, why did Babuschkin choose to leave? Actually, this is not accidental. In an open letter, Babuschkin said, "After helping xAI get on the track, I want to bet my next career on AI safety."

In June this year, when he had dinner with Max Tegmark, the founder of the "Future of Life Institute" (FLI) in Cambridge, they reviewed recent issues of Grok involving dangerous content such as the production of nerve agents and improvised nuclear bombs.

These problems reflected the absence of a "safety guardrail and governance framework," which became the trigger for Babuschkin's decision to leave and focus on the AI safety field.

Musk's $200 - million "Bet"

Although Babuschkin's departure is a "pity" for xAI, Musk's support for Babuschkin didn't stop.

With Musk's promise to invest $200 million to become an LP of Babuschkin Ventures, he continued his support in another way. In response, some commentators said that Musk's move is not just a personal support out of "old - colleague" sentiment, but a strategic "bet."

Looking at Babuschkin himself, his departure was not because of insufficient ability or a lack of prospects for xAI, but due to his sharp insight into the emerging and crucial field of AI safety. With the development of artificial intelligence technology, AI safety issues have become extremely urgent: from data privacy protection to algorithmic bias, from malicious use to potential ethical risks, the fields involved in AI safety are extensive and complex.

As a model company, xAI is crucial for enhancing user trust. Musk knows that as the amount of data processed by xAI continues to grow, data security and privacy protection have become key issues. In public, Musk has repeatedly emphasized the potential risks of artificial intelligence, including his criticism of Apple's cooperation with OpenAI.

After Apple integrated OpenAI's ChatGPT into its operating system, Musk strongly criticized it, calling it an "unacceptable security violation" and worrying that it would lead to the leakage of users' private data. In terms of freedom of speech and responsibility, Musk believes that AI needs to balance "excessive political correctness" and "unrestricted freedom" in speech, and proposed that an "accuracy calibrator" should be installed for AI.

In terms of safety, Musk also advocates that when developing AI, strict prevention and control at the ethical and logical levels must be followed to ensure that it has the greatest degree of truth - seeking and curiosity, and to avoid forcing it to lie or do things against axioms.

All these show Musk's attention and optimism towards the AI safety field and announce that he will make great efforts in this area.

In addition to Musk's $200 - million investment, Babuschkin also said that he will personally invest $50 million as the first - closing capital for the general partner (GP). In addition, the Teacher Retirement System of Texas and the Musk Family Office plan to invest $500 million and $200 million respectively. The target size of the fund is $1 billion, with a term of "10 + 2" years.

Booming Venture Capital in the AI Safety Field

The cooperation between Musk and Babuschkin, from xAI to Babuschkin Ventures, is regarded as the "romance" between two men.

During their time at xAI, Musk and Babuschkin established a deep friendship and trust. Musk once publicly said, "Without you (Babuschkin), there would be no xAI today," which fully reflects the mutual appreciation between the two.

For Babuschkin, Musk's trust is not only financial support but also spiritual encouragement, which is equivalent to a form of cooperation and inheritance.

In the AI safety field, startups are showing an active financing trend. According to incomplete statistics, several AI safety unicorns have emerged in this field.

Among them, Anthropic is the most prominent. Although this company is a newcomer in the AI field, its founders include former OpenAI employees and an Italian - American brother - sister pair. Headquartered in San Francisco, it focuses on developing safe artificial intelligence technologies.

Currently, this company is actively preparing for a financing round of up to $2 billion, led by Lightspeed Venture Partners. After this round of financing is completed, Anthropic's valuation will soar to $60 billion, more than three times its valuation of $18 billion a year ago, making it one of the top five most - valued startups in the United States.

Moreover, the AI defense technology company ShieldAI completed a $240 - million Series F financing in March this year. This round of financing was co - led by L3Harris Technologies and Hanwha Aerospace, with follow - on investments from firms such as Andreessen Horowitz and the US Innovation Technology Fund, and its valuation reached $5.3 billion. The AI platform company Cyera, which focuses on data security, completed a $300 - million Series C financing in 2024, with a valuation of $1.4 billion. Chainguard, which provides software supply - chain security solutions, completed a $140 - million Series C financing in 2024, with a valuation of $1.12 billion. And the cybersecurity company Huntress also completed an $180 - million Series D financing in 2024, with a valuation of $1.55 billion...

These financings are sufficient to illustrate the importance and popularity of AI safety in the industry's development. Babuschkin, now an "investor," will also explore this field with Musk's trust. Maybe, in the next year, a new AI safety unicorn will be born from Babuschkin Ventures' fund pool.

This article is from the WeChat official account "Dongshisi Tiao Capital" (ID: DsstCapital), author: Chen Mei, published by 36Kr with authorization.