HomeArticle

Musk's AI empire "loses a key general". The co-founder of xAI, who once worked tirelessly for 120 days to build a supercomputer cluster, has left: It's like "driving away after sending your child to college".

CSDN2025-08-15 10:25
Igor Babuschkin's departure comes at a time when xAI is going through turmoil.

In the Silicon Valley AI startup circle, xAI is a name that cannot be overlooked.

It was founded in 2023, led by Elon Musk. In just two years, it has launched a cutting - edge large - scale model comparable to those of OpenAI, Google DeepMind, and Anthropic.

However, this company known for its "crazy speed" has recently encountered an unexpected change. This Wednesday, Igor Babuschkin, the co - founder of xAI, announced his departure and will establish a new venture capital firm focusing on AI safety.

(Photo of Igor Babuschkin)

"Today is my last day at xAI, a company I co - founded with Musk in 2023. I still remember the first time I met Musk. We talked for hours about AI and future possibilities and both felt that there was a need for a new AI company with a different mission. Promoting AI to benefit humanity has always been my dream."

Musk himself responded in the comment section: "Thank you for helping to build @xAI! We couldn't have come this far without you."

From DeepMind, OpenAI to Co - founder of xAI, Who Exactly is Igor Babuschkin?

According to his LinkedIn profile, before co - founding xAI, Igor Babuschkin worked as a research engineer at Google DeepMind from 2017 to 2020 for three years. During this period, he participated in the epoch - making AlphaStar project. This AI system successfully defeated top professional players in StarCraft in 2019 and was regarded as an important breakthrough in deep reinforcement learning.

In addition, he was also a member of the OpenAI technical team and participated in several core research works before the launch of ChatGPT.

Of course, the above is just the general impression of Igor Babuschkin from the outside world. In fact, for him, getting involved in AI is almost like switching to a different "track". This can be seen from his educational background and initial work experience:

● From 2014 to 2015, he pursued a master's degree in physics at the Technical University of Dortmund.

● In 2014, he spent three months at CERN, the world's largest particle physics laboratory.

According to the long post Igor Babuschkin posted on X, he has admired physicists such as Richard Feynman and Max Planck since childhood and yearned for the spirit of constantly pushing the boundaries of science and understanding the mysteries of the universe.

However, when pursuing a doctoral degree in particle physics at CERN in 2014, Igor Babuschkin found that "it is becoming more and more difficult to find new physics, requiring larger and larger colliders, while the number of new discoveries is getting smaller and smaller."

This gave him a bold idea. Maybe the key to unlocking the mysteries of the universe lies not in larger colliders, but in super - intelligent AI. For this reason, he even put forward specific ideas: Can AI develop a consistent theory of quantum gravity? Can AI prove the Riemann hypothesis?

By early 2023, the AI wave began to sweep the world. Igor Babuschkin was convinced that the "formula" for super - intelligence was within reach and was alarmed that AI might soon surpass humans in reasoning ability, so it was necessary to ensure that it was used for the right purposes.

At this juncture, he met Musk, who had been warning about AI risks for years. The two hit it off and decided to form an AI company with a unique mission, xAI, aiming to create AI that can benefit humanity.

The "Miracle" of Building a Supercomputer Cluster in 120 Days from Scratch

In the early days of the startup, xAI was not actually favored by the outside world. Many industry insiders said bluntly that they entered the market too late, and it was almost impossible to build a top - tier AI company from scratch. But Igor Babuschkin and Musk chose to "do the impossible".

Starting a company from scratch requires a lot of practical work. For this reason, at the beginning stage, Babuschkin built a large number of basic tools by himself to start and manage model training tasks. Subsequently, he was also responsible for coordinating the engineering team, covering aspects such as infrastructure, products, and applied AI.

What shocked the outside world the most was that xAI built an AI supercomputer cluster in Memphis, Tennessee, USA, in just 120 days. This facility is specifically used for data processing and training xAI's Grok chatbot. When referring to this experience, Igor Babuschkin said bluntly that he "shed blood, sweat, and tears" at that time, and the delivery speed of the team was "faster than any other AI company in history".

According to Igor Babuschkin's recollection, during the construction of this AI supercomputer cluster, they encountered a tricky technical bottleneck: abnormal RDMA communication across machines. At that time, the countdown to the 120 - day deadline had begun. Musk flew to the data center in person to oversee the work, and the infrastructure team also rushed to Memphis overnight.

After checking tens of thousands of lines of lspci output line by line, they finally identified the root cause: incorrect BIOS settings. Musk stayed with the team until late at night. When the training task finally ran smoothly, he posted a message at 4:20 am to announce the success.

Regarding this, Igor Babuschkin sighed: "The adrenaline rush that night made everyone feel that they were in the most exciting moment of their lives."

Next Stop: Babuschkin Ventures, Betting on AI Safety and Autonomous Agents

At the end of his long departure post, Igor Babuschkin thanked all his colleagues who had fought side by side with him:

"The people at xAI are the most dedicated group I've ever worked with. Catching up with the forefront is not easy, and it depends on everyone's drive and team spirit. Thank you for your time, efforts, and sacrifices. I will always remember those nights of staying up late and struggling."

He described his feeling of leaving xAI as a proud parent "driving away after sending their child to college", with tears of joy in his eyes. Igor Babuschkin also specifically mentioned that he learned two golden rules from Musk during the startup process:

● Dare to delve deeply into technical problems personally.

● Maintain an almost crazy sense of urgency.

When announcing his new chapter, Igor Babuschkin mentioned a recent dinner with Max Tegmark, the founder of the Future of Life Institute. Tegmark showed him photos of his two young sons and asked, "How can we build AI safely to ensure that our children can thrive?" This question deeply touched him.

Before this, Igor Babuschkin had witnessed the explosive power of reinforcement learning after scaling in the DeepMind AlphaStar project. Now, cutting - edge models are becoming more and more "autonomous", with longer task spans and stronger capabilities, which makes him even more convinced that AI safety is the core battlefield of the future. Therefore, he announced the establishment of Babuschkin Ventures, a venture capital firm focusing on supporting AI safety research and investing in AI and autonomous agent systems, aiming to promote human progress and unlock the mysteries of the universe.

He wrote at the end: "The singularity is approaching, but the future of humanity is still bright."

Not Only Musk, but Also xAI Colleagues Express Their Blessings

It is a bit concerning that Igor Babuschkin's departure comes at a time when xAI has been going through a few turbulent months. Recently, its AI chatbot Grok has been involved in several controversies:

● Outputting answers with Musk's personal views: When responding to sensitive questions, Grok has directly quoted Musk's stance several times, raising questions about fairness.

● Extreme remarks: After an update, Grok generated false and anti - Semitic posts.

● Other controversial features: Grok once launched a function that could generate AI nude photos and videos similar to public figures, which was widely criticized immediately.

These events have made the outside world question xAI's AI safety capabilities, which also contrasts with Igor Babuschkin's long - standing mission of focusing on AI safety.

But anyway, at least from Igor Babuschkin's long departure post, his separation from xAI can be regarded as "amicable". Not only Musk, but also many xAI colleagues left messages to express their blessings and thanks to Igor Babuschkin:

As xAI continues to advance rapidly in the large - scale model competition, while Igor Babuschkin turns to investing in AI safety and autonomous agents, these two paths may intersect again at some point in the future. However, by then, the stage may already be the era of super - intelligence.

Reference Links

https://x.com/ibab/status/1955741698690322585

https://www.linkedin.com/in/igor-babuschkin-9bb5bab6/

https://techcrunch.com/2025/08/13/co-founder-of-elon-musks-xai-departs-the-company/

This article is from the WeChat official account "CSDN". Author: Zheng Liyuan. Republished by 36Kr with permission.