Holding 10,000 H100 cards, this AI company has raised $200 million in funding with a valuation of $1 billion, and NVIDIA has invested again!

A new unicorn has emerged in the field of AI.

According to media reports on Thursday, Imbue, an AI startup and research laboratory, has secured $200 million in Series B funding, valuing the company at $1 billion and making it one of the few unicorns led by women.

The company also holds 10,000 NVIDIA H100 chips, but its founder stated that it may take several years to launch a product.

This round of funding was led by Astera Institute, a non-profit organization founded by cryptocurrency billionaire Jed McCaleb, with participation from NVIDIA, as well as Kyle Vogt, CEO of Cruise, an autonomous driving company under General Motors, and Simon Last, co-founder of Notion. The funding amount is ten times the previous funds raised by Imbue.

Imbue primarily trains foundational models optimized for reasoning abilities, which the company first applies to the development of encodable AI agents. The company aims to enable anyone to build customized AI agents.

This round of funding brings Imbue's total funding to $220 million, making it one of the best-funded AI startups in recent months, second only to AI21 Labs ($283 million), Cohere ($435 million), and Adept ($415 million), among other generative AI unicorns.

In the context of the ongoing AI boom, it is not unheard of for a startup to raise hundreds of millions of dollars without any revenue. However, this is a high-risk bet that has drawn more attention to Imbue's next steps.

At the same time, there are some doubts among investors about Imbue, especially in light of recent discussions about "NVIDIA's deception" [link in Chinese]. The outside world is particularly concerned about whether Imbue's ambition to release AI agent models as commercial projects can set it apart from other well-resourced laboratories in the long run.

Targeting AI agents, focusing on AI reasoning capabilities

Unlike companies like OpenAI that are striving to build large-scale AI foundational models like GPT-4, Imbue has chosen the path of AI agents: computational systems that can simulate human decision-making to complete complex tasks.

In a blog post, Imbue wrote:

"This latest funding will accelerate our development of AI systems capable of reasoning and encoding, so that they can help us achieve greater goals."

"Our goal has always been to build practical AI agents that can help us achieve greater goals and work safely in the real world." Imbue co-founder and CEO Kanjun Qiu believes that optimizing the reasoning capability of AI systems is crucial for building efficient AI agents.

In an interview with the media, she said:

"By optimizing the system for reasoning, we can have AI that truly helps us do what we want to do, rather than just having a conversation."

Imbue's "mega-scale" model, trained with over 100 billion parameters, has required the accumulation of 10,000 NVIDIA H100 GPUs, roughly the same number of processors used by OpenAI to train GPT-3.

Last autumn, the company released an open-source training environment called Avalon for training these tools. The founder stated that there will be more prototypes and products released in the coming months.

Currently, Imbue does not intend to put most of its in-development products into production. Instead, it views these tools and models as a way to improve future, more general-purpose AI and lays the foundation for a platform that people can use to create their own custom models.

Imbue co-founders Kanjun Qiu and Josh Albrecht

NVIDIA, another investor, leads to skepticism as a non-profit organization

There are some doubts about Imbue's recent financing round.

Firstly, NVIDIA is once again one of the investors.

This year, NVIDIA has invested in seven AI unicorn companies, including Imbue. The other six are: Adept, Coreweave, Cohere, Inflection, Runway, and AI21 Labs.

Recently, there have been conspiracy theories on social media questioning NVIDIA's involvement in Coreweave, an AI startup that they invested in, assisting in manipulating Q2 revenue. There are many "questionable points" in the financial report that suggest collusion among major companies to boost stock prices.

Some skeptics even believe that Coreweave, which placed orders worth billions of dollars with NVIDIA, may not even be a real company, but rather a shell organization created and supported by BlackRock, NVIDIA's largest shareholder, and the asset management giant. Of course, these claims have been refuted by professional analysts on Wall Street, who believe that it proves the intense struggle between the bulls and bears in the field of NVIDIA. The underlying logic behind this concern is the fear that once the AI bubble bursts, it will drag down the entire market.

Furthermore, despite its valuation as a unicorn, Imbue's development is still in a surprisingly early stage. The startup company has only about 20 employees. Several AI investors in Silicon Valley doubt whether the team is capable of running a serious artificial intelligence research laboratory.

Someone familiar with the founders of Imbue believes that this concern is not significant because venture capitalists tend to support founders with famous backgrounds.

Qiu and Albrecht have no concerns about the research capabilities of their team. They point out a few employees with academic backgrounds in artificial intelligence research, neuroscience, and plasma physics, considering their breadth as an advantage.

Finally, the founders stated that their product is not yet ready for the public. The main investor in this round of financing for the company is the non-profit organization Astera Institute, which is dedicated to supporting technology projects. This is different from the well-known venture capital firms and large technology cloud providers that have recently poured into other AI projects.

The two founders said that during Imbue's recent financing process, they deliberately did not have any formal meetings with venture capital firms. This is largely because they acknowledge that their work may take several years to develop mature commercial projects, and non-profit organizations can be more patient with their commercialization timeline.

Their largest investor, cryptocurrency billionaire Jed McCaleb, agrees with this:

"I think it's beneficial to look at this issue with a fresh perspective as an outsider."

As for a nascent project like Imbue having a staggering valuation, McCaleb said that after seeing Qiu and Albrecht's demo PPT on how to build AI agents, he was satisfied. However, this is different from how he views other artificial intelligence laboratories, where he wants to see "99% of the (research) effort."

McCaleb said:

"To really push this research into the next stage, to see if we can build it and then productize it, you need a lot of money, because you need GPUs, right?"

"Either we continue researching, and all the other labs make great progress, or we can make a bet here and go do it. So, it seems like it's time to do that now."

Perhaps one day, just like when OpenAI launched the popular ChatGPT, Imbue will also have its shining moment in the public eye. These things take time.

Qiu and Albrecht point out that ten years from now, the early excitement over PPTs will translate into more applications. Albrecht claims that mainstream AI agent providers like Imbue may not take ten years; Of course, it is also impossible to spend only a few months.

Albrecht said:

"We will actively explore, we want to wait until we are ready, and we think it is really good, we believe in it, it is powerful, secure, and great, and then we will launch it."