The startup, one of the very few AI unicorns led by women, carries a valuation of $1 billion and possesses access to 10,000 Nvidia H100 GPUs, but its creators indicate that it may take years before a product is unveiled.
by Alex Konrad, Forbes Staff and Kenrick Cai, Forbes Staff
It was at a gathering in San Francisco hosted by Greg Brockman nearly a decade ago — around the period when the Stripe CTO departed to co-establish an AI Agents research institution known as OpenAI — that entrepreneurs Kanjun Qiu and Josh Albrecht encountered a cryptocurrency tycoon named Jed McCaleb.
As Qiu and Albrecht initiated two now-discontinued startups, and McCaleb became a frequent visitor at The Archive, their shared residence for founders and AI researchers near Dolores Park, they occasionally discussed developing their own AI lab. Several of their housemates had become involved in Brockman and Sam Altman’s OpenAI initiative. Nonetheless, it seemed premature, “a little too wild,” to abruptly abandon their existing projects, according to Qiu. However, when their second startup, Sourceress, experienced a slowdown in growth, they reevaluated their situation. And when the pair secured $20 million for a new AI research enterprise, Generally Intelligent, last year, McCaleb provided the largest funding amount.
Now, Qiu and Albrecht are redoubling their efforts under the new label, Imbue — and this time, McCaleb, a billionaire following his co-founding of the crypto firm Ripple, is making an even larger investment. Imbue has confirmed that the Astera Institute, his nonprofit organization devoted to supporting science and technology initiatives, is spearheading a $200 million Series B investment round in the startup. This all-cash round, which features AI chip manufacturer Nvidia, Cruise cofounder Kyle Vogt, and Notion cofounder Simon Last, values Imbue at over $1 billion, positioning the company among the exclusive group of women-led AI research unicorns.
While OpenAI and competitors such as Anthropic and Google strive to construct substantial AI foundational models like GPT-4, Qiu and Albrecht are navigating a distinct trajectory. Imbue’s emphasis is on an AI agent: a type of computing system capable of emulating human decision-making to accomplish intricate tasks. Chatbots like ChatGPT receive a user’s inquiry and produce a nearly instantaneous reply. Imbue’s agents would function more like a virtual research assistant that can analyze data, suggest follow-up experiments, and even organize them autonomously.
This level of autonomy could be beneficial in various scenarios, from biological research to travel arrangements and complex coding tasks, as noted by Albrecht. Agents “independently venture off and perform tasks,” Qiu elaborated.
“We believe AI has the capability to lessen the gap between ideas and execution.”
Imbue cofounder Kanjun Qiu
To create such agents, Imbue has secured access to 10,000 of Nvidia’s H100 GPUs — roughly the same quantity of processors that OpenAI utilized to develop GPT-3. It launched an open-source training environment for educating these tools, named Avalon, last fall; numerous prototypes and updates are anticipated in the coming months, according to its founders.
Imbue is also astonishingly early in its quest to join the ranks of billion-dollar startups. The company currently employs approximately 20 individuals. There is no demonstration of their AI agents available for public viewing yet, their founders mentioned. Additionally, their primary investor, McCaleb’s Astera, stands out as an unconventional source compared to the renowned venture capital firms and Big Tech providers that have flocked to other recent AI ventures — partly due to Qiu and Albrecht’s assertion that a nonprofit can afford to be more tolerant with their commercialization schedule. (McCaleb informs Forbes that he is still aspiring for “venture-style” financial returns in the long run.)
In the exuberant AI landscape, a startup acquiring hundreds of millions without any revenue is not unusual. Nevertheless, it represents a high-stakes gamble that places significantly more scrutiny on Imbue’s future activities. For Qiu, that ambition is ambitious: to establish AI agents in the same pioneering way that the Xerox PARC lab did for personal computers half a century ago.
At that time, computers were costly and complicated to operate, but the PARC researchers crafted tools that made them accessible to non-technical users, ultimately leading to PCs becoming commonplace in homes. Imbue aims to achieve the same for generative AI. “We believe AI Agents has the potential to narrow the divide between ideas and execution,” Qiu expressed. “A truly personal computer that performs tasks for you and liberates individuals.”
Imbue’s creators first crossed paths at a conference at UC Berkeley in 2014. Qiu had a position at Dropbox, where she began in business operations before becoming the inaugural chief of staff to CEO Drew Houston. Albrecht had co-founded and served as chief technology officer for several startups prior to joining the data team at wealth management firm Addepar. Both studied machine learning during their college years, with Albrecht co-authoring papers at the University of Pittsburgh, and Qiu focusing on probability theory while developing and executing high-frequency trading algorithms to help finance her education at MIT.
The pair bonded over sophisticated discussions surrounding the future of human decision-making capabilities in an AI-driven society. Eventually, Albrecht persuaded Qiu to collaborate on Ember Hardware, his startup project aimed at creating virtual reality hardware, which, however, never materialized. Following the establishment of The Archive house, which would eventually become a residence for many early OpenAI researchers, they resolved to apply machine learning to tackle a challenge they had faced at the now-defunct Ember: encouraging talented individuals who were not actively job hunting to apply nonetheless. Their startup, named Sourceress, went through the Y Combinator accelerator in 2017; Houston and his Dropbox cofounder Arash Ferdowsi invested as Sourceress secured a $3.5 million seed round upon completing the program.
“Every decade or so, a new era of computing begins, and everyone is exploring together.”
Dropbox cofounder and CEO Drew Houston
The company performed reasonably well for a period, accumulating millions in revenue, before growth stagnated. After several attempts to revive the momentum failed, the founders and Threshold investor Josh Stein engaged in a “mature conversation” regarding shifting to another, potentially more significant concept. “There was nothing inherently wrong with Sourceress, but it became evident that it wouldn’t evolve into a breakout success,” Stein remarked.
In late 2020, Qiu and Albrecht terminated Sourceress, returning approximately $4 million to investors and inviting those interested to transition some of their investments into equity in the duo’s new venture, Generally Intelligent. On paper, those equity stakes appear promising given Imbue’s recent valuation: “It appears that this will become a very lucrative investment for us,” Stein commented.
It’s a pleasant August evening in San Francisco as Houston takes the stage to discuss all things AI with Qiu and Ali Rohde, who oversees a small venture fund that Imbue’s founders created to support other early-stage AI startups. Several dozen attendees, primarily entrepreneurs and engineers, gather at productivity software firm Notion’s sleek Mission District office to listen to Houston’s talk.
The event, part of a weekly series, resembles the former dinners at The Archive, albeit on a larger scale. “Every decade or so, a new era of computing kicks off, and everyone is fumbling around together,” Houston mentioned later. “Being part of that community is exhilarating.”
Houston didn’t anticipate his former employee transitioning from recruitment automation to an AI research lab; now, in light of the recent enthusiasm surrounding generative AI Agents, he expressed that the transition has shifted from appearing as a “science project” to being “right in the fairway.” He is joined by a prominent group of supporters from that fund, Outset Capital, which includes Quora.
CEO and OpenAI board member Adam D’Angelo, Hollywood tycoon Jeffrey Katzenberg, and Google DeepMind’s policy chief (and former Dropbox PR head) Dorothy Chou.
“We could either continue researching while all these other laboratories were making significant advancements, or we could take a risk here, and pursue it.”
Imbue investor Jed McCaleb
Despite their extensive Silicon Valley connections, numerous prominent investors within the field – all of whom requested anonymity to speak candidly – expressed skepticism about the team’s qualifications to manage a credible AI research laboratory. Others hold a different view. One individual, familiar with Imbue’s founders, dismissed such apprehensions as venture capitalist bias towards accumulating “baseball cards,” indicating a tendency to only support founders from a specific elite background. Nevertheless, they questioned whether Imbue’s goal to launch agents as marketable projects would set it apart from other well-funded labs in the long term.
Qiu and Albrecht remain unfazed by any doubts regarding their team’s research skills. They highlighted several employees possessing academic qualifications in AI research, along with backgrounds in neuroscience and plasma physics, asserting that their diversity is indeed a strength.
Imbue’s absence of conventional investors on its capitalization table (with Stein’s Threshold being the only exception) has also sparked inquiries. The founders assert they intentionally avoided any official meetings with venture capital firms during Imbue’s recent fundraising endeavors, largely due to their acknowledgment that their work may require years to yield verified commercial projects. “I believe it’s actually advantageous to be somewhat of an outsider, to adopt a fresh perspective,” their investor, McCaleb, concurred.
Regarding Imbue’s staggering valuation for such an emerging endeavor, McCaleb expressed satisfaction with the extent of the potential after witnessing a demonstration that illustrated how Qiu and Albrecht might ultimately develop agents; it contrasted with “99% of the [research] initiatives” he observed at other AI laboratories. (Qiu and Albrecht indicated that they could not publicly disclose that demonstration yet).
“To genuinely advance this research to the next stage and see if we can construct it and then subsequently commercialize it, substantial funding is needed, as GPUs are essential, right?” McCaleb remarked. “We could either persist with our research while all these other labs proceeded with substantial progress, or we could take a gamble here and move forward. Thus, it just felt like the right moment to do so.”
Imbue may one day experience its own public-facing epiphany, akin to OpenAI’s successful debut of ChatGPT. Such developments require time. Qiu and Albrecht referenced Cruise, the self-driving car enterprise whose cofounder is now an investor, and how early enthusiasm surrounding demonstrations is currently leading to broader acceptance, a decade later. Mainstreaming agents like Imbue likely won’t span a decade, Albrecht asserted; however, it won’t occur within just a few months either.
“We will be actively investigating,” Albrecht stated. “We want to hold off until we are confident, believing that this is truly exceptional, we trust it, and it’s robust, safe, and outstanding, before we present it to the public.”
MORE FROM FORBES
2 Trackbacks / Pingbacks