Paris-based startup H raised $220M to build AI agents in the race towards AGI

H, the French formerly known as Holistic AI, was founded some months ago by Charles Kantor, a researcher at Stanford, and four DeepMind alum: Karl Tuyls, former research director, Laurent Sifre, ex-principal scientist, Daan Wierstra, who was a DeepMind funding member and will become chief scientist at H, and Julien Perolat, a researcher who worked on game theory and multi-agent research. In addition to this all-star funding team, H's mission to develop frontier action models as part of the path toward artificial general intelligence (AGI) seems to have caught everyone's attention, as the startup has managed to put together a 25-engineer team, and raise a $220 million seed funding round.

H is focused on developing multi-agent models, a structure where several AI systems interact with each other to perform tasks requiring memory, planning, and a certain degree of decision-making. It is now a well-known fact that current AI systems struggle with those same capabilities, and most researchers consider them essential for an AGI that can perform at the same rate as a human or better at any task. Although the startup may be overreaching with its promises about AGI, AI agent systems have established themselves as the new frontier. Thus, it makes sense that H would focus its efforts there.

Investors in H's seed funding round form an outstanding list of billionaires (and their offices), VC funds, and industrial investors. Notable billionaires include Eric Schmidt, Xavier Niel, Yuri Milner, Bernard Arnault (via Aglaé Ventures), and Motier Ventures. VC backers include the round leader Accel, Bpifrance’s Large Venture fund, Creandum, Elaia Partners, Eurazeo, FirstMark Capital, and Visionaries Club. Finally, Amazon, Samsung, and UiPath are among the select industrial partners.

In addition to covering its operations, a substantial part of the funding will be invested in buying computing power. According to a report by Bloomberg, the company plans to train a model as quickly as possible to start going after several markets and revenue streams.