The artificial intelligence (AI) sector is on the cusp of a significant transformation, with a wave of new startups emerging from the fragmentation of established AI labs. This trend has been highlighted in the latest State of AI report by Air Street Capital, which sheds light on the internal dynamics prompting this shift.
Large-scale AI labs are dissolving into smaller entities due to a myriad of reasons—ranging from interpersonal disagreements and philosophical divergences to mounting commercial pressures. Such fragmentations are becoming a breeding ground for new ventures, encouraging researchers to branch out and establish their own startups. Consequently, these nascent companies are attracting substantial investment from venture capitalists, eager to back projects with high potential in the AI domain.
One of the most notable figures to embark on such a venture is Ilya Sutskever, formerly the chief scientist at OpenAI. In September 2024, Sutskever managed to raise a staggering $1 billion for his startup, Safe Superintelligence, focused on AI safety. The project drew investments from prominent venture capital firms such as Andreessen Horowitz and Sequoia, even before it officially launched a product. This reflects the robust investor interest in pioneering AI solutions and underscores the appeal these new enterprises hold for financiers.
Similarly, other startups have followed suit. Black Forest Labs, created by former executives of Stable Diffusion, is allegedly in negotiations to raise $100 million, aiming for a valuation of $1 billion. Paris-based enterprise H, initiated by ex-DeepMind employees, closed a significant pre-seed funding round of $220 million in May, backed by Accel and UiPath. Meanwhile, SakanaAI, led by StabilityAI’s former chief operating officer, garnered $200 million from New Enterprise Associates (NEA). Another noteworthy example is Anthropic, a lab founded in 2021 by ex-OpenAI executives, marking the beginning of this fragmentation trend.
PitchBook data reveals that startups focused on building foundational AI models have raised a record $22.9 billion in venture capital funding in 2024, surpassing the $18.4 billion raised the previous year. This surge underscores investor confidence in the potential of these newly created entities.
The trend is partly driven by a blend of scientific conflicts, commercial demands, personality clashes, and the ready availability of capital, according to Air Street Capital's report. These factors signal an evolution within the AI ecosystem, reflecting a broadening and diversification of AI innovation and application.
Personality conflicts appear to be a significant catalyst in the division of these labs. For instance, three cofounders of H left the startup shortly after its inception due to operational differences, prior to the launch of any products. A publicized rift between OpenAI's Sutskever and his cofounder Sam Altman exemplifies how personal value differences and egos can lead to departures.
The influence of venture capital in moulding these dynamics cannot be understated. Noel Hurley, an ex-vice president at Arm and currently CEO of the AI startup Literal Labs, notes that venture capitalists often inflate the prominence of these so-called 'superstars' in the industry, further encouraging them to venture independently.
Moreover, a key challenge within larger AI organisations is the struggle with agility due to bureaucratic limitations. Former DeepMind and Meta employees, for instance, faced similar issues, prompting them to establish independent ventures like H and Mistral. According to Samir Kumar of Touring Capital, researchers often find more freedom and control in directing project priorities when operating independently.
Despite existing organisations holding structural advantages, Kumar predicts a rise in companies at the forefront of generative AI shifts. The landscape remains uncertain, as initial success does not guarantee long-term dominance in the field.
Peter J. Liu, previously with DeepMind, together with former Google Brain researchers, announced his departure to explore new initiatives. Liu noted that, although in-house research is crucial, smaller, focused teams can be more productive.
However, a potential re-consolidation of these spin-outs into a few major entities could eventually occur, cautioned Noel Hurley. As operating AI labs can be prohibitively expensive, collaboration or mergers might be inevitable unless labs specialise in niche markets.
As new AI labs emerge and mature, they have the potential to achieve significant revenue, potentially transforming sceptics into believers within the investment community. This ongoing churn within the AI sector presents a dynamic market environment where public companies may increasingly seek exposure to these burgeoning labs to stay competitive.
Source: Noah Wire Services