AI Labs Face Profitability Test Amid Industry Pressure

This is a distinct era for AI companies developing their own foundation models.

On one hand, a cohort of seasoned industry veterans, having built their reputations at major tech firms, are now striking out on their own. Simultaneously, legendary researchers with deep expertise but unclear commercial goals are entering the fray. While there is a clear possibility that some of these new labs will grow to rival giants like OpenAI, there is also ample space for them to focus on fascinating research without the immediate pressure to commercialize.

The outcome? It's becoming increasingly difficult to discern who is genuinely focused on building a profitable business.

To clarify this landscape, I propose a sliding scale for any company crafting a foundation model. This five-tier framework isn't about whether a company is currently profitable, but whether it is actively trying to be. The goal is to gauge ambition, not necessarily success.

Consider the scale as follows:

• Level 5: We are already generating millions in daily revenue.
• Level 4: We possess a detailed, multi-phase strategy to achieve market dominance and vast scale.
• Level 3: We have numerous promising product concepts that will be unveiled when the time is right.
• Level 2: We have the preliminary sketch of a potential plan.
• Level 1: Our true value lies in the pursuit of knowledge and self-defined purpose.

The established players—OpenAI, Anthropic, Gemini, and others—firmly occupy Level 5. The scale becomes more revealing when applied to the newest generation of labs, which often have grand visions but more ambiguous commercial trajectories.

Importantly, the founders of these labs often have the freedom to choose their level. With so much capital flowing into AI, few investors are demanding rigid business plans upfront. Even a purely research-oriented lab can attract eager backers. If your primary drive isn't amassing a fortune, you might find greater satisfaction operating at Level 2 than grinding at Level 5.

Disrupt 2026 Tickets: Limited-Time Opportunity

Tickets are now available! Secure savings of up to $680 at our current rates. Be among the first 500 registrants to receive a 50% discount on a guest pass. TechCrunch Disrupt unites top executives from Google Cloud, Netflix, Microsoft, Box, a16z, Hugging Face, and more for over 250 sessions aimed at accelerating growth and refining your competitive advantage. Network with hundreds of pioneering startups and participate in curated sessions designed to spark deals, insights, and inspiration.

Disrupt 2026 Tickets: Limited-Time Opportunity

Tickets are now available! Secure savings of up to $680 at our current rates. Be among the first 500 registrants to receive a 50% discount on a guest pass. TechCrunch Disrupt unites top executives from Google Cloud, Netflix, Microsoft, Box, a16z, Hugging Face, and more for over 250 sessions aimed at accelerating growth and refining your competitive advantage. Network with hundreds of pioneering startups and participate in curated sessions designed to spark deals, insights, and inspiration.

Complications arise because an AI lab's position on this scale is often unclear—and much of the industry's current intrigue stems from this uncertainty. The anxiety surrounding OpenAI's transition from a non-profit stemmed from its leap from Level 1 to Level 5 almost overnight. Conversely, one could argue that Meta's early AI research was solidly at Level 2, while the corporate ambition was always Level 4.

With this framework in mind, here is an assessment of four prominent contemporary AI labs and where they might fall on the scale.

Humans&

Humans& was a major AI story this week and part of the inspiration for creating this scale. Its founders present a compelling vision for the next wave of AI, shifting focus from pure scaling laws to tools that enhance communication and coordination.

Despite positive coverage, Humans& has been vague about how this vision translates into market-ready products. The team suggests they intend to build products but avoids committing to specifics. Their stated goal is to create AI-powered workplace tools intended to replace platforms like Slack, Jira, and Google Docs, fundamentally reimagining how such software functions—essentially, workplace software for a future beyond traditional software.

Deciphering tech visions is part of my job, and I must admit the last part leaves me somewhat puzzled. However, the description is just specific enough to place them tentatively at Level 3.

Thinking Machines Lab

This one is challenging to rate. Typically, when a former ChatGPT CTO and project lead secures a $2 billion seed round, you assume a concrete roadmap exists. Mira Murati does not seem like someone who proceeds without a plan, so entering 2026, I would have confidently placed TML at Level 4.

But then the last two weeks unfolded. The exit of CTO and co-founder Barret Zoph made headlines, partly due to unusual circumstances. At least five other employees departed with Zoph, many expressing concerns about the company's direction. Just one year in, nearly half of TML's founding executive team has left. One interpretation is that they believed they had a solid plan to become a top-tier AI lab, only to discover it was less firm than expected. On our scale, they aimed for Level 4 but found themselves at Level 2 or 3.

While there isn't conclusive evidence for a formal downgrade yet, the situation is approaching that point.

World Labs

Fei-Fei Li is one of the most esteemed figures in AI research, renowned for creating the ImageNet challenge that catalyzed modern deep learning. She holds a Sequoia-endowed chair at Stanford, co-directing two AI labs. The list of her honors and positions is extensive, but suffice to say, she could easily spend her career accepting awards and accolades. Her book is also excellent!

So, in 2024, when Li announced raising $230 million for a spatial AI venture named World Labs, it might have seemed like a Level 2 or lower operation.

But that was over a year ago—a lifetime in AI. Since then, World Labs has launched both a comprehensive world-generation model and a commercial product built upon it. Concurrently, we've seen genuine demand for world-modeling technology from video game and special effects industries, with no major labs offering direct competition. The result increasingly resembles a Level 4 company, potentially on the cusp of reaching Level 5.

Safe Superintelligence (SSI)

Founded by former OpenAI chief scientist Ilya Sutskever, Safe Superintelligence (SSI) appears to be a quintessential Level 1 endeavor. Sutskever has worked diligently to shield SSI from commercial pressures, even declining an acquisition offer from Meta. There are no product cycles, and beyond the core research into a superintelligent foundation model, there seems to be no immediate product. With this pure-research pitch, he raised $3 billion. Sutskever has consistently prioritized the science of AI over business, and all signs indicate this project is fundamentally a scientific pursuit.

That said, the AI field evolves rapidly, and it would be unwise to completely exclude SSI from future commercial considerations. In a recent interview, Sutskever suggested two scenarios that could trigger a pivot: if research timelines prove unexpectedly long, or if the value of deploying the world's most powerful AI to impact society becomes overwhelming. In other words, whether the research goes exceptionally well or faces significant hurdles, we might see SSI ascend several levels quickly.