AMI Labs CEO Skips 'AGI' Hype, Eyes Robotics Partners
17 Jul 2026
A world model startup that won't say 'AGI'
Alexandre LeBrun, CEO of world model startup AMI Labs — co-founded by Yann LeCun — is deliberately steering clear of the industry's favorite buzzwords. Speaking at the International Conference on Machine Learning (ICML) in Seoul last week, LeBrun said AMI Labs does not use the terms "AGI" or "superintelligence" to describe its work.
His reasoning is pointed: LeBrun noted the industry appears to have shifted from "AGI" to "superintelligence" as its term of art, but he argues there's no good definition of "superintelligence," calling it "not a very useful word."
Big money, no product yet
The restraint on terminology comes despite — or perhaps because of — a massive vote of investor confidence. AMI Labs raised $1.03 billion in March at a $3.5 billion pre-money valuation. Yet the company remains pre-product, with no commercial offering to sell today.
When asked about a launch timeline, LeBrun offered little beyond a tease: "we'll make a surprise when we're ready."
The partner strategy
Rather than build and ship a product solo, AMI Labs is courting robotics, manufacturing, and electronics companies as partners. LeBrun said these partnerships provide easier access to the real world — a critical input for training and validating world models against physical environments.
This approach lands in a moment of major regional momentum. Seoul announced a plan in June mobilizing roughly $880 billion for chips, AI data centers, and physical AI. LeBrun referenced Korea's rapid adoption of the internet 25 years ago as a comparison point for how fast the country can move on new technology waves.
AMI Labs already has a foothold in the region: JP Lee, CEO of SBVA and one of the company's Asian backers, is encouraging AMI to establish a presence in Korea.
The safety caveat
For all the funding and partnership momentum, LeBrun was candid about a core technical barrier. He said robots are "not safe right now" and that there is currently no solution for that problem — a striking admission from the CEO of a company actively pursuing robotics partnerships.
Why founders should care
- Language matters for positioning. AMI Labs' avoidance of "AGI" and "superintelligence" suggests some founders may see strategic value in resisting hype-driven labels to manage investor and market expectations — though it's unclear how widely this approach will be adopted.
- Pedigree and pre-product raises can coexist. A $1.03 billion round at a $3.5 billion pre-money valuation, without a shipping product, likely signals that some investors are willing to bet heavily on founding-team credentials and technology thesis alone — a pattern worth watching for its rarity.
- Partnership-led go-to-market may be capital-efficient. By seeking robotics, manufacturing, and electronics partners instead of building end-to-end products, AMI Labs' model could offer a template for founders navigating capital-intensive physical AI deployment.
- Safety gaps are likely a gating factor. LeBrun's admission that robots are "not safe right now" suggests founders in physical AI should probably factor safety-solution development into commercialization roadmaps well before launch.
- Regional opportunity may be expanding. Seoul's ~$880 billion plan and interest from Korean investors like JP Lee point to what could be a growing hub for AI startups seeking international expansion, particularly in chips, data centers, and physical AI.
What's still unclear
The report leaves several open questions: no details were given on the specific technology or "world model" AMI Labs is building, no timeline exists for a first product launch, and it's unclear what safety solutions the company is pursuing for robotics deployment. The terms of its prospective partnerships with robotics and manufacturing firms also remain unspecified, as does how the March funding round connects to AMI Labs' current pre-product stage and planned use of capital.
The risks on the table
AMI Labs' pre-product status means commercial viability is still unproven. LeBrun's own acknowledgment that robot safety remains unsolved points to unresolved technical barriers standing between the company and real-world deployment. And its stated reliance on external partners for real-world data access could introduce dependency risks down the line.