Skild Has Amazon Backing, a Massive Valuation, and a Ton of Talent – But What Is It?

Skild AI has the early markings of a behemoth. Founded in 2023 by talent at Silicon Valley flagships, it made waves last year with a $300 million Series A featuring Amazon and Softbank that valued the firm at $1.5 billion. A rumored Series B, featuring NVIDIA and Samsung, bumps that figure to an even more unfathomable $4.5 billion.

In the interim, it hasn’t been clear what the firm is actually building, aside from a pile of venture capital. The term “physical AI” may be enough to get your pitch deck an audience with an eager investor, but the truth is in the execution.

For many, the phrase reflects the next frontier beyond the buzzy LLMs that have dominated the last few VC hype cycles. Simply put, these are companies looking to move generative artificial intelligence into the physical realm – a big next step positioned as the future of AI and robotics, alike. It also marks a move toward more versatile, multi-purpose robots.

Skild this week offered a glimpse at what it has been stealthily toiling away on at its Pittsburgh offices. Skild Brain is the source of some lofty promises. The company sent me a statement this week that refers to the platform as a, "one stop AI Model for all different robot forms.” The company believes it has taken fundamental steps toward a kind of platform agnosticism, allowing for a version of physical intelligence that can be applied to a broad array of form factors.

“ Skild’s Brain can be adapted to different tasks and different robot morphologies," the statement continues, "from quadrupeds to humanoids, from table-top arms to bimanual AMRs; from home tasks such as dishwashers to physically challenging tasks such climbing a slippery slope."

To drive the point home, the company celebrated the soft launch with a video featuring humanoids, quadrupeds, bi-dexterous manipulations, and AMRs performing a wide range of challenging tasks.

A sneak peek is, however, exactly that. In a blog post accompanying the news on Tuesday, the company writes, “Over the next month, we’ll dive into the capabilities previewed in the teaser, and show how robust our foundation model has gotten through continued training and algorithmic innovation, as well as all the emergent behaviors developed by the model.”

Skild goes on to state that much of what is currently being passed off as robotics foundational models at the moment, “start[s] with an existing vision-and-language model and sprinkle[s] in less than 1% of real-world robot data to build a ‘robotics foundation model.’

Skild Brain's “true robotics foundation model” utilizes familiar avenues for training and the gathering of huge sums of data. A combination of simulation and video data culled from the internet is used to pre-train the system, before post-training with real-world data.

Sizzle reels are not real world performance, of course. I’ve reached out to the company to discuss the offering in more detail, and will share that conversation as soon as I’m able.