Another Stealth “AI Acquisition”

Over the U.S. Labor Day holiday weekend, another stealth acquisition took place in the artificial intelligence (AI) industry.

I used the word “stealth” intentionally, because nowhere in the headlines will we find the word acquisition. Even though an acquisition is, in fact, what’s happening.

Amazon announced that it “hired” (I would use the word “acquired”) Covariant’s founders Pieter Abbeel, Peter Chen, and Rocky Duan.

Included in the “hiring” is also about 25% of Covariant’s employees.

Historically, when we saw a large, select team from a private company get acquired en masse, it usually meant that the company was in trouble. Oftentimes, these deals were structured as asset sales – to pay very little for whatever products were developed and to hire the best people while leaving the rest behind to fend for themselves.

But that’s not the case here…

Covariant’s Success

Covariant is one of the leaders in robotics automation technology.

Its technology is particularly good at picking and sorting just about any item imaginable. Its technology can be seen in action in the short video below in a logistics facility.

Source: Covariant

Covariant has been building an AI foundational model for picking and sorting objects built upon the largest dataset in the industry.

The company demonstrated impressive progress, even winning a robotic picking competition held by ABB, besting 19 other competing companies.

One executive in the competition even stated, “We’ve never seen this quality of AI before,” referring to Covariant’s technology.

Covariant was a company that I was watching closely. It had already raised $245 million to date and was backed by investors like Bill Gates, Industry Ventures, Index Ventures, Temasek Holdings (sovereign wealth fund of Singapore), Lux Capital, and even two of the godfathers of deep learning.

The team at Covariant was clearly on to something big. In fact, it was in the process of raising additional capital this summer…

That was… until Amazon came along.

Unofficial

In addition to the “hires” that will all be joining Amazon’s Fulfillment Technologies and Robotics team, Amazon struck a deal to license, on a non-exclusive basis, Covariant’s robotic automation foundation models.

And because this was struck privately between the two companies – and wasn’t officially an acquisition – we don’t know much else regarding the terms and conditions of the deal… only that two of the executives remaining at Covariant will step up into leadership roles to support Covariant’s existing customers.

Robotics automation technology is critically important to Amazon’s entire e-commerce business. Its strategy to employ robots to help address labor shortages and improve logistics efficiency began in earnest in 2012 when it purchased Kiva Systems for $775 million. Kiva’s technology became the core of Amazon’s robotics division.

It’s not a stretch at all to say that Amazon wouldn’t be anywhere near as large and successful today if it hadn’t made that investment into robotic automation. Which is why the deal with Covariant comes as no surprise.

Amazon’s deal with Covariant is remarkably similar to Amazon’s deal with Adept AI that happened this June. I wrote about it in The Bleeding Edge – Amazon’s AI Strategy Unfolds.

Adept was building multimodal foundation AI models with a focus on applications in the workplace.

The team that Amazon selected from Adept AI joined Amazon’s AGI Autonomy team (Artificial General Intelligence). Those who remained at Adept will continue to build out the product for real-world applications. A similar license structure was put in place with Adept AI, as is with Covariant.

And it’s not just Amazon doing these kinds of deals.

The AI Shell Game

This March, Microsoft hired a large group from Inflection AI, which was also one of the most interesting AI startups to watch.

In 2022, Inflection raised $265 million at a $1.23 billion valuation right out of the gates.

That was one of the largest initial funding rounds in history. And then last summer, it pulled in an additional $1.3 billion given the progress that it had made.

Inflection AI was building its own large language model (LLM) with a focus on creating personalized AIs for everyone on the planet. Inflection was working on more human-like AIs designed to be helpful and empathetic.

I wrote about Inflection AI on March 26 this year in Outer Limits – The AI Shell Game. In that issue, I explained how Microsoft paid Inflection a $650 million “licensing fee” that will make Inflection’s personalized AI available for sale on Microsoft’s cloud service, Azure.

The deal was clever in that all investors from both funding rounds received their initial capital back, plus a modest profit.

We also can’t forget about Microsoft’s $13 billion worth of investment in OpenAI, which grants Microsoft an exclusive license to the source code of OpenAI’s large language models. This gives Microsoft an incredible competitive advantage by being able to incorporate OpenAI’s technology into all of its software products.

We dug into the details of that particular deal in Outer Limits – The Drivers Behind Sam Altman’s Ousting.

And just about a month ago, Google stepped up an hired the cofounders of Character AI… along with a large team that will be joining Google’s Gemini development team.

Just like Microsoft’s licensing fee to Inflection AI, Google paid Character AI a licensing fee for its AI technology that will return about 2.5 times the original investments made by venture capitalists in Character’s Series A round.

The pattern is clear with these stealth acquisitions.

Structure these deals not as acquisitions – but as licensing deals – which allow access and control over the technology, as well as the hiring of the most important talent. It’s a way to ensure that the next generation of product development will remain 100% in their control and avoid being seen as an acquisition.

Why? Because if there is no acquisition, there is no need for Microsoft, Alphabet (Google), Amazon, and others to file with the Federal Trade Commission (FTC).

This is intentional. These companies are making every effort to stay out of regulatory crosshairs. Had they filed with acquisitions, it would have guaranteed a regulatory review… upped the risk that the deal wouldn’t be approved… and cost millions in legal fees dealing with the whole process.

It would not only introduce the risk that the FTC would deny the acquisition, but it would also slow everything down. And in the world of artificial intelligence, speed is everything right now.

These deals are devious and clever. Legal shenanigans. A sleight of hand.

And it’s probably not a coincidence that it was announced over a holiday weekend in the U.S., while everyone was distracted.

It’s the AI Shell Game.

And it makes the “acquirers” that much more attractive as investments as they gain control of the most promising artificial intelligence and teams on the planet…