Meet NEO Beta, a recently announced humanoid robot designed for the home.
NEO Beta | Source: 1X Technologies
Designed and developed by Norway-based 1X Technologies, NEO is one of the three primary contenders racing to develop a mass-market, general-purpose humanoid robot.
1X has currently raised about $140 million to date, a relatively small sum for such an ambitious goal, primarily backed by Norway-based investors.
There are, however, three strategic investors worth noting.
Who’s taken an interest in NEO? And what for?
Aside from a large number of hardware improvements in NEO Beta compared to the previous version Eve 2, the most obvious major change has been to evolve its humanoid robot form from operating on two wheels to a bipedal (two legs) humanoid design.
NEO Beta | Source: 1X Technologies
1X has also made a leap in its training approach, using the real world and human examples as the most important inputs to training its neural network to achieve assigned tasks.
1X went so far as to say that as it trains NEO Beta, it probably won’t use past training data used with Eve 2, suggesting that its approach to software design has undergone a major change.
This is the same approach that Tesla is using for Optimus, and Figure AI is using for its latest humanoid robot Figure 02.
Figure 02 | Source: Figure AI
In fact, Figure makes an interesting case study for what we might be seeing happening with 1X’s new NEO robot today.
Not too long ago, Figure AI was on par with 1X Technologies.
However, Figure became a serious contender in the race to develop a general-purpose humanoid robot earlier this year when it raised $675 million. (To date, Figure AI has raised $845 million in total.)
Far more telling were the two lead investors in the last round: Microsoft and OpenAI, which – as my readers know – are linked at the hip. Microsoft has already invested almost $14 billion in OpenAI and is the only company to have an exclusive license to OpenAI’s source code.
The round was an absolute pile-on with 39 different investors contributing to the round, including Amazon and Bezos Expeditions (Jeff Bezos’ venture capital firm). Amazon and Jeff Bezos clearly want an alternative to Musk’s Optimus.
Figure AI needed the capital. While it had made solid progress on the hardware for its humanoid robot, it was woefully behind in artificial intelligence. The reality was that Figure 01 didn’t have much of a “brain”.
So Figure AI made a “shortcut” to catch up with Tesla’s Optimus – by incorporating OpenAI’s latest multi-modal large language models into its humanoid robot.
It makes perfect sense to leverage the incredible work being done at OpenAI, and with the pace of improvements. Every new release of OpenAI’s LLMs will be relevant to Figure’s own humanoid development – and can be incorporated into Figure 02.
And in a recent interview, Figure AI’s CEO Brett Adcock revealed even more…
He said the value that Microsoft is bringing to the table is well above and beyond just capital. “Microsoft has given us as many H100 NDs as we need.” Said Adcock. He was, of course, referring to NVIDIA’s H100 GPUs, which are the workhorse semiconductors used to train AI.
And there’s more…
Adcock went further when he said, “So we’ve scaled up now three times in the last 90 days into larger clusters.” Gibberish to most…
He’s referring to computational clusters of GPUs.
Figure is ramping quickly. And this suggests rapid progress is being made on the software.
Figure knows what every company in this space knows: If it spends more capital on computational power for training the AI models, the software that will run on Figure 02 will improve quickly. And the robot will learn how to do more… faster.
More capital…
More computational power…
More electricity…
Equals more training…
I think we can all see the common thread by now: More… now.
It has been incredible to watch how quickly the industry has gravitated towards Tesla’s design and approach to software for Optimus.
All three players are leaning heavily on leveraging real-world data to train the humanoid robots on how to perform economically valuable tasks.
This, of course, is nothing new to Tesla. It’s the same product development strategy it has been using for years.
As long-time readers know, every Tesla car is equipped with seven cameras placed around the car. Each camera has a 250-meter depth of field.
These cameras provide the primary inputs into the Tesla neural network – the backbone behind the company’s full self-driving (FSD) AI.
Tesla collects all this data from its “fleet” of vehicles around the world, feeds it to its supercomputer, and then learns, optimizes, and improves its performance. Tesla has been doing this for years.
And it’s now this software that Tesla is leveraging as the basis for Optimus’ mobility.
The end goal is for a humanoid robot to operate in any unstructured environment much like how a Tesla car with FSD can drive on roads it has never “seen” or been trained on before without any geofencing.
As for humanoid robots, the goal is to have a human provide directions to the robot, like, “NEO, watch me and learn how to assemble this device” or, “NEO, watch me and learn how to clean up this room and get it ready for bedtime.”
Humanoid robots are being designed to be taught by humans. This is a necessity if they are to be general purpose.
All of these hardware platforms – the humanoid robots -can receive software upgrades, just as any Tesla electric vehicle receives its updates today. With one push, every Tesla can download the latest FSD update… and receive all the newly learned capabilities. Humanoid robots will be no different.
And all three companies know that the more humanoid robots they have in the field collecting data and improving in tasks – through trial and error – the faster they will be able to improve their AI software.
1X Technologies is currently gearing up its manufacturing capabilities.
Its goal is to produce 100–200 NEOs a month in 2025 – the year it expects to first commercialize NEO.
2026 will bring even more scale to manufacturing, as well as developing a model of NEO for industrial applications.
Figure AI has said that based on the rapid progress that it’s seeing with its own software development, it is moving up its schedule for in-home use for Figure 02.
The race is heating up…
Given that Figure AI has already been testing Figure 02 at the BMW plant in Spartanburg, South Carolina, a topic that we explored earlier this year in Outer Limits – BMW Just Hired Its First Humanoid Robot, it’s not unreasonable to project that we’ll see commercialization before the end of 2025.
And we know that Tesla has already developed Optimus Gen 3 and has been testing it for months now.
We should have an update soon from Tesla. But we already know that Optimus Gen 3 will have 22 degrees of freedom, even having the ability to play a piano, and will be sold for about $20,000 when production is at scale.
While Tesla plans to absorb all of its Optimus production in 2025 for internal use in offices and factories, external commercialization is planned for 2026.
And this is exactly why Figure AI and 1X Technologies are pushing as fast as they can to get a competitive robot into the market before the end of 2025. They have been trailing behind Tesla’s incredible progress.
Consider this…
The timelines and the technology are converging.
It has already been an incredible year for humanoid robots.
And next year, we will be able to greet the first humanoid robot in our homes.
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
The Bleeding Edge is the only free newsletter that delivers daily insights and information from the high-tech world as well as topics and trends relevant to investments.
