Nvidia is reportedly positioning itself to become the ‘Android for robotics‘. Arm, meanwhile, has created a fully-fledged business unit for ‘Physical AI’ alongside its other two divisions for cloud/AI and edge. The priority in both cases is clear: innovation is moving into the physical domain, with new pioneers for the next step in IT systems. The rhetoric, however, is jumping the gun a little.
It will not have escaped anyone’s notice that CES, normally a consumer show, was almost entirely dominated by business IT this year. But it cannot be seen entirely as simply ‘the AI show’ as it appears at first glance. Nvidia’s presentation focused on the upcoming Vera Rubin platform to drive the next generation of generative AI, but once again there was a striking amount of attention for the supposed robot revolution that merits a separate treatment. Since the emergence of deep learning in the early 2010s, Nvidia’s enterprise efforts has been targeting industrial automation and digitization. You can bundle it with IoT, edge devices, etc., but thanks to the post-ChatGPT view on AI’s capabilities, physical robots with “thinking capabilities” suddenly seem feasible. These would not just be AI agents within IT systems, but objects that grasp, move, and transform the world.
As mentioned, Arm is following a similar path, at least structurally. Executives there have told Reuters about their ambition to increase their presence in the robotics market. The emphasis is on the most obvious application: automotive. Cars are gradually becoming self-sufficient, autonomous vehicles (opinions differ on the pace at which this is happening). To enable them to operate safely and consistently in all conceivable scenarios, substantial computing power on board seems inevitable. That is why Arm sees itself as (excuse the pun) a pioneer for the future technological standard. Without its own division, Arm already has a dominant position in the automotive industry: 94 percent of all car brands use Arm technology in their vehicles.
A less compelling story
At Nvidia in particular, it is striking that CEO Jensen Huang does not think as monolithically about his company as you might expect. Although Nvidia has transformed itself in just a few years from a relatively obscure gaming chip player to the pickaxe seller of the AI gold mine, its setup remains diverse. In addition to the raw hardware, the vertical integration with software is enormous, with Nvidia trying to keep much more than just the drivers in its own hands. AI developers with a certain level of technical expertise think in terms of CUDA, Nvidia’s exclusive technology. But Huang and his team hope to penetrate industrial players, including car brands and related companies, in the same way. This, if succesful, turns Nvidia from a leading IT company to one dominating entire sectors.
In comparison with Android, Arm has taken on the same role, albeit without an iOS/Apple equivalent, while Nvidia is eager to be the Google of this domain. If Nvidia succeeds in scaling its technology to physical robots and cars, it may well succeed. In concrete terms, this means that today’s data center GPUs will eventually also be able to run on board cars, pick-and-place machines, and, yes, humanoid robots. This requires a huge efficiency boost, especially since battery capacity is not increasing at all over the years, as can be seen with (AI) processors.
This makes the robot revolution less compelling, or at the very least far more abstract, for Nvidia than the data center world, where it has long ceased to be the limiting factor. Currently, the capacity of construction companies, the power grid, and chip manufacturer TSMC seem to be a greater limitation to the advance of AI than Nvidia’s chips. Nevertheless, Nvidia is trying to circumvent such limitations with brute force (see: Rubin), connectivity, and software. Within the physical AI world, there is not yet any adoption capable of encountering similar limitations. The evolution of machinery beyond deep learning to its own equivalent of ‘agentic AI’ has not yet happened. Nvidia and Arm hope to stay ahead of this step and position themselves optimally now.
It is too early to proclaim a ‘ChatGPT moment’
It is clear that Nvidia and Arm cannot wait for robotics to drive the same growth that AI is currently driving. Nvidia CEO Huang’s goal is simple: AI must understand the physical world with the same “common sense” as humans. He went on to say during CES that the “ChatGPT moment” for physical AI is already here. That sounds very premature to us. The reasoning is also inconclusive. The ChatGPT breakthrough showed that generative AI was particularly interesting for ordinary users. Underneath, it used the same transformer technology that had already been discovered in 2012 and developed behind the scenes at OpenAI and Google between 2015 and 2022. The first model behind ChatGPT already existed in rough form in 2020 (GPT-3), but needed refinement to function as an interactive chatbot. GPT-3 already showed the same traits as ChatGPT, but its potential was far from being fully exploited. Even GPT-3.5, which came after it, was only a precursor to what GenAI can do today with agentic workflows, “reasoning,” and RAG.
At most, physical AI is in a “GPT-3 moment,” in which the foundation for a breakthrough in physical AI has been laid. This assumes that today’s technologies are sufficient for robotics, as long as enough thought is given to the precise application, guardrails, and ecosystem. Nvidia and Arm realize that these components are being considered right now, and they want that to be the case. In a few years, we may well be looking forward to a real ChatGPT moment for physical AI. Perhaps it will come just in time to answer the question of whether AI is a bubble with a resounding “no, because.”
Read also: How Nvidia is driving the robot revolution