[Music in background throughout opening sequence.]
On screen disclosures:
Please read important information at the end of this program. Recorded on 3/30/2026 & 4/28/2026.
Chris Hyzy
The rapid rise of artificial intelligence we've seen in recent years has felt very conversational. We've gotten to know the technology largely through prompting, chatting and generating. But beneath that digital layer, something much larger has been taking shape.
On screen copy:
Chris Hyzy
Chief Investment Officer
Merrill and Bank of America Private Bank
Chris Hyzy
Hi, I'm Chris Hyzy, Chief Investment Officer for Merrill and Bank of America Private Bank. We are witnessing the application of AI begin to drive profound advances across the economic landscape. The race is on to build the infrastructure required for AI systems to operate in the real world, not just in software.
On screen copy:
Over $41bn invested in physical AI in 2025.
Source: BofA Global Research, AI left the chat!
February 6, 2026
Chris Hyzy
More than 41 billion was invested in physical AI in 2025 alone. This isn't about better chatbots. It's about intelligence moving into robots, vehicles, factories, logistics, and more. In other words, AI has left the chat. To get a picture of where we are and where things are going. I'm talking to strategists from BofA Global Research who are closely tracking the trends.
On screen copy:
London, UK
March 30, 2026
First up, equity analyst Martyn Briggs.
On screen copy:
New York, NY
March 30, 2026
Chris Hyzy
So, Martyn, let's start right at the top. Can you please explain exactly what physical artificial intelligence is?
On screen copy:
Martyn Briggs
Director of Thematic Investing
BofA Global Research
On screen copy:
Physical artificial intelligence
AI systems that interact with, manipulate and operate in the real world.
Martyn Briggs
Sure. Thank you Chris. Delighted to be here with you and the team. So what we mean by talking about physical AI is that AI is moving from screens to the machines. It might be a robot, a car, a drone, essentially bringing that intelligence into the real world.
Chris Hyzy
So everything that we watched early on in life, whether it was cartoons, movies about what could actually happen, is starting to unfold. You talked about the screen to the machine. Can you talk about why the shift now?
Martyn Briggs
Essentially, there's a number of different things converging. What we see now is a combination of things like world models, simulation, and synthetic data that can accelerate the flywheel and bring that into the real world.
On screen copy:
Factors Propelling Physical AI:
- World models
- Expanded simulation
- Synthetic data
Martyn Briggs
So starting with those individually, a world model essentially brings together the tools: vision, language, action, vision, language models, and a number of different software capabilities to allow AI, robots, cars, drones, etc. to understand the laws of physics that if they walk into this table, they're either going to get hurt or fall over or broken, and that they can't walk straight through them.
So a different set of models to bring that to life. I also mentioned things like simulation, which can test, validate this technology much faster. One of the companies showed that one second in simulated environments would have taken 27 minutes to achieve in the real world without this simulation. And of course, that's been slow, expensive and billions of dollars that these companies have had to raise to do it.
So simulation is making things faster. And things like synthetic data. There's a lack of data available for training of these technologies. Now you can start to do that not just with real world data, but also synthetically to bring these products to market faster.
Chris Hyzy
So one area that investors talk about a lot is value and monetization. So what's the value creation and monetization end of physical AI?
On screen copy:
Potential Value Areas
in Physical AI:
- Sensors
- Actuators
- Motors
- Batteries
Martyn Briggs
The value might start to accrue in the short term in some of the picks and shovels the hardware needed. The sensors, the actuators, motors, batteries, AI chips that are required to bring this together. In the mid to long term, it's likely to be more the integrators that can bring together, be it an autonomous car, a robot, a drone, etc. but I think in the short term we're seeing value accrue in some of the development, the sensors in the hardware, the longer term will be some of the integrators of this disruptive technology.
Chris Hyzy
And when you look across the entire arc of physical AI, starting with why is the first mover advantage — no pun intended, or maybe pun intended here — mobility?
Martyn Briggs
The first reason is that we're bringing scale to the market. There's already a 75 billion sensor market in automotive for driver assistance and safety systems or ADAS systems. So we're seeing quite quickly that accelerate towards autonomous.
On screen copy:
ADAS expected in 85% of new car sales in China by 2030.
Martyn Briggs
So level two capabilities now potentially fully autonomous robo-taxi applications in the not too distant future. And there's already ten commercial cities where you can ride in robo-taxis today, and over 200 pilots across either robo-taxis or commercial trucks and logistics applications. So the scale is starting to come now. And it's bringing that intelligence into the roads, into the vehicles.
The other thing is that the centralized compute. So software defined vehicles or high power compute on wheels is a big shift that's really starting to play out now in the last few years only. Where we used to have 150 or 200 distributed electric control units in vehicles, now we're centralizing that compute and making it far more powerful that you can do much more disruptive applications on top of it. Especially around autonomous, which we wouldn't have been capable or viable only a few years ago.
Chris Hyzy
You talked about scale and the shift. The numbers are absolutely staggering when it comes to the production and almost proliferation, we're not there yet, but of humanoids. You want to dive into that a little bit?
Martyn Briggs
Absolutely. Yeah. So humanoids are one of these areas that have been very, a lot of hype and excitement around, certainly for the last year or two.
On screen copy:
Investment in humanoids has grown ~5x in 7 years
2018: $0.7 billion; 2025: $4.3 billion
Martyn Briggs
It's the pure embodiment of AI that we've just discussed throughout this interview. And things are accelerating very fast. So the hardware is getting cheaper. It's humanoid, cost 90 to $100,000, but many of the companies now say they can manufacture them for maybe 20 to 30 thousand in the not too distant future. And the type of numbers here we could move from 20,000 humanoids manufactured last year to maybe a million by the end of the decade, and even 10 million in 2035.
So we're quite quickly seeing an acceleration of the potential volumes here. If we take a step back and think about how they could be used or where is viable in the near term, it's far more in the industrial applications. Right? So warehouses, logistics, some of these areas that are being piloted already and bringing humanoids into assembly lines and low level assembly tasks where the productivity gains will justify the investment.
So it might be still a while a decade plus before we're going to have them cobots in our household or even alongside me here in the office. But for some of the industrial applications, the ROI is being proven out in the pilots and the test.
Chris Hyzy
Okay. We talked about the physical portion, the arc, value shift, monetization, a little bit on opportunities. You want to go a little bit further on the opportunistic side of the equation for investors.
Martyn Briggs
On the one hand it feels like things are becoming commercially viable in months rather than decades now, right, for some of these applications.
So what that means for investors is that we see a number of ways in the hardware and some of the picks and shovels ways to enter into the market for investors on things like sensors, actuators, body motor, batteries, etc. in the short term.
On screen copy:
Potential Opportunities in Physical AI:
- Drones
- Robots
- Interface integrators
Martyn Briggs
I think the bigger opportunity, of course, and the more disruptive one will be who is going to integrate and be the customer interface for a humanoid robot or a self-driving car or drone delivery, these type of applications. There's a mixture of private and public companies that we've identified that do this already. But to do that more successfully commercially at scale, is a combination of a few things that need to happen to get to a bigger scale regulation overseas to be a bit more amenable in certain locations.
And obviously the technology viability needs to be proven out. But the ultimately, the integrators of this technology will be the largest beneficiaries in the longer term. For now, we've identified a number of areas in the picks and shovels in hardware and companies that are developing the tools you need to develop and deliver physical AI.
Chris Hyzy
Let's shift to something that is also very important when you start talking particularly about humanoids and robots, controlling certain things themselves, which is the risks. Well, what are the top risks that we should all be aware of?
On screen copy:
Risks of Physical AI:
- Potential job loss
- Hallucinations
- Pace of regulation
Martyn Briggs
The biggest risk, of course, will be impact on employment and the pace of that. Also the risk of the technology hallucinating, not as much of a risk in a digital application on a 2D screen and a chatbot, you know, you either do the calculations yourself and come up with an alternative. If there is any kind of hallucination or, in the real world, be that on a road or if a humanoid falls over, there's real safety risks.
So the technology needs to be proven out to, you know, far higher degree of safety than you would have into the chat bot application clearly. And the final risk, I think, as we alluded to earlier, is the pace of regulation and standards, and safety standards for these services to operate. Every region has a slightly different view on that, and there's nothing really exists that standardized globally, which might slow the pace of development of commercialization of some of the technologies. So they're the three main ones.
Chris Hyzy
That reminds me of some 30 plus years ago with the build out of the internet itself Martyn. Final question for you. Tough one here. One word answer about physical AI and its future. What do you say?
Martyn Briggs
Exciting.
Chris Hyzy
Martyn, I want to thank you for joining me today.
Martyn Briggs
Thanks, Chris. Good to be with you.
On screen copy:
New York, NY
April 28, 2026
Chris Hyzy
Let's turn the page now from one aspect of physical AI to another, where AI is moving from software into machines, robots, and mobility, what then comes into view are areas like transportation and semiconductors. To explore these themes, I'm joined by two analysts from BofA Global Research, Ken Hoexter, who covers transportation, and Vivek Arya, covering semiconductors. Ken, Vivek, welcome.
So let's start with you, Ken. Take us through what you're seeing in the transportation sector right now, simply from is it being deployed? What's in the way? How is it being managed? Take us from the top.
On screen copy:
Ken Hoexter
Research Analyst
BofA Global Research
Ken Hoexter
Well, there's a couple of ways. One is is you've got a lot of physical assets that are moving freight around the country. Trucking moves 73% of all stuff in the U.S. And you've got a lot of empty capacity that's trying just to get from one place to another to pick it up. So AI, is able to match that supply and demand in a brokerage capacity and really try to that's the first place.
Chris Hyzy
So first is logistics.
On screen copy:
AI Use Cases in Transportation:
- Logistics optimization
- Autonomous trucking
- Route efficiency
Ken Hoexter
Logistics. Absolutely. That's the first place you're seeing it really roll out. Then you take that from just the existing capacity to then the mobility of the future. We have autonomous trucking that's, you know, on the highway. So now you've got nobody in the cab and driving from a starting point onto the highway and then going all the way to a terminal. So we're doing terminal to terminal at this point in the testing phase in the south where the weather's good. And that's kind of the first phase of where we're seeing it. We're not yet at the point of going all the way through the local roads, you know, and doing drop and hook, but right now it's going from the open road.
Chris Hyzy
Yeah, certainly transformational. But let's talk about the infrastructure. Ken just talked about logistics as an initial starting phase. There's a massive infrastructure needed for that, semiconductors in particular. Can you take us through your thoughts around that?
On screen copy:
Vivek Arya
Research Analyst
BofA Global Research
On screen copy:
Building the AI "brain"
1. Training phase
Vivek Arya
Sure. I think there are two aspects to it, Chris. One is that even before we start deploying that infrastructure, we first have to create that brain, right? So we first go through what's called the training phase, where, you know, they will define the problem that, okay, now you need, that physical AI, whether it's a drone, whether it's a truck or something else, you need to be able to teach it about it. So it's almost like training a child. You have to teach it about its surroundings. You know, teaching a child the difference between a dog and cat is easy. Teaching that to a computer. Insanely difficult. And then once that brain is start, then you will try to go through a second phase, which is simulation, right?
On screen copy:
Building the AI "brain"
- Training phase
- Simulation phase
Vivek Arya
And in many ways, the interesting thing is that the companies that are involved in this are many of the same graphics companies that are also involved in the gaming industry. Because they have to take the car through a set of obstacles and make sure that it can react very quickly to, somebody coming in front of it.
Chris Hyzy
And you're creating a synthetic environment.
On screen copy:
Building the AI "brain"
- Training phase
- Simulation phase
- Trials
Vivek Arya
Exactly. So you create that. And then the third phase is now that you are satisfied with the quality of the brain, you are able to put it out there. Now you start doing test trials. So it's when you get to scale is now you have this backup infrastructure, right, that you can call on. That infrastructure needs to be flexible, you know. So for example, if it is supporting self-driving cars, you have a lot more cars driving during the day than at night. So what do you do with that infrastructure at night?
Chris Hyzy
That downtime.
Vivek Arya
Maybe you use it to train the next brain. So you're absolutely right. I think having that infrastructure and these different phases of how sensing, computing, you know, storage, networking, I think all these are very important components and semiconductors are involved in in each of those.
Chris Hyzy
And that's semiconductors. You mentioned graphic cards. You mentioned graphic. What I call a networking way to learn on itself. Right. That's very different than 20, 30 years ago when it was memory, memory, memory for the most part. How has the semiconductor industry changed, and how is it going to change in the next 2 or 3 years?
On screen copy:
Serial computing: tasks are completed one at a time, sequentially.
Parallel computing: many tasks run at the same time, across processors.
Vivek Arya
I think the biggest transformation has been in this move from serial to parallel computing. Okay. Right. So serial computing is like you have an army that has just one soldier. It breaks the tasks into small steps and it does that task one after another. That's what the CPUs in the PCs used to do. Now the problem is that data is not stored in such a simple way.
So if you take, you know, your favorite social networking application. It knows 5000 things about you. How do you put that in a simple database? You can't. So now you have to train the computer to think in a very parallel way. And that is what a lot of these graphics chips are insanely good at. Thinking about what is called unstructured data, right?
And AI is one of the best examples of unstructured data, right? When we go and ask these AI chat bots a question that isn't like a simple database that they can just look into and give us an answer. They're able to construct answers. So what is different is this move from serial to parallel computing. The second big difference is that doing parallel computing is easy to say, but it is very power intensive. That takes a lot of space, power, transportation, utilities. So now you have to worry about a lot of that infrastructure making. So those are probably the constraints.
And that's why a number of the companies that we track, now they have gone from being just a chip company to being systems companies. Right. They have to think about every aspect of the stack, how to bring it all together. And the ones that are successful are the ones that I think can think about this problem as an end-to-end systems problem, as opposed to, "oh, I just make the chip, throw it at a customer and let them worry about the rest."
Ken Hoexter
And take that one step further. That's exactly what the trucks are doing, is if you find a construction detour, the first truck that runs it now is telling all the other trucks in that system, hey, this you've got to learn this. And in the downtime, they're training each other on new scenarios. So it's constantly learning.
Chris Hyzy
What about the cost line for these companies? Are there ways that the transportation sector is already using this intelligence, if you will, at the cost line to expand margins?
Ken Hoexter
You just think about the excess utilization. 10-15% of your capacity is just deadhead empty driving. So if you can refine the routing and to keep the drivers busy, using AI to refine that matching of the supply and the demand getting just much more efficient. So that's the that's certainly the first phase. But you still have depreciation. You still have gas until we go all electric. You still have all the other costs and you still have congestion issues on the highway. But if you're autonomous, you can do things at different times of day. So you were talking about at night driving. So if you can do that, you can just get more efficient with use of the asset.
Chris Hyzy
As safety standards improve and with autonomous vehicles, is that something that could potentially change and bend the cost curve a little bit better?
Ken Hoexter
Presumably, a computer would be a better driver in proven circumstances than a human because it can react a lot quicker and can anticipate certain things. If people are doing what they're supposed to be doing, as opposed to not right. So in those cases, absolutely, you can start to see insurance costs come down, labor costs come down. But the regulatory environment has to adapt over time. Right. So you definitely have this corridor down south that is willing to test it. There are certain states that are holding out, they want to see it get a little bit further, but we're having robotaxis and autonomous trucks testing in many, many different markets around the country. So it's moving faster than the regulatory environment can adapt.
Chris Hyzy
That's amazing. Let's talk about the final question. Let's talk about risks. Clearly there's risks in the transportation sector. As you get more and more physical artificial intelligence infused into the business. And what are the top few that come to mind?
Ken Hoexter
Well, look, I mean, every safe, safety has to trump everything. Yeah. There is no room right? You get one accident and it can set the industry back years in development. Right. Or months, quarters, years. So you have to make sure that it is ready for prime time as you roll this out. When you're putting it on, if you're going to have your family of five driving next to a truck, an 80,000 pound truck with nobody in that cab, right?
On screen copy:
Risks for AI in Transportation:
- Safety incidents
- Hacking potential
Ken Hoexter
You have to make sure that it is buckled up. So that's the only risk that matters to start, right? Yeah. Then the risk is, you know, can it be hacked? What's the protection of the of the asset? You have to remember, this is this is a company's freight. That's their lifeline. So if I'm going to give it to somebody it's my brand name on it. It's their freight. It's their goods.
Chris Hyzy
That's a great point. Vivek. How about you?
Vivek Arya
Yeah, I think the same thing in that, it has to be evolutionary, right? I mean, technology do many, many things. Right? But as I mentioned, how quickly can a specific society adopt it? Right. And we see different behavior in Asian versus Western right societies. Right. There's a lot more deployment and adoption of robotics. Right in, in the East. Right in Japan. Korea. Right. There's a lot more, deployment and adoption. Right. Because it's just tougher to find help. There there's a lot more older people in certain places. But one thing the automotive industry has done, is go through different levels. So cars come in different levels of safety. So level one is just kind of a plain vanilla car. Level five is your ultimate self driving go anywhere and then you have level two, level two plus, level three. So every step brings about more safety features. And they are very careful. And calling it copilots and assisted driving. Right. Like you know we have copilots in the sky. So I think it's going to be evolutionary, but I think it is inevitable.
Chris Hyzy
This has been fascinating. Ken, Vivek, thanks for joining me today.
As AI moves out of the digital world and into the physical one. Execution, reliability and scale begin to matter more than just novelty. This next phase may be harder, but it's also where the real economic impact will be created. Thanks for joining us. We'll see you next time.
On screen disclosures:
Important Disclosures
The opinions expressed are as of 3/30/2026 and 4/28/2026 and are subject to change.
Investing involves risk, including the possible loss of principal. Past performance is no guarantee of future results.
Asset allocation, diversification, and rebalancing do not ensure a profit or protect against loss in declining markets.
Investments have varying degrees of risk. Some of the risks involved with equity securities include the possibility that the value of the stocks may fluctuate in response to events specific to the companies or markets, as well as economic, political or social events in the U.S. or abroad. Investments in a certain industry or sector may pose additional risk due to lack of diversification and sector concentration. Investments in foreign securities involve special risks, including foreign currency risk and the possibility of substantial volatility due to adverse political, economic or other developments. These risks are magnified for investments made in emerging markets.
This information should not be construed as investment advice and is subject to change. It is provided for informational purposes only and is not intended to be either a specific offer by Bank of America, Merrill or any affiliate to sell or provide, or a specific invitation for a consumer to apply for, any particular retail financial product or service that may be available.
The Chief Investment Office (CIO) provides thought leadership on wealth management, investment strategy and global markets; portfolio management solutions; due diligence; and solutions oversight and data analytics. CIO viewpoints are developed for Bank of America Private Bank, a division of Bank of America, N.A., ("Bank of America") and Merrill Lynch, Pierce, Fenner & Smith Incorporated ("MLPF&S" or "Merrill"), a registered broker-dealer, registered investment adviser and a wholly owned subsidiary of Bank of America Corporation ("BofA Corp.").
Merrill makes available certain investment products sponsored, managed, distributed or provided by companies that are affiliates of BofA Corp. MLPF&S is a registered broker-dealer, registered investment adviser, Member SIPC and a wholly owned subsidiary of BofA Corp.
Merrill Private Wealth Management is a division of MLPF&S that offers a broad array of personalized wealth management products and services. Both brokerage and investment advisory services are offered by the Private Wealth Advisors through MLPF&S. The nature and degree of advice and assistance provided, the fees charged, and client rights and Merrill's obligations will differ among these services. The banking, credit and trust services sold by the Private Wealth Advisors are offered by licensed banks and trust companies, including Bank of America, N.A., Member FDIC, and other affiliated banks.
Bank of America Private Bank is a division of Bank of America, N.A., Member FDIC and a wholly owned subsidiary of BofA Corp. Trust and fiduciary services are provided by wholly owned banking affiliates of BofA Corp., including Bank of America, N.A.
BofA Global Research is research produced by BofA Securities, Inc. ("BofAS") and/or one or more of its affiliates. BofAS is a registered broker-dealer, Member SIPC, and wholly owned subsidiary of Bank of America Corporation.
Investment products:
| Are Not FDIC Insured |
Are Not Bank Guaranteed |
May Lose Value |
© 2026 Bank of America Corporation. All rights reserved. 8894963 - 06/2026
[End of transcript]
"AI is moving from screens to the machines," says Martyn Briggs, director, Thematic Investing, BofA Global Research. "It might be a robot, a car, a drone, essentially bringing that intelligence into the real world."
As artificial intelligence (AI) rapidly evolves, its impact is expanding far beyond chatbots and digital applications. A new phase is emerging — one where AI moves into the physical world, powering machines, vehicles, logistics and industrial systems.
This shift is driving a surge in infrastructure investment and innovation, as companies and countries race to build the foundation that will support AI at scale. From semiconductors and data centers to transportation networks and autonomous systems, the implications are broad and increasingly tangible.
"We are witnessing the application of AI begin to drive profound advances across the economic landscape," says Chris Hyzy, Chief Investment Officer for Merrill and Bank of America Private Bank.
Watch the video above to understand how quickly physical AI is developing and how it could drive the next phase of economic growth and potential investment opportunity. In it, Hyzy is joined by a trio of BofA Global Research analysts: Martyn Briggs on the broader expansion into the physical world, Ken Hoexter on the impact on transportation and Vivek Arya on how semiconductors and infrastructure tie it all together. Together, they explore how AI is transitioning from software into the physical economy — the risks as well as potential opportunities — and what it could mean for markets, industries and investors.