Robots - horizontal or vertical play?
Robotics companies are starting to scale... which way will they go?
Right now whole robot systems are where the value is, even if they’re single purpose. Components can be a race to the bottom, but then many humanoid companies were actuator companies first and a vertically integrated company has a lot of benefits. And that’s the opposite of aiming for general purpose robots.
I’ve been wondering about this and how much it relates to the stages of category creation and development. There’s evidence of successive cascades in vertical/horizontal approaches - for example the rise of TSMC and the contract foundries vs the chip manufacturers like Intel owning the whole stack.
BTW - the Acquired podcast just remastered their TSMC episode and I totally recommend it. We normally don’t hear enough about complex hardware companies. I’m going to paste in a small section, but you also need to place this in historical context. TSMC has just started with a hope to provide contract fab services to startups and they’re surviving by doing the low grade overruns of the main chip players. Basically their real market doesn’t exist yet, and it’s a very lucky and/or smart and/or timely bet that it all came together for them.
Acquired podcast - TSMC - excerpt
So the fabs market starts to grow, which there exceeding and enabling it. As that happens, T SMCs revenue grows and because they have 50% gross margins and 40% operating margins, they can take that profit and buy more advanced machinery back into CapEx and build more fabs..
Advance the level of their technology. Remember they were starting from behind and technology within about 10 years, they catch up and then they start to exceed everybody else. So as they push the manufacturing process technology forward, they're manufacturing better chips with smaller wave, you know, process lengths. They're enabling their customers, which are the fabs companies to get better and better performance. As they get better performance. The fabs companies can address more of the market and more use cases. So their existing customers get bigger and new Fbu customers start, which gives them more revenue, which repeats the whole cycle. And you know, it goes slowly like any flywheel, it takes a lot of effort and a lot of time to start turning it.
But fast forward to now. So in the early two thousands when TSMC finally caught up to the bleeding edge level of technology with other semiconductor companies, there were 22 companies that were at the leading edge. I think it was like, I don't know, let's call it 150 nanometer process or something like that. At that point in time, 22 and TSMC finally broke into the pack. They were one of the 22. By the late two thousands it had gone from 22 down to 14 that were at the leading edge by the mid 2010s. There are six.
It's basically Samsung and TSMC. Right.
Today there are two at five nanometer processes. The current leading edge only TSMC and Samsung. Intel has been trying to get there, but they haven't been able to. They've fallen behind. And the next process is gonna be three nanometers. TSMC is gonna launch that next year.
Sam, which by the way just slipped six months.
Ah, interesting. Well Samsung has already slipped to 2024. So very likely in the next process is just gonna be TSMC.
Which means that you will see that on an Apple slide somewhere announcing the next iPhone, talking about how it's a three, three nanometer process. They'll take all the credit for it. And TSMC is totally fine with that because their job is not to market. It's to empower their customers
This flywheel. It's just unreal what happens here. They run the table and the whole industry.
It is interesting, the industry went from vertical to horizontally integrated where the very best products in the market became horizontally integrated. And it's interesting how, I'm trying to figure out what drove that. Because at some point, I guess there's a couple components to it. One is the speed at which Moore's law happens makes it such that you can't be good at everything. You can't be good at everything from EDA to making the manufacturing equipment, to running the manufacturing process, to designing the chips. You're not gonna write your own instruction set. Architecture people did need to break into best of class. Oh,
Morris has got this great quote about this that I have in here. So he says the semiconductor business is like a treadmill that speeds up all the time. If you can't keep up, you fall off. And that's Moore's law from 22 down to two, down to one. Even when their competitors are only doing the one thing that TSMC has done. Like if you fall behind by a step, you're toast.
Right. And it's because there's this big part of it that you're talking about that hasn't come up on other episodes. 'cause we tend not to talk about companies that require a lot of manufacturing prowess. But in order to stay on that treadmill, the number of tens of billions of dollars that you need to be spending into CapEx is going up. So you need to be enormously profitable so you can build the factories for the next generation.
Yeah, I Mean, well there's two things. So yes, that is a hundred percent true. And the scale of this now, I Mean TSMC just announced they're gonna spend a hundred billion dollars in CapEx over the next three years. $30 billion this year. 60 over the next two. And I bet that keeps going up. So that's a lot of billions. You might even say like this is so strategically important and people are talking about this, certainly China's talking about this, the US government's now talking about this. Governments might need to come in with a bazooka of money and create other options. 'cause almost all their manufacturing's in Taiwan, it's in this strategically geopolitically challenged location. We need to reshore some of this in the us. China Of course wants their own. You can't just spend the money and do this.
US government could come in and say we're gonna spend a trillion dollars this year to do this. They can't do it because we're gonna get to powers later. But there's this marriage of scale, economies and process power. The TSMC, like in this industry, there is no amount of money you could spend to catch up next year. You can't because the engineering is so hard and the learning curve takes decades to get to this point. I was listening to a podcast, a Bloomberg Odd Lots podcast about this where they were talking about this and their reporter who covers TSMC, who's great China, they asking the question like the, well China could just spend a billion dollars and do this, create their own fabs. And they're doing this,
What's the company called? SMIC?
etc. etc. Timing is everything, but if you get it right you’re going to own the table. Now how does this apply to robotics technologies? Love to hear your thoughts.
This timely piece from CBInsights just happened to land in my inbox straight after the TSMC episode. And then Stargate dropped. Read on below for The Information’s take.
CBInsights - Chart of the Day - excerpt
Nvidia has leapfrogged other big tech companies like Microsoft and Amazon in overall venture activity.
That gap is even more pronounced when looking at AI dealmaking: Nvidia’s AI startup investments nearly 5x’d between 2022 and 2023.
CB Insights customers can use this deal search to track all of big tech’s AI investments.
Although Google remains far more active generally (it backed 133 deals in 2024 vs. Nvidia's 55), Nvidia’s upswing indicates the strategic importance it’s placed on being a player in the AI startup landscape.
Of course, many of its investments — like Perplexity and xAI — are in turn using its chips.
CB Insights customers can check out Nvidia’s growth strategy here.
All systems go
AI and industrial automation dominated the fastest-growing tech markets of 2024.
Out of 1,400+ tech markets CB Insights tracks, autonomous agents and generative AI for customer support saw the highest rate of YoY deal growth (150%).
Generative AI advancements also fueled activity in areas like humanoid robots and autonomous driving.
See the full State of Venture 2024 Report for more on the markets reshaping the tech landscape.
OpenAI, SoftBank Each Commit $19 Billion to Stargate Data Center Venture
OpenAI and Japanese conglomerate SoftBank will each commit $19 billion to fund a joint venture to develop data centers for artificial intelligence in the U.S., OpenAI CEO Sam Altman told some colleagues Wednesday.
The ChatGPT developer would effectively hold a 40% interest in the venture, Stargate, and it would act as a kind of extension of OpenAI, he said. Stargate stems from Altman’s longtime concern about his company’s access to servers as it faces rising competition from Elon Musk and others. Altman’s comments imply SoftBank would also have a 40% interest.
On Tuesday at the White House, Altman and the CEOs of SoftBank and Oracle, another member of the joint venture, said the venture would raise $100 billion initially, and $400 billion more over several years, to develop new facilities that would boost AI for curing diseases and solving other problems. The companies didn’t immediately disclose details about the venture but Altman told later colleagues the goal was to produce a relatively cheap source of servers to help OpenAI develop its technology.
Altman’s comments on Tuesday suggest the company will have to raise $19 billion through equity or debt, and company leaders previously told colleagues they were prepared to raise debt for data center projects. The company projected to generate about $4 billion in revenue last year while still burning considerable sums of cash due to high computing costs.
OpenAI raised $6.6 billion in equity funding last fall at a $157 billion valuation and has raised a total of about $20 billion, mostly from Microsoft, which has rights to OpenAI’s intellectual property but is largely unaffiliated with Stargate.
The scale of the data center venture is unprecedented. On Wednesday, Altman likened Stargate to a venture fund, with OpenAI and SoftBank as two of the general partners. Two other general partners will be Oracle and MGX, an Abu Dhabi fund that also owns a piece of the new Delaware-registered venture. GPs would commit $45 billion in total to the project, he said, implying that $7 billion would come from Oracle and MGX.
The rest of the money for Stargate would come from investors categorized as limited partners as well as several types of debt financing. Such debt could eventually trade publicly, he said.
Altman said OpenAI would be a customer of Stargate but would also have operational control of it. It’s possible that Stargate would raise even more capital than the $500 billion that Son had stipulated, at which point OpenAI would lose some control.
One more gem in the Acquired podcast that I think you can apply to some of these snippets: David Rosenthal offered up this rule “Don’t make strategic decisions based on cost/price”.
In other words, if something is a strategic purchase, don’t nickel and dime the valuation. You aren’t making a balance sheet decision. You’re making an existential one.