That's how much TSMC generated in revenue last quarter. Sales jumped 41.2% from a year earlier, and its gross margin reached an almost absurd 66.4%.
Meanwhile, ASML, the Dutch company that makes some of the world's most advanced chipmaking equipment, says its planned production capacity for 2027 and 2028 won't be enough to meet projected customer demand.
Now here's the other side of the story.
American factories are running at just 75.9% of their production capacity.
One corner of manufacturing is desperate for more capacity. Much of the rest has capacity to spare.
And if you're investing in robotics or physical AI, understanding that difference might be more important than watching another humanoid robot do backflips.
The market isn't buying every AI story
The Black Scarab Physical AI 15 dropped 4.77% over the five trading sessions ending October 20, according to our internal calculations.
That's considerably worse than the S&P 500's 1.19% decline and the Nasdaq's 2.87% loss over the same period.
And underneath those numbers, things got interesting.
NXP Semiconductors managed to gain 4.98%. Ouster, which makes lidar sensors, fell 16.08%. Teradyne dropped 11.49%.
The broader AI selloff following October 14’s inflation report likely contributed to Teradyne’s decline, given its semiconductor testing exposure. Higher borrowing costs can also weigh on valuations and customers’ automation budgets across the index.
Ouster’s August results showed an $18 million quarterly net loss despite 56% revenue growth. Its continuing losses may make its valuation especially sensitive to higher rates, although that does not establish why investors sold that week.
Owning a semiconductor company benefiting from AI infrastructure spending is not the same investment as owning a robot manufacturer trying to grow its installed base.
Both might belong in the physical AI ecosystem. Their economics can be completely different.
And right now, some parts of that ecosystem are having a much easier time making money than others.
The week in markets
One ecosystem. Very different trades.
Price returns · October 13 close to October 20 close
Black Scarab Physical AI 15
-4.77%
S&P 500
-1.19%
Nasdaq Composite
-2.87%
NXP Semiconductors
+4.98%
Ouster
-16.08%
Teradyne
-11.49%
Black Scarab calculations. Index returns exclude currency movements. Individual shares use their tracked listings.
The AI gold rush has some very clear winners
While investors debate whether AI is a bubble, the companies supplying the semiconductor industry continue reporting extraordinary numbers.
TSMC's latest earnings were spectacular. Revenue reached $46.72 billion, and management expects another $52.3 billion to $54.3 billion next quarter.
ASML wasn't exactly struggling either. It generated €11.7 billion in quarterly sales, with a 57.5% gross margin.
But one detail in ASML's report deserves more attention.
Demand for upgrades to equipment customers already own helped push profitability above management's expectations.
Think about that.
When a semiconductor factory is running near its limits, making an existing machine more productive can be incredibly valuable. Customers don't necessarily need a revolutionary new system. Sometimes they need the equipment they already paid for to work faster.
That's a lesson for the entire physical AI industry.
Solving an expensive bottleneck today can be a much better business than promising to revolutionize a factory tomorrow.
The rest of manufacturing hasn't received the same memo
The Federal Reserve's latest industrial production report painted a very different picture outside the hottest parts of the economy.
U.S. manufacturing output grew at an annualized rate of just 0.7% in the third quarter.
Semiconductor and related electronic-component production? Up 28.4% annualized.
Meanwhile, motor vehicle and parts production fell 3% in September alone.
Same economy. Completely different demand environments.
Inside U.S. manufacturing
The boom is concentrated.
Third quarter 2026 output growth · annualized
Semiconductors & related components28.4%
All manufacturing0.7%
75.9%
of U.S. manufacturing capacity was in use in September.
Source: Federal Reserve, October 16, 2026. Manufacturing includes the semiconductor industry.
Imagine you're selling a robotic assembly system.
Your first customer is a semiconductor manufacturer trying to squeeze every additional unit out of a production line. Your second is a traditional manufacturer with underused equipment and uncertain demand.
Who do you think is more eager to write a check?
That doesn't mean factories with spare capacity won't automate.
But the sales pitch changes.
Instead of promising more production, you need to demonstrate lower operating costs, fewer defects, reduced labor requirements, or faster changeovers.
And those benefits need to survive scrutiny from someone who cares a lot more about payback periods than how impressive your robot looks on YouTube.
Yaskawa's numbers reveal the uncomfortable part
Japanese automation giant Yaskawa offers a fascinating example.
Its October 9 earnings, released just before our reporting window, showed second-quarter companywide orders jumping 38% year over year.
Sounds excellent.
Its Motion Control business, which supplies equipment including motors and drives, reported first-half revenue growth of 21.7%.
But its Robotics division tells a different story.
Revenue slipped 0.7%, while operating margin fell from 8.8% to just 5%.
To be fair, Yaskawa had internal problems. A new ERP implementation disrupted production and shipments, while European restructuring costs weighed on profitability.
So this isn't proof that demand for industrial robots has collapsed.
It's evidence that strong industrial demand doesn't automatically translate into strong robotics profits.
The interesting question is how much of that margin pressure disappears once the operational problems are fixed.
Yaskawa also has something working in its favor: growing demand for automation, new production capacity, and its push into more advanced physical AI systems.
There's plenty of opportunity here.
But investors need to distinguish between a company developing promising technology and a company consistently earning attractive margins from selling it.
Those aren't always the same thing.
The robot is only part of the bill
This week, Siemens and NVIDIA announced something that could eventually matter enormously to industrial automation.
At Siemens' electronics factory in Erlangen, Germany, the companies are using AI and digital twins to prepare circuit-board inspection before physical prototypes even exist.
The project targets a reduction of up to 85% in certain manual engineering work involved in preparing inspection programs.
That's not an 85% reduction in total factory costs or robot installation expenses. It's a specific manufacturing application, with results that still need to be demonstrated at scale.
But the economics are interesting.
If AI can reduce the engineering work required to introduce new products, configure systems, and prepare production, manufacturers could spend less money getting automation into useful operation.
That matters because buying the robot is only one part of the investment.
Someone still needs to integrate it, program it, maintain it, and deal with everything that goes wrong along the way.
FANUC provided a rather expensive reminder of that reality.
The Japanese automation company announced the opening of a new Michigan training academy representing more than $18 million in investment. Its broader training network now has capacity for approximately 17,500 participants annually.
There's a certain irony in spending millions to train people to operate machines designed to reduce labor requirements.
But it makes perfect economic sense.
A robot that nobody can install or maintain isn't particularly useful, no matter how intelligent it is.
The next generation of industrial AI may create significant value not just by making robots smarter, but by making the entire process of deploying them cheaper.
Next week, Teradyne gets its turn
One company I'm particularly interested in watching is Teradyne.
Last quarter, its Robotics division generated $100 million in revenue, up 33.4% from a year earlier.
That's encouraging.
But put it next to the company's semiconductor testing business, which generated $1.122 billion in the same quarter.
Robotics represented only about 7.5% of Teradyne's total quarterly revenue.
The AI infrastructure boom is still doing most of the heavy lifting.
And there's another interesting connection: electronics manufacturing and semiconductor customers have become the largest end market for Teradyne's Robotics division.
The pattern keeps repeating.
The strongest near-term automation opportunities appear closely connected to industries already spending aggressively on AI infrastructure.
Teradyne reports third-quarter results on October 27, with its earnings call the following morning.
I'll be watching whether robotics revenue keeps growing, whether profitability improves, and whether demand broadens beyond electronics.
More importantly, I want to see evidence that growing robot sales can eventually translate into a scalable, profitable business.
That's the part shareholders actually get paid for.
The bottom line
The AI infrastructure boom is real. The financial results make that difficult to dispute.
But the physical AI investment opportunity is not one uniform trade.
Some companies are selling into industries where customers urgently need more output. Others are trying to convince factories that automation makes economic sense even when demand is uncertain.
Both groups could benefit enormously from the next decade of robotics adoption.
They just aren't operating on the same timetable.
For shareholders, the useful evidence will come from customers placing repeat orders and suppliers earning better margins as deployments grow. Teradyne’s results next week will offer one indication of whether the business is moving in that direction.
AI Is Printing Money. The Robot Makers Are Still Fighting for Their Cut.
In this edition +
Explore the Physical AI 15 ↗$46.7 billion.
That's how much TSMC generated in revenue last quarter. Sales jumped 41.2% from a year earlier, and its gross margin reached an almost absurd 66.4%.
Meanwhile, ASML, the Dutch company that makes some of the world's most advanced chipmaking equipment, says its planned production capacity for 2027 and 2028 won't be enough to meet projected customer demand.
Now here's the other side of the story.
American factories are running at just 75.9% of their production capacity.
One corner of manufacturing is desperate for more capacity. Much of the rest has capacity to spare.
And if you're investing in robotics or physical AI, understanding that difference might be more important than watching another humanoid robot do backflips.
The market isn't buying every AI story
The Black Scarab Physical AI 15 dropped 4.77% over the five trading sessions ending October 20, according to our internal calculations.
That's considerably worse than the S&P 500's 1.19% decline and the Nasdaq's 2.87% loss over the same period.
And underneath those numbers, things got interesting.
NXP Semiconductors managed to gain 4.98%. Ouster, which makes lidar sensors, fell 16.08%. Teradyne dropped 11.49%.
The broader AI selloff following October 14’s inflation report likely contributed to Teradyne’s decline, given its semiconductor testing exposure. Higher borrowing costs can also weigh on valuations and customers’ automation budgets across the index.
Ouster’s August results showed an $18 million quarterly net loss despite 56% revenue growth. Its continuing losses may make its valuation especially sensitive to higher rates, although that does not establish why investors sold that week.
Owning a semiconductor company benefiting from AI infrastructure spending is not the same investment as owning a robot manufacturer trying to grow its installed base.
Both might belong in the physical AI ecosystem. Their economics can be completely different.
And right now, some parts of that ecosystem are having a much easier time making money than others.
One ecosystem. Very different trades.
Price returns · October 13 close to October 20 close
The AI gold rush has some very clear winners
While investors debate whether AI is a bubble, the companies supplying the semiconductor industry continue reporting extraordinary numbers.
TSMC's latest earnings were spectacular. Revenue reached $46.72 billion, and management expects another $52.3 billion to $54.3 billion next quarter.
ASML wasn't exactly struggling either. It generated €11.7 billion in quarterly sales, with a 57.5% gross margin.
But one detail in ASML's report deserves more attention.
Demand for upgrades to equipment customers already own helped push profitability above management's expectations.
Think about that.
When a semiconductor factory is running near its limits, making an existing machine more productive can be incredibly valuable. Customers don't necessarily need a revolutionary new system. Sometimes they need the equipment they already paid for to work faster.
That's a lesson for the entire physical AI industry.
Solving an expensive bottleneck today can be a much better business than promising to revolutionize a factory tomorrow.
The rest of manufacturing hasn't received the same memo
The Federal Reserve's latest industrial production report painted a very different picture outside the hottest parts of the economy.
U.S. manufacturing output grew at an annualized rate of just 0.7% in the third quarter.
Semiconductor and related electronic-component production? Up 28.4% annualized.
Meanwhile, motor vehicle and parts production fell 3% in September alone.
Same economy. Completely different demand environments.
The boom is concentrated.
Third quarter 2026 output growth · annualized
of U.S. manufacturing capacity
was in use in September.
Imagine you're selling a robotic assembly system.
Your first customer is a semiconductor manufacturer trying to squeeze every additional unit out of a production line. Your second is a traditional manufacturer with underused equipment and uncertain demand.
Who do you think is more eager to write a check?
That doesn't mean factories with spare capacity won't automate.
But the sales pitch changes.
Instead of promising more production, you need to demonstrate lower operating costs, fewer defects, reduced labor requirements, or faster changeovers.
And those benefits need to survive scrutiny from someone who cares a lot more about payback periods than how impressive your robot looks on YouTube.
Yaskawa's numbers reveal the uncomfortable part
Japanese automation giant Yaskawa offers a fascinating example.
Its October 9 earnings, released just before our reporting window, showed second-quarter companywide orders jumping 38% year over year.
Sounds excellent.
Its Motion Control business, which supplies equipment including motors and drives, reported first-half revenue growth of 21.7%.
But its Robotics division tells a different story.
Revenue slipped 0.7%, while operating margin fell from 8.8% to just 5%.
To be fair, Yaskawa had internal problems. A new ERP implementation disrupted production and shipments, while European restructuring costs weighed on profitability.
So this isn't proof that demand for industrial robots has collapsed.
It's evidence that strong industrial demand doesn't automatically translate into strong robotics profits.
The interesting question is how much of that margin pressure disappears once the operational problems are fixed.
Yaskawa also has something working in its favor: growing demand for automation, new production capacity, and its push into more advanced physical AI systems.
There's plenty of opportunity here.
But investors need to distinguish between a company developing promising technology and a company consistently earning attractive margins from selling it.
Those aren't always the same thing.
The robot is only part of the bill
This week, Siemens and NVIDIA announced something that could eventually matter enormously to industrial automation.
At Siemens' electronics factory in Erlangen, Germany, the companies are using AI and digital twins to prepare circuit-board inspection before physical prototypes even exist.
The project targets a reduction of up to 85% in certain manual engineering work involved in preparing inspection programs.
That's not an 85% reduction in total factory costs or robot installation expenses. It's a specific manufacturing application, with results that still need to be demonstrated at scale.
But the economics are interesting.
If AI can reduce the engineering work required to introduce new products, configure systems, and prepare production, manufacturers could spend less money getting automation into useful operation.
That matters because buying the robot is only one part of the investment.
Someone still needs to integrate it, program it, maintain it, and deal with everything that goes wrong along the way.
FANUC provided a rather expensive reminder of that reality.
The Japanese automation company announced the opening of a new Michigan training academy representing more than $18 million in investment. Its broader training network now has capacity for approximately 17,500 participants annually.
There's a certain irony in spending millions to train people to operate machines designed to reduce labor requirements.
But it makes perfect economic sense.
A robot that nobody can install or maintain isn't particularly useful, no matter how intelligent it is.
The next generation of industrial AI may create significant value not just by making robots smarter, but by making the entire process of deploying them cheaper.
Next week, Teradyne gets its turn
One company I'm particularly interested in watching is Teradyne.
Last quarter, its Robotics division generated $100 million in revenue, up 33.4% from a year earlier.
That's encouraging.
But put it next to the company's semiconductor testing business, which generated $1.122 billion in the same quarter.
Robotics represented only about 7.5% of Teradyne's total quarterly revenue.
The AI infrastructure boom is still doing most of the heavy lifting.
And there's another interesting connection: electronics manufacturing and semiconductor customers have become the largest end market for Teradyne's Robotics division.
The pattern keeps repeating.
The strongest near-term automation opportunities appear closely connected to industries already spending aggressively on AI infrastructure.
Teradyne reports third-quarter results on October 27, with its earnings call the following morning.
I'll be watching whether robotics revenue keeps growing, whether profitability improves, and whether demand broadens beyond electronics.
More importantly, I want to see evidence that growing robot sales can eventually translate into a scalable, profitable business.
That's the part shareholders actually get paid for.
The bottom line
The AI infrastructure boom is real. The financial results make that difficult to dispute.
But the physical AI investment opportunity is not one uniform trade.
Some companies are selling into industries where customers urgently need more output. Others are trying to convince factories that automation makes economic sense even when demand is uncertain.
Both groups could benefit enormously from the next decade of robotics adoption.
They just aren't operating on the same timetable.
For shareholders, the useful evidence will come from customers placing repeat orders and suppliers earning better margins as deployments grow. Teradyne’s results next week will offer one indication of whether the business is moving in that direction.
Sources
About the author
Rodolfo Garcia Calderoni, CFAFounder, Black Scarab