Over the past few weeks I’ve been walking you through the biggest ideas in this market one at a time:

The robotics revolution…

The supply chain feeding it…

And the companies making the components underneath it all.

This week, companies at both ends of that story reported earnings, giving us a glimpse of what the next few years could look like.

Both delivered results that reinforced everything we’ve been talking about.  Yet the market sold them.

Why?

Because the market doesn’t look ahead far enough.

In one case, fear drove the sell-off.

The market persists with the view of an AI bubble.

Yet every avenue of the supply chain says otherwise.

Demand for semiconductors, particularly memory, continues to outstrip supply, and that’s likely to remain the case for a long time.

If you want to argue the AI boom is getting even bigger, there’s certainly a case to be made. I’ll have more to say on that on Wednesday.

In the meantime, it might be worth paying attention to what the companies actually building the AI infrastructure are saying.

When a carmaker builds its own chip fab

Let’s start with Tesla, because Elon Musk spoke about the most important technology device that we’ll see this generation . In my view this will be as impactful, if not more, than the smartphone revolution the iPhone kicked off.

Musk told analysts that Terafab, Tesla’s planned semiconductor fabrication plant, is a necessary project because, without it, “we simply won’t have enough AI chips” to scale production of Optimus. He pointed to three key constraints: memory, logic, and advanced packaging.

Think about what that means.

The world’s most valuable carmaker is spending billions to build its own chip fabrication plant because it doesn’t believe enough chips will exist to build its robots.

And the wider supply chain is behaving exactly as you’d expect if that were true.

Samsung is building capacity in Texas and TSMC in Arizona, with tens of billions aimed squarely at compute for Optimus and Robotaxi.

Musk even went out of his way to publicly thank Micron for granting Tesla a memory allocation , a courtesy you only extend when memory has become so scarce that an allocation counts as a favour.

Then came the production ambitions. They may sound extraordinary, but consider what they would mean if Tesla comes anywhere close to achieving them.

Optimus 3 carries an aspiration of a million units a year.

Optimus 4, to be built in Austin, aims for 10 million. And remember this is just one of many humanoid robot companies.

Musk was upfront that these are aspirations, not forecasts, and that scaling production of a robot built almost entirely from new components will be brutally difficult.

But even a fraction of those numbers changes the economics for the whole semiconductor industry.

Each robot needs inference compute, memory, sensors, batteries, and power.

Last week I showed you the picks and shovels of the robotics thematic, from Nvidia’s Thor platform to the memory demand heading Micron or SK Hynix’s way.

Physical AI is a whole new layer of demand arriving on top of the data centre build-out.

Speaking of car manufacturers…

When Henry Ford began mass-producing cars, what industry unexpectedly became one of the biggest winners?
Oil
Rubber
Glass
Road construction


Find out what your friends and family think as well. Share this email with them so they can answer today’s poll and subscribe as well.

Google can’t build fast enough, and the memory makers agree

Which brings me to Alphabet’s numbers from the same evening.

While Elon Musk was outlining Tesla’s ambitions, Sundar Pichai was doing much the same at Alphabet.

Revenue up 24% year on the year to US$119.8 billion. Google Cloud up 82% to US$24.8 billion. Operating income up 30%, with group margins expanding from 32% to 34%, and Cloud margins leaping from around 21% to more than 35 %.

Then came the number the market really hated: capital expenditure.

Quarterly capex more than doubled to US$44.9 billion and free cash flow swung negative, while full-year capex guidance was lifted to US$205 billion with a warning that 2027 will be higher again.

The stock fell around 5% after hours.

So the sellers are effectively arguing that a company growing its most important division at 82%, with expanding margins, should spend less on the thing driving that growth.

Good one.

Meanwhile Google’s own CFO told analysts the company remains supply constrained, with demand so strong that Alphabet will rent third-party cloud capacity  as a bridge while it builds its own infrastructure.

And if you want the view from the coalface, listen to SK Hynix.

CEO Kwak Noh-jung told Reuters this month that 2027 is shaping up to be “the worst year in the industry’s history ” for supply, and that memory demand will outstrip supply beyond 2030.

Analysts at Meritz reckon DRAM makers can meet only 75% to 80% of current demand, and that could fall to 60% by 2027.

That hunger for capacity is why rumours surfaced this week that SK Hynix could take over Intel’s long-delayed Ohio semiconductor fabrication site.

 The shares jumped 9% on the rumour before the company formally denied it, though it’s suggested the two are still exploring an operational partnership for the site.

Either way, a company that has just raised US$26.5 billion through a Nasdaq listing is chasing every possible avenue for capacity and funds to build because it cannot make memory fast enough for the demand in front of it.

The market thinks this level of capex is reckless.

I think it’s exactly what’s needed. Thinking this investment won’t generate returns many times greater than the initial expenditure is what strikes me as reckless.

I think the sellers are wrong on every level.

Every party in this supply chain is telling us the same thing from a different perspective.

The buyers of compute say they can’t get enough.

The builder of robots says the chips don’t exist yet and won’t be enough.

And the makers of memory say they’ll be short into the next decade and can’t make enough.

Heavy capex is a rational response to a shortage that the market can’t see lasting a decade.

These aren’t the only two by the way. Amazon, Meta, Microsoft, Nvidia… they’re all spending aggressively to build this all out. You can’t tell me they’re flying blind here.

They’re all looking ahead, not one year, not two, not even five. They’re planning for the next 10 to 20 years, investing today to ensure they remain relevant, market-leading, and profitable for decades to come.

Yet the market keeps treating every capex headline as a reason to sell the exact companies building out the biggest technology revolution in history.

Here’s the thing: Every one of those sell-offs is a chance to build positions in the compute, memory, component, and infrastructure names we’ve been covering, at better prices, for a build-out that will run for a decade or more.

I can’t remember a more exciting time to be an investor.

The scale of what’s coming is bigger than the market understands, and that’s an opportunity for you.

Until next time,


Sam Volkering
Investment Director, Southbank Investment Research

PS If this week’s earnings proved anything, it’s that the companies building AI are planning for the next decade, not the next quarter. That’s exactly why James Altucher, Editor of Altucher’s Investment Network, has been digging beyond the obvious winners. History suggests some of the biggest returns from Elon’s projects haven’t come from Tesla itself, but from the suppliers making it all possible. He’s identified the company he believes could be next. You can see his full research here.