Enterprise AI Demand and Supply Chain Bottlenecks (with Jonathan Ferro, Lisa Abramowicz, Annmarie Hordern)

Heath Terry of Citigroup sees enterprise AI demand as incredibly robust with accelerating growth from 28% to 141% year-over-year. He argues hyperscalers are generating 29% returns on AI investment, making continued spending rational despite stock price underperformance. Key bottlenecks exist throughout the supply chain - memory (especially HBM), interconnect, storage, labor (both technical talent and construction workers), power, and semi-cap equipment.

Yields

explicit

RUT

Oil

Metals

USD

Citigroup Inc. 8.5
Investment Bank $1800.00B
Heath Terry 8.0

Enterprise AI Demand and Supply Chain Bottlenecks (with Jonathan Ferro, Lisa Abramowicz, Annmarie Hordern)

ndx

Enterprise demand is incredibly robust with growth accelerating from 28% to 141% year-over-year. Hyperscalers are generating 29% returns on AI investment.

Micron shares surging on blockbuster forecast. Heath Terry writes the health of the AI trade will be defined by irresistible enterprise demand meeting immovable inference capacity.

Jonathan Ferro
Heath Terry

Enterprise demand is incredibly robust. Growth has accelerated from 28% year-over-year in Q1 last year to 141% in Q1 this year, and we believe it will accelerate again in Q2. Anyone at a Fortune 500 pushing this into production sees it in real time.
Some companies like Micron with 85% margins benefit, but companies doing the spending are being punished. Does that change the trajectory for spending?

Jonathan Ferro
Heath Terry

Not anytime soon. Companies are getting real returns. Hyperscalers are generating 29% cash return on cash invested - very stable. If you can borrow at 6% and get 29% return, you're supposed to do that regardless of stock price.

Google issued $85 billion in equity and the market didn't blink. The gap between when you start spending and when you see returns is the issue, not the returns themselves.
Microsoft down 18% for worst monthly performance since 2008, Oracle down more than 30%. Are we seeing beginnings of investor pushback?

Lisa Abramowicz
Heath Terry

Microsoft and Oracle are software companies. The market is responding to the other 70% of Microsoft's business that is software, not Azure's acceleration. This should encourage more AI investment as Azure revenues accelerate through year-end.
Everyone's looking for choke points after memory. What's next?

Annmarie Hordern
Heath Terry

Choke points exist throughout the supply chain. Memory is the biggest - Anthropic chose Samsung, Micron, and SK Hynix as strategic partners. Next is interconnect, storage, labor (both technical talent and construction workers), power, and semi-cap equipment.

Anyone surprised by Micron today didn't pay attention to what Anthropic told them. We need more memory, more chips, more compute, more storage.
Are some of these supply chain companies undervalued right now?

Jonathan Ferro
Heath Terry

You want to be where the bottlenecks are - where people are spending money, not necessarily with the people spending it. It's not even undervalued, estimates are just too low. With irresistible demand and immovable supply, the only thing that breaks is price. 85% gross margins at Micron - memory companies aren't supposed to have that.

Eventually that will settle because cyclicals do, but we're a long way from that happening.
When you said labor is a choke point, are you talking about AI researchers or construction workers?

Annmarie Hordern
Heath Terry

Both. There's not much talent advancing these models - Meta was doing billion-dollar pay packages last summer. But I'm also referring to electricians, plumbers, construction workers who build data centers. Very few people have the skills to build at the level these data centers require, and they're in massive demand getting paid much more.

SignalTube

markets at a glance