What caught our eye this week

Texas Instruments shared this long term chart on semiconductor cyclicality. Historically, slumps have ended around the 40 week mark, which would be May this year. 

New AI capable phones and computers could trigger an upgrade cycle, but inference on a phone would be power intensive, drain battery life, and make devices uncomfortably hot. Better to do the calculations in datacenters and stream the output. But at some point people will need new phones and PCs and perhaps it's finally time.Hyperscaler capex will soon hit $200 billionAnd growing. META sold off after saying they were going to increase capex more than expected. Which is good news for the receipients of that spend!

Microsoft's call this week: 

“We expect capital expenditures to increase materially on a sequential basis driven by cloud and AI infrastructure investments”

“Currently, near-term AI demand is a bit higher than our available capacity”

Google's call: 

“Our reported CapEx in the first quarter was $12 billion, once again driven overwhelmingly by investment in our technical infrastructure, with the largest component for servers followed by data centers”Google spent $12 billion on capex this quarter, twice as much as last year. 

Facebook's call:

“We anticipate our full year 2024 capital expenditures will be in the range of $35 billion to $40 billion, increased from our prior range of $30 billion to $37 billion as we continue to accelerate our infrastructure investments to support our AI road map”

Omdia research shared this chart of H100 estimates:

The biggest standout is META, as it doesn't have a cloud business like AWS, Azure or GCP, but has one of the largest horde's of GPUs. Well done Zuck.  From Jefferies 

"... training GPT-3 in MSFT’s US datacenters can directly evaporate 700k liters of clean freshwater. In Iowa, an OpenAI data center consumed 6% of a district’s water a month before the release of GPT-4"Elon Musk on GPUS

“We are, at this point, no longer training-constrained”

“We've installed ….. 35,000 H100 computers or GPUs. Roughly 35,000 H100S are active, and we expect that to be probably 85,000 or thereabouts by the end of this year in training, just for training”

So Tesla is going to more than double their historic purchase quantity of H100s by the end of the year.

They still need cars to turn around for the stock to be a buy, in my opinion. It's a shame for Tesla shareholders that Elon is putting much of his AI efforts outside the company. Richard Tornetta, who held just 9 shares and successfully sued Elon in Delaware, has done no favours to his fellow shareholders. 

Spotify reported solid results with revenue up 20%, but shrinking quarter-on-quarter.

Personally, the $60 billion valuation looks way out-of-whack for 20% topline growth and marginal profitability.On the plus side, I'm sure they could lift prices with minimal churn, which must be what holders are playing for here.Perhaps then you can map out a path to a $60 billion valuation in a few years, but to own it today you need to see $120 or $160 billion. A much tougher ask, however you cut it. 

Australian inflation came in hot

The market is now pricing in a small but positive chance of a rate rise as soon as August. 

Hope you have a good weekend

Michael