What a time to be live. The main thing happening today is Apple warning that rising memory costs will keep squeezing its profit margin. Live from 12 GMT. Here's your Order of Play plus an Op-Ed on why Apple aren’t buying data centres.

Apple has warned that its profit margin will keep falling as memory costs rise.

On its earnings call, the company projected slower services growth and a lower gross margin this quarter, sending shares down as much as 4% in after-hours trading. That followed a September quarter in which revenue grew 12% and gross margin reached 47.9%, both slightly ahead of Wall Street expectations.

CEO John Ternus and finance chief Kevan Parekh said Apple's stockpile of cheaper components is running down. It raised prices across the board last quarter to help cover higher costs. But demand for its $1,999 iPhone Duo foldable has outstripped supply despite mixed reviews, with shipping dates pushed back to January soon after launch.

Apple’s services business has problems of its own. Courts and regulators have limited what Apple can charge developers to distribute apps through the App Store, the segment's biggest component. OpenAI has also started directing potential ChatGPT subscribers to its website to pay, avoiding Apple's fees.

Ternus took over on 1 September, inheriting some expensive expectations. Oppenheimer analyst Martin Yang noted that Apple trades at 36 times next year's expected earnings, despite earnings growth forecasts of just 1%. He wants to see faster iPhone sales growth and a services recovery before becoming more bullish. “We see insufficient upside to justify the premium,” he wrote.

We're getting into all of it on today's show.

Microsoft, Alphabet, Amazon and Meta will spend $720B - $750B on the AI infra build out this year. Apple spent just $9B, down 29% on the year before, and returned $102B to shareholders through dividends and buybacks. Before, the market rewarded this as discipline, but Apple’s stock fell 4% after their earnings call on Monday so sentiment has changed. The stock is still trading at 36 times next year's earnings which is a premium compared to the rest of the market especially for 1% expected growth, but there’s a lot of worry that the growth engine might be slowing down.

Apple's strategy has been to rent not buy. It made sense for their on-device, privacy-first thesis and their premium multiple was built on the business's capital efficiency which probably wouldn’t survive tens of billions a year in GPU and infrastructure spend. So Apple has been renting a custom Gemini model to sit under Siri for $1B a year from Google. That was a good bet but now there are a few reasons why the market is pricing it negatively.

Firstly, Siri AI is considerably behind the agents it competes with (Instinct, Muse etc), because building frontier pace on rented compute and a licensed core model is pretty impossible. Renting is also hurting the margin. Ternus confirmed Siri AI will carry usage caps, with paid tiers following on soon. They have to ration the flagship feature because they don’t own the asset and so every query is a cost paid to a third party supplier. Thirdly, their App Store moat on tolls has started getting bypassed. OpenAI now steers ChatGPT subscribers through their website instead of the App Store. If the agent eventually becomes the interface, the party that owns the agent collects the toll.

Of course there’s the argument that the hyperscalers are overbuilding, compute will commoditise, and in three years Apple will rent frontier capability for a lot less. So far the compute owners are using their capacity to own the agent, and the agent is where intent is expressed and purchases are made. For twenty years Apple's growth came from controlling that moment but the low capex decision Cook made ceded this new version of it to the hyperscalers - mainly Google.

I think Ternus has two options. Either keep renting and accept that Apple becomes a premium distribution channel for other people's intelligence or do the opposite of Tim Cook and build real training and inference capacity. Ternus would basically need to sell shareholders that the next few years look more like Microsoft. I’m bullish on Microsoft of course ;)

See you on the show

— Luke