One week / one topic: (Deep)Seeking answers

What happened?

Last week, DeepSeek took the world by storm by releasing a new leading edge AI model that is competitive vs OpenAI’s chatGPT and only costed $6m to train.

Understandably, investors were faced with key questions such as:

1. Wait: You guys (Big Tech) are spending hundreds of billions on AI CapEx, and this was achieved with only $6m? Oh my…

2. There are export restrictions in place vs China: how did they even get hold of the necessary Nvidia chips to do this?

ROI vs AGI concerns have ebbed and flowed for months now, and so the price reaction was very sharp on Monday.

That said, the moves also looked very thought-through as opposed to a wholesale, knee-jerk reaction… Adopters rose, providers fell.

Given the size of Mag Seven companies and their large impact even on a macro level, it’s worth breaking down the move for each one beyond what happened at the index level:

• Nvidia: Your whole “selling picks and shovels in a gold rush” positioning is in deep trouble here! Get me out!

• Alphabet: Not good, if you were planning to give away your AI ‘for free’ to protect/boost your ad-led business model.

• Tesla: Despite Elon’s purported technical miracles, this doesn’t sound too good… I think?

• Microsoft: Okay, you have been distancing yourselves from OpenAI but still… How much money have you given them so far already?

• Amazon: Who knows… You have been struggling to develop your own models and Jeff is busy elsewhere, but now maybe it will cost less to integrate AI in the business?

• Meta: Now we are talking… That old fox Zuck is gonna make money with AI no matter what!

• Apple: Yay, you already have the best hardware to deploy these new models!

(Information should not be considered a recommendation to purchase or sell any particular security.)

What are the wider implications then?

Is this a watershed moment in the all-consuming AI race, or just something to be expected when technology advances so quickly?

Our observations

• Fundamentals: Suddenly, Jevons’ paradox is all the rage. The implications – we will use even more AI than expected – look entirely plausible… Figuring out who’s going to make money however remains unclear.

• Price action: Treasuries rallied on the news, but it feels like there were already heading in that direction anyway…

• Investor beliefs: It would take much more than this to dislodge the cult of Nvidia… but for good reason perhaps, as this is not the Dot Com bubble. (Plus, t shirts)

So what?

While some investors would (secretly?) like to be able to ignore the AI theme and bring back the good old days, it is all simply too large, fast-moving and pervasive.

At the same time, from a macro perspective, equities remain ‘stuck’ between the floor provided by high (US) growth and the ceiling of high (US) rates. In other words, the delivery of earnings is key for continued appreciation given high valuations and strong competition from risk-free bond yields.

This is especially true for US equities, and indeed a key reason in support of our current equity positioning geared towards ex-US opportunities.

Not only the ‘US premium’ looks quite extended, but it is also largely predicated on where a company’s stock is listed… Everything else equal, having a higher percentage of US sales actually has the opposite effect!

(Valuation differentials shown as ‘US Premium’ below were regressed against other common risk factor for the largest 1,000 stocks globally.)

However, the fact remains that the automation of work has historically generated significant value and there is little reason to believe that it should be very different for cognitive work…

So – if it’s just a matter of ‘when’ – there is likely a lot of room for growth from here… and all the (investable) winners are US companies.

As long as AI keeps exceeding our expectations and performing genuinely (and upsetting) mind-blowing feats, we have no choice but to try to navigate these waters.

Mood music: Pat Benatar – Heartbreaker