Micron FQ3 25 & Nvidia – Party On!

AI picks and shovels continue to do very nicely.

• The Western AI industry has clearly put the DeepSeek wobble behind it with no sign of a fall in demand or cut in capex meaning that for now, the AI party continues in full swing.
• Micron reported excellent results, handily beating expectations and with more to come, there is no sign of a top in the notoriously cyclical memory industry.
• FQ3 25 revenues / Adj-EPS were $9.3bn / $1.91, nicely ahead of expectations of $8.9bn / $1.60, and the good news did not stop there.
• FQ43 25 revenues / Adj-EPS are expected to be $10.4bn – $11.0bn ($10.7bn) / $2.35 – $2.65 ($2.50), also ahead of expectations of $10bn / $2.12.
• At the same time, Jensen Huang put on a confident performance at the company’s AGM, adding confidence to shareholders’ view that the opportunity in AI and robotics will continue to rapidly expand.
• This put to bed any fears that the technical efficiencies that DeepSeek pioneered (and that everyone else has replicated) would cause a weakening in demand as a result of needing less compute.
• The alternative theory, which is one I generally agree with, is that when AI becomes cheaper in terms of tokens per dollar, more tokens are demanded, and the lower price ends up expanding the opportunity overall.
• This has been observed many times in the technology sector, most notably in mobile phones, where in the early days it was thought that the high cost of a mobile phone would limit its adoption to a few million units per year.
• The semiconductor industry has also benefited from this phenomenon on several occasions, and I don’t see why AI should be any different.
• Even without super-intelligent machines, of which I am highly sceptical, the new AI superpowers of data ingestion and understanding and the ability to converse with humans using natural language are more than enough to generate many highly lucrative use cases.
• Consequently, the outlook is that the cloud companies are going to continue spending like crazy, and the picks and shovels such as Nvidia, Micron, SK Hynix and so on will continue to benefit for a few quarters yet to come.
• Where I have concerns is with the likes of Anthropic, Mistral, OpenAI, Safe Superintelligence, and so on, who all have valuations that assume that they will invent AGI but who are also facing commoditisation and financial distress should they fail to do so.
• There are signs of this commoditisation everywhere as OpenAI and Google keep offering me more and more services for the same price.
• For example, when I first signed up for OpenAI’s Plus tier, I had 8 deep research enquiries per month.
• This quickly grew to 10 and now I have 25, which, for my use case, is plenty.
• Google’s Gemini Pro tier now offers me unlimited Deep Research queries and has taken to reminding me to use it every time I log in in a bid to ensure that I stay with Google and don’t go elsewhere.  
• This tells me that the differentiation between the different players is rapidly eroding, meaning that they have to keep offering more and more in order to prevent price erosion.
• At the same time, the cost of producing tokens is also falling very rapidly, but if the valuations are correct, this should be contributing to financial performance, not giving users like me much more bang for the buck.
• I still think that this ends in a correction and a shakeout as the weaker players fail to keep up and get absorbed by someone with poor in-house AI (e.g. Amazon and Apple) but with a fat bank account.
• However, in the meantime, money continues to pour into the sector from both companies and governments, and so the likes of OpenAI and so on will remain hot commodities.
• However, it is the picks and shovels where the real money will be made, and here the outlook for Nvidia, Micron, SKHynix, and so on remains pretty good.
• In the long run, I think that a lot of AI will be run on edge devices and here the picks and shovels are Qualcomm, MediaTek, Samsung and also Micron, which does much more than just datacentres.
• I continue to hold a position in Qualcomm as my exposure to AI at the edge.