Artificial Intelligence – The Money Waterfall

Be at the bottom, not the top.

• The news of a gargantuan $100bn fund raise by OpenAI coupled with a 30% increase in HBM memory prices at Samsung, remind one very clearly that the place to be is at the bottom of the money waterfall not the top.
• OpenAI has been on the money raising trail again and it looks like this time it will raise $100bn at a pre-money valuation of $730bn, which sounds good now, but raises the bar for the valuation that will be demanded at the IPO.
• Amazon, SoftBank, Nvidia, Microsoft, and MGX are all thought to be participating, and between them, they should make up the bulk of the $100bn needed.
• For those who were fretting about the $100bn deal with Nvidia, I suspect that the mooted $20bn from Nvidia will make up the first tranche of this deal, leaving $80bn still to be filled down the line.
• This will be Amazon’s first dance with OpenAI, and the trade-off here is thought to be that OpenAI will give Amazon’s custom Trainium and Inferentia silicon chips a spot in its data centres.
• This is important for Amazon as it means that OpenAI will ensure that its models work well on Amazon silicon, which in turn means that developers using OpenAI will now be able to run their workloads within AWS.
• At the same time, from the other end of the supply chain comes the news from Chosun Ilbo (see here) in Korea that, despite making a mess of the 3^rd generation of HBM chips, Samsung has been able to increase the price for HBM4 by 30% to $700 per chip.
• Most people had assumed that this would price somewhere around $500 per chip, meaning that there is also a boost here for SK Hynix and Micron.
• This is yet another clear signal that Samsung has knuckled down, fixed its problems and come out of the gate with all guns blazing.
• The good news is that even though the HBM market goes from two players to three, there is so much demand that even with another supplier, there is still not enough to go around.
• Micron, for example, is already sold out for 2026 and is looking at taking firm orders for chips that will ship in 2027.
• Furthermore, the shortage is impacting supply for devices that require DRAM just at the time when the amount of DRAM that these devices is starting to trend upwards due to the requirement to run some AI services on device.
• Consequently, what we are looking at is the biggest upcycle in the notoriously cyclical memory market that we have seen for a generation, and there is clearly more to come.
• It takes time to build new fabs, and so the shortage is likely to continue for at least 2026 and potentially 2027.
• This means that the memory companies are going to hit or exceed their 2026 estimates, and on that basis, they still look very cheap.
• From an investment perspective, the bottom of the money waterfall is clearly the place to be, as this is where money is being made now and the crazier the investment rounds become, the more money these companies stand to make.  
• Furthermore, I continue to think that the business model of compute needs adjusting before it can make any money, and there are also real questions around how OpenAI, Anthropic et al will make money and earn a return worthy of their massive valuations.
• This is why I have very little desire to be positioned at the top of the waterfall.
• In the middle are the GPU suppliers Nvidia, AMD, Cerebras et al who earn the most revenues from the waterfall, but whose valuations are more reflective of how the AI bonanza will impact their bottom line.
• Here, newcomers like Qualcomm and MediaTek also look interesting as the market has not really appreciated what they may be able to achieve in the data centre, meaning that their valuations are also very reasonable.
• I have a position in Qualcomm and am looking closely at MediaTek, having solved its invisibility problem.
• I also have a position in Samsung, which I bought in early 2025 thinking that it would rerate once it had solved its HBM problem.
• This, it has done, but I greatly underestimated the short to medium term demand for memory, and as such, my numbers for Samsung remain way too low.
• Hence, I think the stock can continue to rise from here, and so I have no intention of selling it just yet.
• I continue to steer well clear of the top of the waterfall.