Mark | AAIG Research

We’ve got a gem for you tomorrow 👀 With data center delays putting delivery timelines under pressure, we’re dropping a free deep dive into Nebius’s $NBIS Birmingham site ahead of Thursday’s earnings. @Mark_AAIG broke down Nebius’s latest video to assess construction progress, what still needs to happen, and where the challenges lie. Can they deliver on time? We’ve previously published two WhiteFiber $WYFI site breakdowns, and management gave us credit for the accuracy of our timing assessments. Both are available on our Substack if you want to see the work behind those calls. Also going live tomorrow: my WhiteFiber earnings breakdown and how I view the steps they are making and what it means for 2027 and beyond. And today’s Macro Pulse is already live, covering the AI sell-off, the midterms and the financing behind the AI buildout. It is something I rarely talk about but is very well written up by @JochemAAIG and keeps you up to speed with the market. You can subscribe to AAIG Research on Substack for free. Don’t miss tomorrow’s Nebius drop 👇 Link is in my bio!

1/ Everyone take a deep breath. Texas data centers aren't going anywhere. The Texas governor has very limited authority to stop them. Calls for a pause or moratorium must be approved by TCEQ—the permitting authority of Abbott appointees. Here's how Texas permitting really works👇

The newest AAIG Technical Series is now live. This time I went through what it actually takes to design and build a data center campus like the $NUAI TCDC site. • the permit roadmap, and which permits have been cleared • the equipment involved in the build, and its lead times • is it realistic for the site to be powered by Q4 2027? Link in the comments 👇 $IREN $NBIS

New interview dropping on Substack today 45 minutes before market open! Freshly recorded 1h ago. I sat down with @MarkosAAIG & @Mark_AAIG from AAIG to discuss AI infrastructure investing, portfolio strategies, memory stocks, next-gen data center cooling, and defense.

The Trump administration plans to offer a roughly $4-billion loan package to Vistra Corp. to upgrade three of its nuclear plants, according to people familiar the matter - Bloomberg

A note to the market: The market and ignorance go hand in hand again. I just listened to Micron’s earnings call, and a big part of the market is still underestimating how durable Micron’s, SK hynix’s and Samsung’s earnings could become. We got 80%, 87% adjusted gross margins, and next quarter, 86.25% gross margin. And this is where the focus will be. I’m going to pull you away from the numbers because the underlying story is much more important. There’s a bandwidth problem, and there’s a supply problem. Then we’ve got demand, which has been skyrocketing over the past one and a half years due to the data-centre buildout and server-level memory components. That’s the baseline. That buildout will continue for years. On the constraint side for data centres, we also have the power shortage, which needs to be solved. So that’s a double-up story that extends the timeline. At the same time, we’ve got the tailwind of broader demand coming in from different subsectors, like physical AI. Mehrotra pointed out that self-driving cars typically need over 200 GB of memory and multiple terabytes of storage. Humanoids, about the same. But those are the headline topics that still make people doubt: will we have fully self-driving cars in three or four years? Is big humanoid adoption possible in three or four years? That misses the point, in my opinion. A while ago, I wrote an article about ambient AI. That’s what I call it: AI that is around you everywhere. NVIDIA worked with NoTraffic, using local Jetson computers connected to a cloud-based system. It needs some low-power DRAM, such as LPDDR5X, to deliver that bandwidth performance within power and heat limits. To give you an example, and you probably want to read it again.. "On average, NVIDIA’s NoTraffic helps save over 12 tons of carbon dioxide and generates $45 million in economic productivity per year for a typical city of 100,000 residents. And 99% of the world’s traffic signals run on fixed timing plans, which leads to unnecessary traffic jams and delays" These small, real-world examples don’t make the headlines. Of course, humanoids and autonomous driving are interesting. But look at NoTraffic. With over 300,000 signalised intersections in the US, the company estimates that optimising them with this platform could result in a total of $14 billion in economic savings per year. And it needs memory. Again, this is just a small example that you can observe in the physical world around you, which is going to change what happens in it. Visualise it. Use my article below to get a little bit of a feeling for it, but look past the headlines. For the memory players, this broadens the potential demand base across their whole memory portfolio. There is so much volume and memory needed per use case. It’s just hard to wrap your head around it. This quarter, some clients extended their SCAs by one year into 2031. That already gives you a five-year baseline revenue view on the 26 strategic customer agreements expected to represent over 35% of revenue through 2030. And this is just the beginning. It’s not even the end, in my opinion. The majority will be data-centre-heavy for the upcoming years, as this is the core of the AI buildout needed to run these kinds of optimisations, like NoTraffic shows. We need compute for that. The CFO said that even at floor prices, we expect margins meaningfully above any prior cycle peak margins. That says everything about the economics you can expect in the upcoming years. 2027 and 2028 will be tighter than 2026. Yeah, I know. 2029 will be also, and 2030 probably also. And then we haven’t even talked about the underlying ecosystem that struggles so much to keep up: substrate materials, where to go with stacking and bonding, thermal issues. all these things that strengthen the strategic position of memory and the co-design depth needed to produce the best products. I’ve been a memory bull for a long while. I started accumulating Micron heavily in 2024, SK and Samsung in the fall of 2025 and the thesis is even bigger than my perception back then. Bullish memory. $MU $SNDK $NVDA $SKHY

Fascinating. Chief Economist at Apollo: agents could cause a bank run by sweeping household cash into accounts paying 3-5% instead of the 0.1% national average, causing banks to lose a large share of their cheap deposits.