We just recorded a podcast with @daniel_koss and @Mark_AAIG Daniel hosted the podcast. We talked about portfolio management, the AI trade, $NBIS the technical side of liquid cooling, and why counter-drone is so interesting, especially Will be live soon👌 Thanks for having us @daniel_koss

EOS was already 34% of my portfolio before this news. my biggest position. The whole defense sector has been down, but EOS has held up because management keeps executing. And investors focus on the wrong tailwind. I believe this company is going to re-rate hard as it goes from a weapons supplier into a provider of complete turnkey defense solutions for NATIONS. My first entry was at A$2.80. I presented it to our members in July 2025 at around A$4. My confidence in this company is sky-high. Europe’s push toward non-ITAR solutions and plans to spend €800 billion to €1 trillion on defense and specifically counter-drone. And this creates massive tailwinds for EOS. This latest contract, while still subject to conditions, reinforces exactly why I have such strong conviction. Counterdrone is THE defense trade $EOS.AX

The lithium trade! We talk business cases, geopolitical tailwinds and emerging opportunities Join us tonight on Discord for a LibertyStream presentation and live Q&A with fund manager @LucasSacerdote_ and our analyst @JochemAAIG . Lucas recently interviewed the CEO, and Jochem closely follows the company. Bring your questions! 🕗 8pm Amsterdam (CEST) · 7pm London (BST) · 2pm New York (EDT) · 11am Los Angeles (PDT) Exclusively for AAIG paid members. Access our Discord through our website or Substack. $LIB.V $VLTLF $SLI $ETL.V $EEMMF $LBNK.V $LBNKF

Chat, for the people following Penguin Solutions $PENG or $RXT closely small question👇 I’m researching into managed services for GPU clusters and would appreciate insights into their focus, what they manage for customers, and how they charge. Have they disclosed any pricing per GPU-hour, or enough contract detail to estimate it? Particularly interested in the management fee separately from hardware rental and power. Any relevant earnings-call comments, presentations or contract disclosures would be helpful. Thanks people👍

We’re hosting a members-only presentation and Q&A with fund manager @LucasSacerdote_ on our discord this Thursday, 8–9 p.m. Amsterdam time, to discuss our One to Watch report on LibertyStream and his recent deep dive with its CEO. Ahead of the session, we’ve made our One to Watch report free to read on Substack. The thesis? Turning oilfield wastewater into a source of lithium. Oil production brings vast quantities of salty water to the surface. That water contains lithiuma key ingredient in batteries. LibertyStream aims to extract it using modular plants connected to existing infrastructure. The company says this approach could reduce upfront investment and accelerate expansion across multiple sites. Oil producers already bring the water to the surface. Water companies can earn royalties on the lithium recovered. And the US could gain a domestic lithium source as it seeks to reduce reliance on foreign supplies. But can LibertyStream scale this into profitable production? @LucasSacerdote_ brings in-depth research into the companies his fund invests in and interviewed LibertyStream CEO Alex Wylie last week. Thursday is an opportunity to dig into the investment case, examine the execution challenges, and ask your questions. Read or watch ahead of the session: 📖 Free One to Watch report: 🎥 Lucas interviews CEO Alex Wylie: Want to join? Become an AAIG member: See you Thursday! $LIB.V $LAC $ALB $SLI

Okay, you want counter-drone stocks over drone makers, because it’s just simple economics and the importance of the solutions. Yet it seems hard to understand for a lot of people while it’s so simple. Defense against it is the no.1 priority now in the world in terms of defense spending. There are tons of drone makers that can make a drone as cheaply as possible and produce it in volume. This is the biggest issue in Ukraine, and innovation is super fast there. You want to focus on counter-drone solutions, as shooting down these drones costs way more than making them. So there’s an issue there, not only from a pure defense capability perspective, but also from a simple economic perspective. Countries want to be able to shoot down drones more accurately and way more cheaply than they currently do, because those cheap drones deplete countries’ financial reserves. That’s why you see all the drone makers going down: people think the defense trade is there. No, it’s in counter-drone. We were at Eurosatory in Paris this year, where every single company we spoke to said counter-drone was going to be the biggest spending focus for every country. We pitched to members in July last year at around A$4.50. It’s now at A$11.50. We pitched Smart Shooter in August because it focuses on the soldier level. So with EOS and SmartShooter so we cover CUA-S systems with on a multi layered country level (and mobile units and ships) and the soldier level with Smart Shooter. Imo that’s the whole spectrum for now where the big spending is. Smart Shooter recently won a three-year contract from the U.S. Joint Interagency Task Force 401, or JIATF-401, for its SMASH counter-drone systems, with a ceiling of up to $150 million in future orders. $EOS.AX $SMSH.TA

Finishing up the $NUAI thesis as we speak with @Mark_AAIG and @JochemAAIG We’ve been following and investing in the company since Q3 2025, and it was time to write up where we stand with the thesis today. There’s been plenty of speculation since the latest 8-K about the direct lease opportunity and what’s happening with the Stream Data Centers JV. we focus of course also on the risks as signing a tenant is the most crucial part of the thesis. We work through three scenarios, looking at what we think is most plausible for the equity split and how the development could be financed. Some questions remain open, and we’ll need official company announcements to answer them. Simple as that. The sector tailwind for powered land is strong. As you saw recently, OpenAI committed more than $518 billion through 2030 to reach its goals and that’s Antrhopic not even considering the hyperscalers, Nvidia, OpenAi etc. An interesting thesis, with plenty to unpack. Live this week on our platform

Good beat. Had revenue at 57b on the bull side on larger volume and blended asp mix. $MU Mostly looking forward to the call on productmix, architecture or system level solutions arround the Memory wall and HBM developments.

I see a lot of posts about power these days and how much we’re going to need, but much less about the optimisation everybody is working on. After our conversations with APR Technologies, ZutaCore, and Accelsius with different cooling solutions, I think the cooling efficiency of next-generation AI data centres will improve significantly. The ZutaCore/Munters comparison gives a sense of the potential: modelled annual cooling energy roughly 74% below its specified single-phase chilled-water baseline, with partial PUE falling from 1.115 to 1.03. Just to be clear, this is cooling-energy savings, not a 74% reduction in the facility’s total electricity consumption. But it would still be a huge step. The figures are modelled for a particular design and climate, with a reference-design proof of concept planned for Q1 2027. From our liquid-cooling sector research, we see how much time, energy and engineering effort is being committed to solving thermal issues. Accelsius reported B200 server benchmarks in which its two-phase system matched the single-phase baseline’s thermal performance with facility water at 54°C rather than 45°C. Warmer water gives operators more opportunity to reject heat without running energy-intensive chillers. Accelsius benchmark Part of APR’s approach is local coolant control. Management described directing coolant towards hotspots to reduce pumping energy and improve heat transfer at the chip interface. These designs target energy savings, thermal control and, depending on the system, opportunities to reduce cooling infrastructure and CAPEX. If you look at the route to market, there are interesting options for how these technologies could scale. In our call with APR, management said that even when large companies send their best engineers, they still need to teach them how to use and integrate the technology at a really deep level. Management stressed the need to get involved early in the hardware design, helping with simulations, coolant selection and materials. To me, that gives these niche players significant strategic value as partners to server manufacturers and chip designers. There’s a real specialist edge here that’s what we’ve gathered from multiple calls. So I see specialists teaming up with larger partners as a likely route, alongside potential M&A. Time to market is still something we’re researching. We have follow-up calls planned with these companies, but also with a few major players to look at it from their side. The combination of commercial scale and getting more compute out of the electricity a site can access is really interesting in these times. Lowering cooling overhead can leave more power available for running the servers. In my view, this sector is still very undercovered by analysts and people on X. $NVDA $NBIS $IREN $VRT

Of my three holdings in the neo-cloud space ($NBIS , $IREN and $WYFI ) Nebius is the clear strategic winner from SemiAnalysis’ new ClusterMAX 3.0. Which isn’t a suprise ofc. Nebius is proving it can serve both ends of the market: multi-hundred-megawatt hyperscaler contracts and smaller, shorter-term clusters for AI labs and startups. Its move into Platinum validates the full-stack strategy, strong infrastructure, orchestration and customer experience translating into premium pricing and higher revenue per MW. Build by a strong, experienced team. IREN’s result is clearly weaker, but it is a bit more nuanced on site and older DC infra imo. Most of the negative feedback relates to Prince George and Mackenzie, while the newer Childress and Sweetwater builds appear materially stronger. I had hoped to see clearer validation from Sweetwater but we have to wait when first cluster is online. IREN’s strategic strength remains its ability to build large-scale, competitively priced capacity. The next step is proving that its managed-cloud layer can catch up or remaining disciplined around bare metal plus where demand is already strong. For WhiteFiber, there is not enough recent evidence for a high-level judgment. ClusterMAX still relies largely on earlier testing. WhiteFiber is now signing new cloud contracts on an improved platform, so the next meaningful signal must come from actual customer experience. Can’t judge it properly with recent developments but the road they are taking is promising.

Saw this move coming. These are the fastest to bring online. Can be retrofit or modulair building. You can purchase existing factory load sites with a smaller MW grid connection much easier due to availability. Much more for sale. There has already been a lot of site inspections by players like Krambu, Duos technologies and much more. Private company Giga energy also sees huge growth in their modulair building. We talk with the CTO soon. $WYFI $DUOT $IREN

If you want to play the Nscale IPO, look at the JV partner $AKER.OL They still own a 22.7% ownership position in Nscale. Nscale scaling will be very bumpy though because they have a huge backlog, and they will carry very large losses before their infrastructure has reached scale. You also have some dilution risk. CEO of Aker is still on the board though. If the IPO is valued at a big higher valuation it gives immediately upside for Aker also. They online have 5% of $NVDA GPUs online tho so they will be very power hungry and aggresive.

A small extension on yesterday. I personally found the memory market to be one of the most difficult markets to research out of all the research I have done, because there are just so many moving parts. I have heard the same from members, subscribers and clients. I also see it daily on my X feed: 10 or 20 posts about Samsung expanding here, SK hynix moving to this technology, Micron moving to that, Samsung taking the lead on yield, then SK hynix having the yield lead, then Samsung not being qualified, and then substrates being in a huge shortage… But it is impossible to tie all of this together if you do not put in the work to map it from every angle and put it into a live system. For most private investors, depending on how much money you have exposed to memory, it is simply too big of a task. Ideally and this is what I built for our members and clients you use all the most important angles that actually move something, allowing you to cancel out the noise and focus on one overview. Otherwise, the entire memory trade, at a high level, is almost impossible to follow. $SKHY $MU $SNDK

Like I said 3 days ago. Before we talk about full AI ROI we need to learn the basics.✍️ (must read for the people that need to bridge all these tech post on models, GPUs, etc to human psychology and adoption) Yesterday at AI Infra Summit, I listened to a panel called “Operationalizing AI: Turning Data, GPUs, and Inference into Enterprise ROI.” It was moderated by Radhika Malik of Dell Technologies Capital, with Kaushik Shirhatti of NVIDIA, Boris Lukashev of WhiteFiber, Ruben Bryon of Verda and Alex Saroyan of Netris on the panel. A lot of things where so inline with my own thinking that i feel the industry is moving in the right direction. I was very clear on something a while back: you need to give people time to adopt AI and get used to the basic ways of using the tools. Otherwise, it becomes way too fast, and that essentially stops adoption. Thats not rocket science. Kaushik compared generative AI to a vending machine. You press a button and get an answer. An agent, on the other hand, is more like a new employee. It needs to be onboarded. It needs the most current information, access to the right tools and an introduction to the systems. It also needs boundaries, security, supervision and all of that. Kaushik also compared the complete team to a relay team so the compute, network and software may all be super good individually, but if you have four sprinters on your team, you need to practise the handoff. Its as simple as that. That is where a lot of performance, and therefore ROI, is won or lost for the enterprise and, to keep it even simpler, for people using AI. I’m going to show you a few angles here. Ruben focused on speed and adaptability. He compared a neocloud to a fighter jet and a conventional cloud more to an aircraft carrier. This is also so spot on because I’ve been talking about the agility that, for example, Nebius $NBIS provides and WhiteFiber $WYFI also. This panel showed again that you need to adapt fast, with physical deployment, software and operations at the same time, even when billions in capital have just been spent. A lot of people misinterpret the agility that is needed for this build-out. Ruben’s advice was to get people using the tools quickly, including in areas where the return is not yet completely clear. The cost of producing intelligence continues to fall while the models keep improving. I couldn’t agree more with that. This session made so much sense to me, based on how I was already thinking, that I had to share it with you. Boris, for example, said: “Preemptive optimization is the work of the devil.” You need to reach the required level of accuracy first and then optimize for performancenot earlier. So how much proof of ROI should we demand before allowing people to experiment? We can measure GPU utilization, inference speed, token volume, everything but those numbers don’t tell you anything about whether AI improves a decision imo. @tengyanAI talked about it this week as well. I personally see that it is a really fine line that you need to walk. If you ask too early, you are killing promising applications before people really understand what the technology can do. I think we are now at the point where we should start that.And if you leave experimentation unfocused for too long, you burn through a lot of resources without actually learning anything from feedback. The panel agreed, and I was already fully aligned with this. The right approach is to let people explore what becomes useful and then concentrate investments on three or four workflows that materially change businesses. I found this such a good panel discussion. Everybody should read it if it comes outor when it comes out. I’m not sure.👌👌 $IREN $DELL

Some personal Investing lesson on how I research and judge. My investing approach starts by mapping a market in its early stages, then working out what it could look like when it matures. From there, I reverse engineer what a company needs to do to succeed throughout that transition. For example on the AI buildout (subsector power) Early on, securing power and grid access is critical. As the market develops, constraints could increase the need for on-site generation, battery storage and software that coordinates power supply and demand to use available capacity efficiently. I then bring that back to the company level: what is it building today? Which capabilities is it developing? Do those investments position it for the next stage of the market? How does management align their strategy to my end state market vision. That shows me if it’s just a quick trade because they lack it. That’s one part on how I invest: i map the starting point, map the potential destination, and test the companies building the capabilities needed to get there. If you reverse engineer it all will become so much clearer for you all. Give it a try.

Everybody stressing out about the Dario comment and Sam Altman and Elon Musk publicly backing it, but it has zero to do with the AI buildout demand. This is from the model side and even if so it’s temporary. We still need a vast amount of compute. Nothing has changed there, but potentially the market will overreact tomorrow as always.