Thomas Schulz

Memory customers asked for 5x-6x more capacity rather than 2x expansion plans! That's why this cycle will take longer than people think. $MU #skhynix

$NVDA growth is still accelerating even at $100 B revenue per quarter! (from CFO) Trading at a below-market multiple while growing multiples faster than the market and even accelerating with Vera Rubin driving a strong cycle soon.

UBS expects DRAM prices to rise 32% QoQ in Q3 and 18% QoQ in Q4. Note that after prices rose ~95% QoQ in Q1 and ~55% QoQ in Q2 (Trendforce), another 32% increase is like a 97% increase vs Q4 and the 18% in Q4 is another 72% vs Q4. So it looks like the increases are slowing way down, but the absolute price gains are still enormous due to the power of compound interest. $MU #skhynix

Micron's guidance was even stronger than my high expectations. They guide to $50 B with 86% gross margin. After opex, that's $165 B annualized EBIT for 8.5x EBIT with EBIT almost certain to grow a good deal higher in 2027. The long-term agreements appear to have a floor price around 70% gross margin, which suggests a high EBIT margin for 5 years and they expect to earn significantly above that floor. That meaningfully reduces cyclicality and justifies a higher multiple. Crucially, $MU sees tightness beyond 2027 with no line of sight for supply catching up to demand even after that. Around 2030, they see a multi-decade memory demand cycle starting from humanoid robots. Starting in December, they will return 100% of excess FCF to shareholders. At ~5.6x 2027 EBIT, buybacks retire a lot of shares and importantly reduce overbuild risk, especially if SK Hynix and Samsung follow that discipline. $MU is no longer as crazy cheap, but it still looks highly attractive. The strong earning power looks pretty durable if one believes we are still early in the AI buildout like I do.

Roman Chernin on 20VC podcast: If they had 10x the capacity overnight, they could still sell it reasonably quickly, because "we certainly have the demand for it." It's very rare that a company can say this and it is likely true. That gives significant pricing power to the capacity they do have. $NBIS

The 25-30% QoQ DRAM price increase in C3Q26 would compound on top off ~+95% QoQ in Q1 and ~+55% QoQ in Q2. The growth rate gets smaller, but in an absolute sense the EBIT increase will be almost as big. Additionally, Bernstein sees HBM4 pricing at $53/GB in 2027, up from $16.6/GB. This will significantly increase 2027 EBIT after $NVDA locked in HBM-prices before the memory surge on an annual contract. Jensen said the "memory shortage will persist for quite a few more years." Elon sees a huge gap between too little memory capacity being added and future demand. -> Memory prices will rise further and multiples are still too cheap, because the shortage seems unlikely to end before 2030 IMO. $MU #skhynix

$DELL call had a very bullish read-through for memory and AI compute in general. “Demand continues to exceed supply, with memory as the primary constraint.” "We don't know how big the TAM is other than we know it's bigger, it's growing, and we're in the early innings of it." "The premium for computational capability across edge devices, PCs, smartphones, servers, and GPUs continues to grow at a rate we've never seen and is pulling the rest of the ecosystem." "I don't think applying historical models or historical views about the market and how it's going to act are appropriate today. What's the value of adding intelligence into every workflow, every decision, every product, every customer interaction? I would assert the value is pretty darn high. And that's what's been really, I think, the game changer since that October time is what's really happened in agentic. And what you're seeing are new categories of TAM expanding." $MU #skhynix $NVDA $NBIS

Leopold Aschenbrenner made $NBIS his by far largest holding. It's one of my biggest holdings too, because I really like the team and what they are building, AI compute demand really seems insatiable and will keep growing much bigger. Nebius is still cheap based on my longer-term projections, despite the large increase in the stock price, because the great news like the hyperscaler deals and large prepayments keep pushing my intrinsic value estimates higher.

UBS says memory long-term agreements have price floors only slightly below current prices. It's unclear whether the agreements allow for price upside. With 5-year contracts, Micron trading around ~5.6x near-term EBIT-runrate and SK Hynix at an even lower multiple, both still look very attractive. I expect AI demand to be so large that even the supply they bring online end of 2027 won't break the bottleneck. We may be transitioning to a structure where memory makers expand mostly against contracted demand 5 years out, making them less cyclical than in the past and deserving of higher multiples. $MU #skhynix

Morgan Stanley's newest EBIT-forecasts for SK Hynix: FY26: KRW 277 T FY27: KRW 408 T FY28: KRW 403 T Current Market Cap: KRW 866 T Hard to reconcile these earnings with this valuation. I'm long.

$NVDA China sales are back! Jensen also said they will use at least 50% of FCF for buybacks+dividend

$NVDA increased Blackwell + Rubin order visibility from $500 B (2025-2026) to $1 T (2025-2027). Importantly, this excludes Hopper (significant in 2025) and other products (e.g. Groq, Vera CPU racks, self-driving tech). Crucially, this doesn't mean $NVDA expects to earn "only" $500 B in 2027 (~7% above estimates). This is order visibility 9 months before the year even starts! Jensen explicitly said this pipeline keeps growing. The market seems to interpret this as revenue guidance for 2027. But order visibility doubled in just 4.5 months and they can clearly take more orders. It seems VERY likely that they will get more orders for 2027 and do significantly better than $500 B.