Melvin @MelvinInvests

@KobeissiLetter x.com/melvininvests/… This is what’s happening!

Melvin @MelvinInvests

2 hours ago

The South Korea KOSPI just got halted limit down after falling over 8% and to understand why, you have to understand what this market had become (Save this). The KOSPI is no longer a diversified index but rather became a two stock AI trade. Samsung Electronics and SK Hynix,

3 10 20 6K 15

This is one of the most important charts in markets right now (Save this). It shows analyst price forecasts for key electronics components in January 2026 versus where those forecasts were revised by April 2026. The biggest revisions are concentrated in exactly the components that go into AI servers. DRAM was forecast to rise roughly 150% in 2026 and that was revised up to 250–280%, NAND was forecast to rise around 100%, revised to 200–250%. CCL, the copper laminate that printed circuit boards are made from was forecast to rise just 5–10% but it was revised all the way up to 35–40%. ABF substrates were forecast to rise 20–30%, revised to 30–35%, with supply conditions flipping from tight to very tight running into late 2026 and early 2027. And MLCCs, the tiny capacitors that manage power in every circuit were forecast to be essentially flat and the April update revised that to 0–5%, with supply tightening noticeably through the back half of 2026. Memory is still the headline, DRAM and NAND forecasts getting revised that dramatically is extraordinary on its own but the more important signal from this chart is what is happening below memory. The shortage is now spreading into ABF substrates, CCL laminates, and MLCCs, the physical components needed to actually build AI servers at scale. These are not glamorous components but you cannot build an AI data center without them, supply takes 18 to 24 months to meaningfully expand, and the demand shock from the AI buildout arrived in months. Now here are the stocks that benefit directly. Memory (DRAM + NAND): Micron Technology (MU) is the most direct US listed beneficiary of the DRAM and NAND surge, with AI HBM memory sold out through 2026 under long-term contracts. SK Hynix (000660.KRX) is the leading HBM supplier to Nvidia and benefits from the same DRAM pricing tailwind, with arguably even more HBM exposure per dollar of revenue. Samsung Electronics (005930.KRX) is the largest DRAM and NAND manufacturer on earth and captures pricing upside across both memory categories. ABF Substrates: Ibiden (4062.T) is the world's dominant ABF substrate maker for AI GPUs. Bernstein and JPMorgan have both upgraded it to Outperform, and it just committed ¥500 billion to expand capacity over three years. Unimicron (3037.TW) and Kinsus (3189.TW) are aggressively expanding ABF substrate production and are direct beneficiaries of the shortage running into 2027 and 2028. Ajinomoto (2802.T) controls over 90% of global ABF film supply, the raw material substrate makers depend on and just raised prices 30% effective Q3 2026. CCL Copper Laminates: Shengyi Technology (600183.SS) is one of the largest CCL producers globally and benefits directly from prices hitting all-time highs. Rogers Corporation (ROG) makes high-frequency laminates used in advanced AI server PCBs and is the most accessible US-listed CCL beneficiary. MLCCs: Murata Manufacturing (6981.T) is the dominant global MLCC maker with roughly 40% market share. Morgan Stanley just raised their price target from ¥5,100 to ¥12,500, calling it their top Japanese tech pick.kucoin+1 Samsung Electro-Mechanics (009150.KRX) is the second largest MLCC maker and is running near full utilization, with analysts describing it as the most mispriced opportunity in the passive components space. Taiyo Yuden (6976.T) is the third major MLCC player and has moved sharply higher alongside tightening supply conditions.kucoin The common thread across all of these is the same. They make the physical components that you cannot build an AI data center without, supply takes years to meaningfully expand, and the demand shock is already here. Make sure to follow me @MelvinInvests for more AI, semis, and underrated market opportunities.

I’m an analyst at Milk Road, and my job is to find underrated gems before the market catches on. We called names like MU, CRDO, NBIS, and BE over the last 3 months. Join me and my team for just $1. milkroad.com/pro/?utm_mediu…

@quantum_trader of course!

@MilkRoadAI you're welcome

I’m an analyst at Milk Road, and my job is to find underrated gems before the market catches on. We called names like MU, CRDO, NBIS, and BE over the last 3 months. Join me and my team for just $1. milkroad.com/pro/?utm_mediu…

Nebius will become a trillion dollar Neo-Cloud (Save this). The CEO of Nebius just laid out the playbook behind one of the most explosive growth stories in the entire AI infrastructure space and most people still haven't connected the dots. Nebius is not just selling compute but rather building an ascending product stack where each layer up unlocks exponentially more customers. At the bottom you have bare metal, raw GPU clusters, sold to a literal dozen companies on earth, the Metas and Microsofts who bring their own software and just need physical infrastructure at scale. Move up to managed infrastructure and the customer base expands from a dozen to hundreds, mid size AI companies, enterprises, research institutions. Move up again to inference and it jumps to thousands, every startup and developer building an AI product, running an LLM, or calling an API. At the top sits the agentic layer, AI agents running autonomously, continuously and the CEO said tens of thousands of new developers who don't even exist as customers today will live here. Nebius is building a funnel that starts at the most capital-intensive layer and expands outward as the entire ecosystem grows. The financial results validate exactly what he is saying. Nebius AI cloud revenue hit $389.7 million in Q1 2026, an 841% year over year increase from just $50.9 million in Q1 2025. The company exited 2025 with an annualized run rate of $1.25 billion, a 14x increase over the prior year and now projects exiting 2026 with an ARR of $7–9 billion. That implies 540% growth in run rate revenue in a single calendar year. What separates Nebius from a speculative bet is that most of this revenue is already contracted and locked in before the capacity is even built. The company has signed over $46 billion in long-term agreements including a five-year deal with Meta worth up to $27 billion, and a $19.4 billion agreement with Microsoft. The CEO made a point that building at the scale Meta and Microsoft demand is not a commodity. But these companies need gigawatt class power, custom cooling, ultra-low latency networking and the ability to deploy Nvidia's latest GPU generations within weeks of availability. Nebius has contracted over 4 gigawatts of power capacity by end of 2026, with AI factory campuses under construction in Missouri and Pennsylvania. The barriers to executing at that scale, utility relationships, construction timelines, GPU allocation priority, take years to build and cannot be bought overnight. Long Nebius and follow @MelvinInvests for more underrated gems.

@Ferbin08 Compute becomes a commodity only if anyone can supply it. Right now they can’t.

@BitWhitman They will and beyond especially as they build infrastructure around the world.

@LongEquities @MilkRoadAI It certainly is but I also own CoreWeave as well.

@MoMoMacro They’re building an AI native cloud from bare metal all the way up to inference and agentswhile securing power and capacity years in advance. I see a vertically integrated AI infrastructure platform with a much larger long term addressable market.

@BitWhitman From the same place every hyperscaler does, utilities and power developers. Nebius has already contracted power capacity and is building AI factory campuses around that infrastructure.

Because the market keeps assuming they’ll fund 100% of it from the balance sheet and that’s not how hyperscale infrastructure gets built. The longterm contracts are the financing. When you have billions in committed demand from customers like Meta and Microsoft, lenders are far more willing to provide project financing.

x.com/i/article/2070…

@JackSom28304086 lmfaooooo

@BitcoinJesusETH

@MilkRoadAI We def going somewhere!