Register and share your invite link to earn from video plays and referrals.

Melvin
@MelvinInvests
AI Analyst @MilkRoadAI | Finding opportunities across AI, photonics, defense, space, and tech.
123 Following    18.5K Followers
Morgan Stanley mapped out the entire AI infrastructure supply chain and it reveals who actually gets paid at every layer of the trillion dollar buildout (Save this). This heatmap breaks the AI infrastructure value chain into two dimensions those who owns and operates the data centers at the top and what physical and technical components get built underneath to make those data centers function. At the top sit the owners/operators, the hyperscalers like Meta, Alphabet, Amazon and Microsoft, alongside data center REITs, private equity giants like Blackstone and Brookfield, enterprises and neoclouds including CoreWeave and Nebius. These are the companies writing the massive capex checks that fund everything below them. Below that sits the actual build out, split into seven layers, semi production, processors, server components, servers, network, internal power/cooling and power supply. Semiconductor production is dominated by names your audience already knows well, Nvidia and AMD for GPUs, TSMC adjacent foundries, ASML and Applied Materials for capital equipment, and Micron and SK Hynix under memory/storage. But the less obvious money is in the physical infrastructure layers most retail investors never look at. Server components include passive parts from Yageo and Murata, thermal solutions from Sanyo Denki, and PCB substrates from companies like Unimicron. Network infrastructure includes InfiniBand and Ethernet gear from Nvidia and Arista, plus optical/DCI routing from Cisco and Ciena. Internal power and cooling is arguably the most underappreciated category here. It includes liquid cooling specialists like Vertiv and CoolIT, power electronics from Siemens and Eaton, and uninterruptible power supply makers like ABB and Legrand, all companies solving the literal heat and electricity problem created by cramming more GPUs into less space. So who benefits from all of this? Everyone in every box benefits in some way but the real insight is that value doesn't concentrate at just the GPU layer anymore. The hyperscalers at the top are distributing capex across seven distinct physical layers which means the picks and shovels opportunity set has expanded well beyond Nvidia into cooling, grid infrastructure, and power generation. Bullish on AI infrastructure, make sure to follow @MelvinInvests for more AI infrastructure insights, and if you want to see exactly what I'm buying as an analyst at Milk Road Pro, you can check out the link below for more.
Show more
0
82
2.8K
577
Forward to community
Marvell wants to almost triple its revenue in three years and its plan to get there might make today's stock price look like a steal (Save this). Marvell's revenue climbed from $2.3 billion in FY17 to $8.7 billion in the trailing twelve months. Analysts now project it will nearly triple again to $11.5 billion by FY27, $16.7 billion by FY28, and $23.3 billion by FY29. That growth curve has already shown up in results. Q1 FY27 revenue hit a record $2.42 billion, with data center revenue growing 38% year over year and management raising its forecast for next year's data center growth above prior expectations. Marvell's growth plan rests on straddling both halves of the AI infrastructure buildout rather than picking one lane. On the compute side, Marvell designs custom AI silicon for hyperscalers, currently working across more than 50 custom design opportunities spanning over 10 customers, with three nanometer wafer capacity already locked in to support that pipeline. This includes chips built specifically for companies like Amazon, Microsoft, and Google that want AI accelerators tailored to their own workloads instead of buying off the shelf GPUs. On the data movement side, Marvell sells the switching, electrooptics and high speed interconnect silicon that moves data between those chips and across data center racks. Marvell projects the critical interconnect market alone will hit $14 billion by 2028 at a 27% annual growth rate. Layered on top of that is the CXL memory expansion business, where Marvell's Structera controllers and switches sit at the center of a market Morgan Stanley now projects will more than double and triple by 2030. Data center revenue already makes up roughly 76% of Marvell's total sales. That means the plan to reach $23 billion in revenue is really a bet that AI infrastructure spending keeps compounding across compute, connectivity, and memory all at once. Marvell trades at a forward P/E of roughly 42 and a market cap around $166 billion. That sounds rich until you compare it to the growth rate behind it, since revenue is projected to nearly triple in three years, meaning the multiple compresses fast if the company executes even close to plan. With that being said, customer concentration risk is real since Marvell depends heavily on a handful of hyperscaler relationships and the stock has shown it can swing 10% in a day on competitive headlines, like the ByteDance in-house ASIC scare in June. But with data center demand still accelerating, a tripling revenue base already baked into consensus estimates,and a valuation that hasn't fully caught up to that growth path, the setup favors Marvell closing that gap rather than the growth story stalling out. I remain extremely bullish on Marvell, follow me @MelvinInvests for more infrastructure plays and check out the link below for more!
Show more
The scariest chart on Wall Street right now is actually one of the best buying signals in years (Save this). The Morgan Stanley Tech Momentum Index just hit a 17 day rate of change of -35.9%, the worst reading in the index's entire 27 year history. The Goldman Sachs High-Beta Momentum Index is down -24% month-to date, the worst performance since April 2009. Before you panic, you need to understand what these numbers actually measure because this is not the AI buildout collapsing but rather a momentum factor unwind and those are two very different things. A momentum strategy is simple, buy whatever has gone up the most, short whatever has gone down the most and in H1 2026, that strategy returned a historic 57% because AI stocks went nearly straight up. When a trade gets that crowded, Goldman tracked momentum positioning at the 100th percentile of the last five years, it gets fragile. One trigger causes every fund running the same playbook to sell at the same time and you get a cascade that has nothing to do with the actual businesses underneath. The triggers here were textbook, low holiday week liquidity, end of quarter rebalancing, profit taking after a record first half and a widely misread headline about Meta and data center capacity. The fundamental view remains positive and while JPMorgan said buy the dip and UBS called it an orderly de-risking exercise, not a forced liquidation. The data that actually matters hasn't moved. Hyperscalers are still guiding to $1.4 trillion in capex by 2028, a number Morgan Stanley just raised 9-10% for 2027 and 2028. Micron's entire HBM4 supply for 2026 is already sold out under long-term contracts, Nebius has $12 billion locked in from Meta and $17.4 billion from Microsoft. None of those contracts changed because a momentum index printed -35%. Goldman's own historical data shows that momentum selloffs of this magnitude since 2006 were followed by average gains of 1.45% the following week and nearly 23% over the next year. The worst momentum reading in 27 years sounds terrifying but what it actually means is that the trade got too crowded after a historic run and now it's flushing out the weakest hands. The only question left is simple, are you buying from the weak hands or are you one of them? The biggest opportunities are often created by the biggest overreactions and make sure to follow me @MelvinInvests for more.
Show more
0
41
778
120
Forward to community
Apple wants to buy cheap Chinese memory chips and the everyone thinks it's bad news for Micron but it's not (Save this). CXMT is China's largest DRAM manufacturer, headquartered in Hefei and it was essentially irrelevant three years ago but today it holds 8% of the global DRAM market, up from just 3% a year ago. The company makes commodity DRAM, DDR4, DDR5, and LPDDR chips primarily for phones, PCs, and consumer devices. It is heavily subsidized by the Chinese government, operates at costs well below market rates, and has been placed on the Pentagon's Chinese Military Company blacklist due to alleged ties to the People's Liberation Army. Apple raised MacBook Pro prices by 15% this week and is now desperate to find cheaper memory. Tim Cook told the Wall Street Journal that everything needs to be on the table when it comes to memory sourcing, and Apple is lobbying the Trump administration for a license to buy CXMT chips technically possible, but complicated by the Pentagon blacklist. This spooked some memory investors but it shouldn't have. Here are the five reasons why this barely moves the needle for Micron. First, Micron essentially already walked away from this market. Samsung, SK Hynix, and Micron have all shifted more than 70% of their DRAM capacity toward High Bandwidth Memory (HBM) for AI, the premium product running inside every Nvidia GPU and every major AI data center. CXMT is filling a hole that Micron chose to vacate because HBM generates 3–5x more revenue per wafer. Second, CXMT cannot make HBM, its most ambitious stated goal is to begin certifying HBM3 for mass production by end of year, a full generation behind where SK Hynix and Micron already are in HBM3E production today. The AI memory market, which is the entire bull thesis, is simply not accessible to CXMT at any meaningful scale. Third, the dollar opportunity is in a completely different product. The $280–530 billion memory opportunity Micron is selling into over the next two years is almost entirely HBM for data centers, automotive systems, and humanoid robots, none of which CXMT competes in today or realistically can in the near future. Fourth, Apple is a consumer device customer, not an AI customer. The memory that goes into a MacBook Air is commodity LPDDR while the memory that goes into an Nvidia GB200 AI cluster is HBM3E. These products share a name and almost nothing else. Losing Apple to CXMT on consumer DRAM is the equivalent of a luxury steakhouse losing a lunch sandwich order. Fifth, the regulatory outcome is far from certain. The Trump administration has been tightening, not loosening, export controls on Chinese semiconductor companies and even if Apple gets a narrow license, it would likely come with restrictions that limit scale, product type, and duration. Bears will try to scare you but do not fall for it, long Micron. Follow me @MelvinInvests for more AI, semis and the next big market themes.
Show more