저자: DanIves · 외국인
요약 보기
AI 인프라 병목(메모리 부족, 전력망 위기, 데이터센터 용량 부족)을 짚고, 빅테크 자본지출이 2026년 6000억 달러 이상으로 증가해 AI 투자는 아직 초기라고 주장. 마이크론(MU)은 HBM 수요와 장기계약으로 메모리 부족 수혜, 마벨(MRVL)은 커스텀 ASIC·광 인터커넥트·CXL 성장, 콘스텔레이션에너지(CEG)는 원전 기반 무탄소 전력 부족의 수혜주로 제시.
다듬은 글
Memory is the biggest constraint we see, and equilibrium will likely not come until 2028-2029. Now you can argue about stocks. What do they reflect? Is this cyclical? Did this change the cyclicality? Energy is definitely the biggest obstacle, and I think the main problem is that the US needs to build data centers to make all this work.
There is no doubt about it. I think the biggest risk in US- China relations is data centers , because there can be no debate: we win in chips, models, hyperscalers, and intellectual property, and China wins in robotics and energy. But if data centers aren't built here, it's like, you know, a hospital without surgeons or without the technology to do its job.
So this continues to be a real bottleneck . If you've been waiting for the most sensible and realistic voice on what's ahead for the AI market, while the latest market fluctuations have investors panicking. Wall Street veteran Dan Ives just presented an important reality check. While sensational headlines warn of an imminent bursting of the AI bubble due to Chinese models with open scales that eliminate the price advantage, the math suggests otherwise.
Capital spending by big tech companies is not slowing down. These are growing to a projected $600 billion+ in 2026, with over 75% going directly to physical AI infrastructure. We are not at the finish line yet. We are only in the third quarter of a decade-long technological transformation. Leading closed-source labs like OpenAI and Anthropic are building huge corporate moats around their own data that openly available solutions cannot touch.
The real risk to AI investments is not a lack of demand for software or falling model margins. The real risk lies in the massive supply-side constraints that are stifling the entire ecosystem: a shortage of high-speed memory, a serious power grid crisis, and a lack of physical data center capacity. To benefit from this market shift, you need to keep an eye on physical bottlenecks.
Here's a complete analysis of the leading stocks that are positioning themselves to address these three critical infrastructure deficits. Let's start with Micron Technology, ticker MU, a cornerstone of American memory manufacturing and a major player in the high-speed hardware shortage space. Micron is the main beneficiary of the global shortage of high-speed memory, which puts the company directly at the center of AI development.
What makes the current moment so attractive is the sharp disconnect between short-term market sentiment and net operating results. After a recent 26% drop driven largely by broader market volatility and the liquidation of positions in foreign memory indexes, Micron's fundamental engine has not lost its rhythm.
The company is on a roll thanks to unprecedented demand for HBM3e, which is critical for next-generation AI accelerators such as Nvidia's Vera Rubin and AMD's Helios architectures. Modern AI models require significantly expanded context windows, which means that hardware architectures need more, not less, memory bandwidth.
To satisfy this huge appetite, hyperscalers enter into long-term contractual commitments. , 14 of which require a minimum threshold of $100 billion in residual commitments over the next 5 years. Importantly, these floor prices guarantee profitability that exceeds the peak quarterly levels seen in any previous memory market cycle.
Looking ahead, the financial picture is staggering. After a quarter with revenue of $41 billion and a net margin of 68%, Micron's near-term forecasts suggest revenue growth to $50 billion. In addition to the visibility of individual contracts, US hyperscalers prioritize security of supply, which gives Micron a distinct structural advantage along with continued incentive support under the US chip law.
Even as memory price increases naturally normalize over time, Micron's cash flow remains protected by take-or- pay agreements and $22 billion in customer financing. With industry peers like Samsung indicating that global memory shortages could last until 2028, Micron finds itself at a historic tipping point .
The business is converting huge infrastructure demand into high-margin contract cash flows, solidifying its place as an indispensable player in the hardware landscape . Moving from memory chips directly to the limitations of physical connectivity, our next pick is Marvel Technology, ticker MRVL. This is a powerhouse of high-speed data center infrastructure that is undergoing structural transformation.
Marvel is at the intersection of AI cluster scaling, high- bandwidth optical interconnects, and custom semiconductor development, becoming essential where data centers reach thermal and speed limits . Although the stock recently underwent a sharp 50% correction from June highs to around 46 to the price of future earnings, buyers on the downside have begun to return near key technical support levels as long-term fundamentals strengthen.
A fundamental validation for Marll was a major strategic investment and $2 billion partnership with Nvidia that integrates Marll's custom XPUs and optical solutions directly into the NVLink platform and Spectrum 6 Ethernet ecosystem. To sharpen its focus on these hypergrowth opportunities, Marll streamlined its operations by selling non-core automotive assets for $2.5 billion in cash, redirecting capital directly into developing high-yield AI chips.
Marll's competitive advantage is based on three powerful catalysts. The first is the custom ASIC business, where hyperscaler programs including AWS Trainium 2 and 3, Google Axion ARM CPU, and Meta DPU are expected to increase custom semiconductor revenue from $1.5 billion to over $4 billion by fiscal 2028.
The second is leadership in interconnects, where the mass adoption of 1.6 terabit optical DSPs should drive over 70% revenue growth in the interconnect segment. The third is the groundbreaking Compute Express Link (CXL) market position, which directly addresses memory shortages by allowing servers to efficiently pool DRAM.
Marvell forecasts that revenue from CXL alone will reach $2 billion by fiscal year 2028. As AI models scale, triggering thousands of autonomous queries across distributed clusters, demand for Marvell's retimers, network cards, and optical switches is growing exponentially, with management setting revenue targets of $11.5 billion for 2027 and $16.5 billion for fiscal 2028.
Marvell is a vital infrastructure player, turning physical connectivity constraints into multi-year revenue growth. Moving from data center equipment to the electricity needed to power these facilities, our final pick is Constellation Energy, ticker CEG, which is solving the main problem of the entire AI revolution: electricity generation.
As hyperscale data centers grow from tens of megawatts to gigawatt- class campuses, access to reliable, 24/7 carbon-free energy has become the biggest barrier to AI development. Constellation Energy holds an unrivaled competitive position as the largest producer of carbon-free electricity in the U.S., controlling more than half of the country's commercial nuclear fleet.
While critics often mistake Constellation for a regular utility company, the fundamental reality is one of sheer structural deficit. The demand for electricity for data centers in the US is projected to grow to 90 gigawatts by 2030. But the total power that can be supplied
Comments 0