저자: DanIves · 외국인
요약 보기
AI 인프라 병목(메모리, 전력·냉각, 네트워크, 광통신)이 AI 투자의 다음 기회라고 주장. 메모리 수급 불균형은 2028~2029년까지 지속될 것이라 전망하며, 마이크론이 메모리 부족 이후 주가 급등한 사례를 근거로 듦. Vertiv(매출 24% 성장, 2026 가이던스 상향), Arista(매출 30억 달러 돌파, 영업이익률 49.9%), Coherent(매출 34% 성장, 구리→광연결 전환 수혜)를 병목 수혜주로 제시하고, 밸류에이션과 경쟁·고객 집중 리스크도 언급.
다듬은 글
What is getting more expensive? Look, memory. I mean, the reality is that we're seeing a memory supercycle, and I think that's the word that's being used all over Korea right now . And that's why Nvidia will have to pass on some of these costs to its customers. But it's not slowing down, because today the reality is this: demand for chips exceeds supply by 15 times.
And I think that applies not only to Nvidia, but ultimately to their end users. Now, through the supply chain, you will see that some of the costs will be absorbed, but ultimately the end user will pay. These are customers. These are enterprises. This is tokenization. As all this unfolds, the question arises: OK, when do we reach some kind of equilibrium?
We don't expect it before mid- to late- 2028. And I think it's a tough situation right now, whether it's for Nvidia or for memory manufacturers, Micron and everyone else, because hyperscalers, as Lisa said, will continue to attract capital and capital expenditures will continue to increase, because the AI revolution is just beginning, and I think you'll hear that from Nvidia this week, and then it'll come to memory and Apple, which we're already seeing.
These costs will continue to rise. Dan Ives just gave one of the most important warnings for AI investors in 2026, and most of Wall Street may still be looking the wrong way. In a recent interview, Ives stated that we are only 10-15% of the way to the AI revolution, and his message was straightforward.
The biggest opportunity may no longer lie simply in owning companies that create AI models or GPUs. Real money can be made by owning companies that supply what AI cannot function without. Because that's the problem. Big tech companies are investing hundreds of billions of dollars in AI infrastructure.
Amazon has raised its capital spending forecast for 2026 to about $220 billion, much of which is directed toward AI. Microsoft, Alphabet, Meta, and other hyperscalers are also aggressively expanding their AI infrastructure. But Ives says demand is so far outstripping supply that the industry could remain constrained for years.
He specifically points to memory as a critical bottleneck, arguing that supply and demand may not reach equilibrium until 2028 or even 2029. And besides memory, he sees an even bigger problem looming : energy. After all, you can build more GPUs, more servers, and more AI models, but if you don't have enough power, cooling, networking, and optical connectivity to run it all, the entire AI revolution hits a wall.
And this is where the most interesting things begin for investors. Micron has already shown us what happens when Wall Street suddenly realizes that one seemingly boring component becomes critical to the AI supply chain. The company's shares have soared as the memory shortage has fundamentally changed its profitability.
So the question is not who will be the next Nvidia. A better question is: who will be the next Micron? Which companies control the " bottlenecks" that every AI data center needs , and which could face explosive demand if the shortage worsens? Today, we'll look at three hidden AI infrastructure stocks that may be ready for just such an opportunity.
And the first of them is at the intersection of AI's biggest physical problem —power and cooling. Let's start with this. Promotion number one—Vertiv . Let's start with the problem that lies directly between the AI chip and the electrical grid: power and cooling. You can build the world's most powerful AI accelerator, but if the data center doesn't provide enough electricity or remove enough heat, that accelerator won't be able to do anything.
This is where Vertiv gets extremely interesting. Vertiv provides critical digital infrastructure, including power management, cooling systems, thermal management and other equipment required to operate high-density data centers . And the financial results show that this is not just an AI story invented by investors .
In the second quarter of 2026, Vertiv generated approximately $3.27 billion in revenue, up 24% year-over- year. Adjusted operating profit rose 51%, while adjusted diluted earnings per share (EPS) jumped 60% to $1.52. Even more impressive is that adjusted free cash flow reached about $925 million, up 234% from last year.
Management also raised its full-year 2026 forecast, expecting sales of approximately $14 billion, representing approximately 31% organic growth on average. Adjusted earnings per share are expected to be around $6.70. And here is the broader investment thesis. AI racks become much denser. More GPUs means more power.
More electricity means more heat. More heat means more difficult cooling. And this creates a chain reaction of infrastructure demands. The bottleneck is not simply whether companies can buy GPUs, but whether they can physically deploy those GPUs at scale. It is this type of bottleneck that can create a price advantage for infrastructure providers.
The risk, however, lies in the valuation. Vertiv is no longer an " uncharted" asset, and expectations for it are very high. If hyperscaler capital spending slows, project timelines are stretched, or data center construction encounters capacity constraints, stocks could experience significant multiple compression.
But, in essence, Vertiv is one of the most obvious ways to bet on the physical complexity of AI. And that brings us to the next bottleneck . Because even if you have the power, cooling, and GPUs, you still need to connect it all. Promotion number two—Arista Networks. Imagine an AI data center as a giant brain.
GPUs are neurons. And the network is a nervous system. And as AI clusters grow from thousands of accelerators to potentially hundreds of thousands, data transfer between these processors becomes as important as the processors themselves. This is an opportunity for Arista Networks. Arista is one of the main beneficiaries of high-speed Ethernet networks in large cloud and AI environments, and its recent performance is hard to ignore.
In the second quarter of 2026, Arista's revenue was $3.036 billion, surpassing the $ 3 billion mark in a quarter for the first time. Revenue increased by 37.7% year-on-year and by 12.1% compared to the previous quarter . Non-GAAP operating margin reached 49.9%, and non-GAAP earnings per share (EPS) increased to $1.02, approximately 40% higher than last year.
But the most important achievement may not be quarterly revenue. This is a technology roadmap . Arista introduced platforms with 1.6 terabits per second data rates for AI networks, including liquid- cooled configurations for scaling AI networks. This is important because network requirements are growing along with computing power.
Nvidia's latest results provide another clue. Its networking business has shown consistent growth , supporting the idea that networks are becoming a larger part of the AI infrastructure stack. Analysts point squarely to Arista and other network companies as beneficiaries of this trend. And that's what sets Arista apart from speculative AI infrastructure stocks .
It already has exceptional profitability. This is not a company that is waiting for demand for AI to emerge. AI accelerates an existing business model. The main risk is valuation and competition. Cisco, Nvidia, Broadcom, and hyperscalers developing their own specialized networking solutions are creating competitive pressure.
Arista also has significant dependence on a relatively concentrated group of large customers. But if AI clusters continue to scale, networks will not disappear. They are only becoming more important. And that leads us straight to the third action, which is probably the most interesting bottleneck of all three.
Promotion number three—Coherent . If Vertiv solves the power problem and Arista solves the networking problem, Coherent attacks another fundamental limitation: transferring huge amounts of data at incredible speeds as AI data centers scale. Copper connections are increasingly facing limitations in terms of distance, bandwidth, power consumption, and signal integrity.
This is where optical connections become critically important. Coherent is a large company in the field of photonics and optical technologies, and its involvement in data center infrastructure for AI is becoming increasingly significant. The fourth quarter of fiscal year 2026 was particularly successful for the company.
Revenue reached $2.05 billion, up 34% year-over- year. Gross margin (non- GAAP) increased to 40.2%, and earnings per share ( non-GAAP) were $1.74, up 74 cents from a year earlier. However, management's comments are even more important than these numbers. Coherent notes that AI data center architectures are increasingly moving from copper to optical connections.
The company entered fiscal year 2027 with significant customer demand and expanded production capacity. The company is also actively expanding its manufacturing presence in the United States. In June 2026, Coherent announced it had received $50 million in a “chip” program to expand its Sherman, Texas, plant, where it creates optical networking technologies for next-generation AI centers .
The market is already aware of these opportunities. Coherent shares were extremely volatile in August, but the company's securities remain one of the most direct ways to invest in the optical infrastructure layer of AI. As of August 26, the stock closed at around $294.37 after fluctuating significantly throughout the month.
This volatility is important. Coherent is not a low-risk utility company. This is a fast-growing semiconductor and photonics company that is dependent on cycles of rapid technology development. If AI
Comments 0