We toured an AI data center to see how our stocks make these facilities work
The part of artificial intelligence that often gets overlooked lives inside data centers like CoreSite’s sprawling complex just outside New York City. CEO Juan Font thinks of his company’s facilities like shopping malls. “You have multiple stores sharing the same building, the same power, the same cooling, but in the digital sense, as opposed to having every company build their own data centers,” Font told CNBC during a recent tour of CoreSite’s NY3 facility in Secaucus, New Jersey. AI-optimized data centers like NY3 are popping up all around the country as technology companies crave more and more high-powered computing capacity to run their artificial intelligence workloads. They are the physical manifestation that enables the everyday tasks that we conduct on smartphones, computers, wearables, and connected cars. Big Tech — including portfolio names Amazon , Alphabet , Meta Platforms , and Microsoft — has already poured billions upon billions of dollars into data center buildouts. Collectively, these four companies are on track to spend at least $608 billion this year to keep in the AI arms race. That number may move even higher after all four companies report earnings after Wednesday’s closing bell. On top of all that, newcomers OpenAI and Anthropic are also spending at furious clips ahead of their expected initial public offerings (IPOs), which could happen, at the earliest, near the end of this year. You have multiple stores sharing the same building, the same power, the same cooling, but in the digital sense. CoreSite CEO Juan Font For these complex, power-hungry data centers to operate, an ecosystem of technologies must come together. This new AI infrastructure gold rush is creating massive opportunities for Investing Club names tied to major parts of these facilities. To better understand how data centers are built and how some of our holdings — from top chipmakers Nvidia and Broadcom to leading energy generator GE Vernova — come into play, I took an in-depth tour of CoreSite’s Secaucus facility. I also spoke with executives at Amazon and another facility designer, Prime Data Centers (no relation to Amazon’s subscription service Prime). Here’s what I learned. Power It starts with power. “The innermost loop of this intelligence revolution is power,” CoreSite’s Font said as we walked through his company’s Secaucus facility. CoreSite is owned by communications real estate investment tru s t (REIT) American Tower , and operates 30 data centers in 11 markets across the U.S. Before installing a single piece of technology, data center developers source electricity as the first and most critical step of the process. (Energy is also the base layer of Nvidia CEO Jensen Huang’s “five-layer cake” of AI .) “If we go back three years, [clients] would be coming to us and saying, ‘Do you have 10, 20, 30 megawatts?’ … That’s enough power for a sizable town,” said Michael Wall, executive vice president of product delivery at Prime Data Centers, which has 29 facilities in seven American markets and three in Europe. “Now they say, ‘How many hundreds of megawatts do you have available?'” Wall said demand has even scaled into multiple gigawatts (1,000 MW) for really big centers, reflecting the rapid increase in power needs driven by artificial intelligence. Data center size is measured by these energy increments as a nod to power being the limiting factor of how much computing capacity they can deliver. The powerful chips used to train and operate AI models — and the labyrinth of infrastructure to run, connect, and cool them — put pressure on an already constrained public energy grids. “The grid capacity as it stands today has probably been taken up,” said Wall, adding that companies need real estate with either existing grid capacity or can support what he calls behind-the-meter power generation. The supply gap has led companies to seek alternative, supplemental avenues such as natural gas, fuel cell, solar, and wind solutions to keep projects on track. CoreSite selected fuel cell technology , which is the conversion of gas from a utility provider into electrical energy without combustion. GE Vernova , which supplies equipment like natural gas turbines, is another popular choice for data centers. Shares of the Club stock have surged more than 60% year to date after doubling last year. GEV YTD mountain GE Vernova YTD AI demand for energy products used to power data centers helped fuel GE Vernova’s stellar quarter report last week. Revenue grew 16% year over year to $9.3 billion in the first quarter. GE Vernova also saw strong equipment orders, driven by growth across all of its segments. The company added $13 billion to its backlog in 90 days, expecting to reach a $200 billion backlog in 2027, a year sooner than anticipated. We increased our price target on the stock to $1,300 after its more than 12% post-earnings rally. We kept our buy-equivalent 1 rating on shares. Jim Cramer called GE Vernova’s stock “one for the ages.” Power management solutions to run data centers at peak efficiency is where Club name Eaton comes in. “Our PDUs, power distribution units, are provided by Eaton and Schneider,” said Anthony White, CoreSite’s senior manager of data operations. CoreSite also sources the facilities’ remote power panels from Eaton and Schneider. Remote power panels are the last piece of equipment needed before power is delivered to customer racks, said White, who walked me through CoreSite’s NY3 complex. Back in February, Eaton’s fourth-quarter earnings revealed that its Electrical Americas’ order backlog hit an all-time record high, growing 31% year-over-year. “Electrical Americas is seeing unprecedented demand with all-time high backlog and record order intake,” said CEO Paulo Ruiz during the call. Shares of Eaton, which reports first-quarter results next week, are within a few percentage points of last week’s all-time closing high. The stock has gained roughly 28% year to date. ETN YTD mountain Eaton YTD Chips All of that power leads to the brains of the operation, the server rooms. Inside those rooms, you’ll find rows of tall rectangular-shaped racks that house high-performance chips. In CoreSite’s case, its clients choose their preferred chip vendors. The all-around gold standard AI chips from Nvidia and the emerging demand for custom chips from Broadcom are at the top of companies’ wish lists. Nvidia shares had been in a funk for much of this year. But like many of its peers, the stock has surged since the market’s March 30 Iran war lows. Shares have now gained more than 13% year to date — and closed at a record high Monday with a market cap back above $5 trillion . During our April Monthly Meeting , Jim said, ” Why do I cling to Nvidia? I think the better question to ask is, why would you sell it?” He added that Nvidia is “still the most important company in the AI universe, no matter what anyone says.” Over the past three years, Nvidia has skyrocketed more than 650%. Over that same three-year stretch, Broadcom has surged nearly 540%. So far this year, the stock has gained rouhly 15.5% year to date. Hyperscalers are increasingly looking for custom chips, which is emerging as a test of Nvidia’s industry dominance. This month, we trimmed Broadcom three times to right-size the position, which had swelled to an over 5% weighting in the portfolio. We generally like to keep our portion weighting to about 5%. Broadcom co-designs Google’s increasingly popular tensor processing chips (TPUs), which will see their eighth generation split into two chips: One for training and one for inference. In addition to Alphabet’s Google, Broadcom announced last week a multi-year, multi-generation strategic partnership with Meta to support the company’s Meta Training and Inference Accelerator (MTIA) chips through 2029. NVDA AVGO mountain 2022-12-30 Nvidia vs. Broadcom since Dec. 30, 2022 The data centers that hyperscalers are building differ from the CoreSite in a couple of ways — in particular, they’re much larger in scale. CoreSite’s NY3 facility is 138,000-plus square feet. Amazon’s $10 billion data center campus being built in North Carolina will have 20 buildings, each between 200,000 to 225,000 square feet, according to local officials . “We are building at a pace that we haven’t seen before in our very long history in this space,” Amazon Web Services Vice President Kerry Person told me. Amazon designs its own data centers, networking equipment, and some of its own chips, such as Graviton CPUs and AI-focused Trainium chips. To be sure, Amazon still buys a lot of Nvidia. Chips deployed throughout AI-optimized facilities like the one I toured at CoreSite are used for inference, or the use of pre-trained models. Inferencing is essentially what helps models like OpenAI’s ChatGPT or Google’s Gemini produce answers to questions. While inference doesn’t require the same scale as model training, CoreSite said these facilities still need to support dense clusters of GPUs and the high power demands that come with them. Optical fiber Also critical to data center operations is fiber connectivity. Thousands of fiber connections carry massive amounts of information at extremely high speeds. Optical fiber, like that made by Corning , runs through CoreSite’s facility, enabling data to quickly move between servers, cloud providers, and networks. Data centers need more fiber and faster connections than ever before as artificial intelligence workloads grow. Fiber, which runs faster and cooler than copper, is becoming increasingly popular. Faster, cooler runs use less power and reduce the load on the massive cooling equipment at these facilities. Excitement around the growth of fiber has made Corning one of the best stocks in the S & P 500 this year, even including Tuesday’s 8.9% pullback in response to earnings. Corning’s numbers were strong; the issue is that expectations were too high. That created a good entry point for investors, according to Jim, who emphasized that on Wednesday’s Morning Meeting. Corning shares are up more than 70% year to date. In its first-quarter report, Corning said sales grew 18% year over year to $4.35 billion, headlined by a 36% increase in its optical communications segment, which houses its data center business. Notably, Corning also announced two more long-term supply agreements similar to the deal it inked with Meta back in January. Meta agreed to pay Corning up to $6 billion for fiber-optic cables in its AI data centers through 2030. Corning told CNBC it expects about 8 million miles of its latest optical fiber solution, named Contour, to go through Meta’s Hyperion, its largest under-construction data center located in Louisiana. GLW YTD mountain Corning YTD The optical connectivity solutions provider is actually expanding its Hickory, North Carolina, cable factory to accommodate demand from others like Nvidia, Google, Amazon, and Microsoft. Corning is expected to report first-quarter results early next month. Networking As data moves through optical fibers, it still requires direction. Networking equipment is used to direct traffic inside and outside the data center. Network switches and routers ultimately determine where data goes and how fast it gets there. Broadcom is also involved in this part of data centers, producing hardware like network switches that help manage data traffic. In its most recent report on March 4, Broadcom’s networking revenue grew 60% year over year, making up one-third of total AI revenues for its fiscal first quarter. According to Broadcom CEO Hock Tan, the company expects its AI networking business to grow to 40% of total AI revenues. Nvidia also has a massive networking business, thanks in large part to its wise acquisition of Mellanox in April 2020. Its networking revenues alone topped $31 billion in its latest fiscal year — more than 10 times larger than where Mellanox was at the time of the takeover. “Demand for our scale up and scale out technologies reached record levels, both growing double digit sequentially, driven by strong adoption of NVLink, Spectrum-X Ethernet, and InfiniBand,” Nvidia CFO Colette Kress said during the company’s latest earnings call on Feb. 25. “On a year-over-year basis, growth was driven primarily by NVLink 72 scale up switches as Grace Blackwell Systems accounted for roughly two-thirds of data center revenue in the quarter,” Kress added. Former portfolio holding Cisco produces hardware such as routers that permit fast data transfer. The Club exited our position in Cisco in late March to replenish our cash reserves following our annual donation to charity. Cooling All of that power, processing, and networking generates an enormous amount of heat. Prime Data Centers said some server racks produce heat equivalent to “hundreds of hair dryers” running at once. Managing that heat has become another monumental problem for modern data centers. Traditional air cooling, which uses computer room air conditioners, is no longer sufficient for many data center server rooms. Instead, facilities like CoreSite’s NY3 AI-optimized data center are increasingly turning more to liquid cooling systems, where coolant is delivered directly to servers or even individual chips to prevent overheating. “Probably the biggest innovation that AI has brought to data center design is liquid cooling capabilities,” said CoreSite’s Font. Probably the biggest innovation that AI has brought to data center design is liquid cooling capabilities CoreSite CEO Juan Font Companies like Eaton and Dover are important to this transition, supplying cooling systems, power management equipment, and thermal connectors that help keep these facilities running efficiently. Eaton last month closed on a $9.5 billion acquisition of liquid cooling leader Boyd Thermal. Jim thinks the Boyd acquisition was an excellent and strategic move by Eaton, which he also called “the best at everything that’s electrical.” Jim is much more upbeat about Dover after it delivered last Thursday what he called a “shockingly good quarter,” which sent the stock up 5.5% in response. “We can look at this and say that [Dover CEO] Rich Tobin earned the right to stay in our portfolio,” Jim said during last Thursday’s Morning Meeting . It had been on a short leash, but the numbers delivered. Dover, which produces thermal connectors used in data center liquid cooling solutions, said it expects over a $1 billion in revenue this year from applications linked to artificial intelligence and power generation infrastructure. “In data centers, increasing density of thermal requirements are necessitating a shift towards liquid cooling, which directly benefits our connector and heat exchanger businesses,” Tobin said during the company’s earnings call. DOV YTD mountain Dover YTD Bottom line From power to chips to fiber to networking to cooling, the Club portfolio owns stocks in companies in all five buckets. The rush to build data centers is being funded by technology companies looking for greater and greater computing power in the age of AI, which Jensen from Nvidia has called the fourth industrial revolution. The AI boom is not just transforming our daily lives; it’s transforming the fortunes of so all sorts of companies — from the household names like Amazon that also deliver us toothpaste and clothes, to the behind-the-scenes players like Eaton and GE Vernova. As investors, we’re riding both sides of the wave. (Jim Cramer’s Charitable Trust is long GEV, GLW, AVGO, NVDA, META, GOOGL, AMZN, ETN, DOV. See here for a full list of the stocks.) As a subscriber to the CNBC Investing Club with Jim Cramer, you will receive a trade alert before Jim makes a trade. Jim waits 45 minutes after sending a trade alert before buying or selling a stock in his charitable trust’s portfolio. If Jim has talked about a stock on CNBC TV, he waits 72 hours after issuing the trade alert before executing the trade. THE ABOVE INVESTING CLUB INFORMATION IS SUBJECT TO OUR TERMS AND CONDITIONS AND PRIVACY POLICY , TOGETHER WITH OUR DISCLAIMER . NO FIDUCIARY OBLIGATION OR DUTY EXISTS, OR IS CREATED, BY VIRTUE OF YOUR RECEIPT OF ANY INFORMATION PROVIDED IN CONNECTION WITH THE INVESTING CLUB. NO SPECIFIC OUTCOME OR PROFIT IS GUARANTEED.
<
