The Great Rotation, Marvell-Lumentum, Cerebras, And AI Winners

The Great Rotation

Semi stocks sold off heavily post the US midterms. Texas is the main story. Gina Hinojosa won the governor’s race after promising a statewide moratorium on new data center permits. Wells Fargo estimates Texas holds about 27% of planned US data center capacity. Projects already under construction would reportedly be exempt. Making matters worse, voters in 10 states rejected data centers in local ballot measures. Goldman Sachs estimates about 45% of planned capacity could face delays or tighter oversight, mainly in Texas, Georgia and Ohio. From the WSJ:

“Voters backed Democratic candidates, from Ohio and Michigan to Texas and Georgia, who have been critical of data-center construction and pledged to slow or pause their rollout. More than a dozen towns in Ohio voted to halt or restrict new development. Similar bans passed in towns in Kansas, Florida and California—a near universal rejection across dozens of measures that signals growing bipartisan opposition.Texas Gov.-elect Gina Hinojosa lamented during her campaign that the state had become a “wild west” for data-center development. “It stops when I’m governor,” the Democrat said. She called for a pause on data-center construction until the Texas state legislature can pass new regulations.Texas, Georgia and Ohio account for more than 40% of all planned data-center capacity in the U.S., according to investment bank Evercore ISI. OpenAI’s Stargate is under construction in Texas and it is planning another massive data-center effort—Project Camellia—in Georgia. Meta Platforms is building a large data center in Texas and another dubbed Prometheus in Ohio.There were few concrete details from Democrats elected on promises of everything from outright bans to more scrutiny. But slowing the rapid growth of properties that house the hardware for AI systems could upend one of the main drivers of U.S. economic growth. A large number of projects already approved for construction will continue, offering another dose of growth and investment for the near term. Still some future projects, even some well into the planning stages, could be scrapped or at least delayed.“This is going to probably elongate the timelines for the data-center build-outs,” Sequoia Capital partner Pat Grady said at the WSJ Tech conference Wednesday. “That’s probably the most tangible impact of what happened with the elections yesterday.”“Now with this blue wave I think it’ll slow down even more,” Brad Gerstner, a venture capitalist who invests in AI companies, said on the “All-In” tech podcast late Tuesday, adding that roughly 25% to 35% of next year’s data-center build-out in the U.S. could be at risk. “That’s why I think it will be really challenging for the AI infra trade,” he said, referring to shares of data-center companies.About 56% of voters nationally said data centers were a crucial or major factor in their picks for Congress, according to a national survey conducted by polling firm SSRS.”

The sell-off then intensified the next day as rising volatility and losses led hedge funds to cut positions. From the WSJ again:

“After a strong start to the year, hedge funds began struggling in the summer and endured a particularly difficult October, helping to turn some funds’ double-digit gains into losses. This week compounded the difficulties when the market went into whiplash, rallying Wednesday after the blue wave that put Democrats in control of both chambers of Congress before plummeting Thursday.A Wednesday client note from Morgan Stanley laid out the scale of carnage for hedge funds betting on and against stocks last month. October marked the largest active unwinding of positions at stock-picking hedge funds in over a decade, with their market exposure down about 9% since June.Firms in recent weeks have been cutting teams and unwinding trades to manage their risk, actions that in turn continue to pressure popular trades held across funds. Many favored investing styles at multistrategy funds, including fixed-income and macro trading based on big-picture market moves, haven’t been working in recent months.Thursday could have generated even steeper losses, had so many funds not reduced their use of leverage, or money borrowed to amplify their bets, in prior weeks, said one bank executive who works with hedge funds.”

Due to the hedge funds’ intense focus on short term performance, this creates automatic feedback loops in the market. In this case, a negative one, with rising volatility and losses leading to more selling as hedge funds try to bring their VaRs (Value at Risk) under control. In some way, these are fairly flawed measures as they measure volatility of your holdings over the last 60 days or so. So, what seemed like a low risk position will suddenly seem like a high risk position on a volatility spike, making you cut your losses while everyone is selling. In situations like these, we usually wait for volatility to drop to then pick up some jewels in the rubble.

Looking at the fundamentals, there remain still plenty of states for data center expansion such as in the Midwest, the Northwest, and the Southeast. The key voter concern is that data centers raise power bills, however, this can easily be resolved by data centers bringing their own power such as gas turbines, fuel cells and battery capacity. So, we don’t think the data center growth story is over, although growth will likely slow down as more states impose restrictions.

Whereas even a year ago, one might have said, “No problem, we’ll build these data centers in the Middle East in countries such as Saudi-Arabia, the UAE, and the likes”, with the explosive situation the Middle East currently is facing—with both Iran and the Houthis being constant threats—we don’t think that hyperscalers will be looking at this region to invest hundreds of billions of dollars into.

Being long AI capex beta has been one of the most profitable trades in the market over the last three years, and if capex growth indeed slows down from here, we suspect that strong share price performance can now start spreading out to more areas in tech. We’re especially looking for names that can grow recurring revenues by selling automation and other AI powered workloads to customers. One obvious area are the big cloud platforms, where enterprises run their digital workloads, store their data, and are now also purchasing ever more tokens.

The big cloud platforms benefit in two ways from a slowdown in AI capex growth. First, this allows them again to grow their operating cash-flow more rapidly than their capex, giving investors steep FCF growth again. The below graph shows what this can look like—if capex moderates, operating cash-flow (the black line) will continue to see attractive growth. The gap between these two is the attraction, this is the growth in FCF:

The below graph illustrates more clearly the FCF profile. Due to the heavy capex spend in 2026-2027—FCF has nearly completely evaporated for the big cloud platforms. The big winners from this trend over the last years were the AI capex picks-and-shovels providers—names such as Nvidia, Broadcom, Vertiv, Caterpillar, etc. However, when AI capex growth slows down, we expect that the big cloud platforms and other hyperscalers will outperform as they can show investors highly attractive FCF growth again, and with a good recurring revenue story from AI.

The second way that the big cloud platforms and AI hyperscalers benefit from a slowdown in AI capex is that they will be able to raise pricing on capacity they can bring online. While a lot of capacity is covered by long term enterprise agreement, part of the capacity, however, is available to be sold at shorter term and spot pricing contracts. The heavy capex spend has been the main concern in the market over the past few years for these names, and so shares in Amazon and Microsoft rallied strongly post election results.

Overall, our view is that AI capex will likely continue to move higher—capitalism is very good at finding solutions—but that the pace will start to moderate at some stage in the coming years. This would mean that AI supply will likely remain highly constrained, with players that have AI capacity available able to sell it at highly attractive pricing.

In a way, this might not be a bad thing for society. It’s highly likely that AI will hugely transform the global economy, and higher AI pricing and AI compute shortages will slow this process down. This gives governments more time to evaluate societal impacts. For example, if 10-20% of office workers would find themselves unemployed at some stage there is more time to carefully evaluate how society can handle this. At the same time, a slower buildout also limits the risk of an overbuild which carries substantial financial risk. There are tons of neoclouds around the world looking to build AI capacity by gearing up with loads of debt, as they don’t have the money to build this capacity, and history shows that this usually leads to bubbles and a financial crash afterwards. However, currently, the risk of this looks very limited to us given that AI compute remains in a large shortage.

UBS semiconductor analyst Tim Arcuri discussed on the Compute 100 podcast his outlook in semis and AI, this is recorded before the election:

“Everyone thinks that NVIDIA is going to grow to be X size and Micron will go to be Y size. If you back into how many gigawatts that implies, the grid cannot accommodate all of that, or what has been permitted currently is nowhere close to that. So you have to believe that there is this conversion of what is not permitted currently, and believe that there is a significant amount of off-grid capacity that also has to come online to get you to where semiconductor expectations already are. This year, we think they are going to ship somewhere in the range of 6 million to 7 million, GPU units. Now, dies and units are not the same thing, but we think it is 6 million to 7 million units this year. If you look at ASICs, our estimate for Broadcom units looking into next year is somewhere in the range of 6 million. They are going to ship probably 3.5 million or 4 million this year. From a unit perspective, ASICs are getting close to the merchant GPUs, so ASICs are growing faster than merchant GPUs. Now, from a dollar perspective, NVIDIA’s dollar share will be higher than its unit share obviously because the ASPs are higher.So there has to be additional things that happen between now and then on the infrastructure side to get you to where we think semis will be in 2028. Not in 2027, but more in 2028, and certainly in 2030 that actually has to happen. Public markets are weighing where the bottlenecks are. With all this capacity being built, it is obvious to me that demand grossly exceeds supply. Why? Because I look at instances that are 7 to 8 years old, Ampere instances, A100s that are coming off lease, and all you have to know to me is that those re-lease for higher. So I have not had to worry that much about demand because it has been all about supply. Most of my work has been around how fast TSMC is expanding wafer capacity and how many tools the equipment makers can make. It has been predominantly supply-side work.I can tell you how much capacity is going to get added. What I have a harder time telling you is whether the demand will be there. I suspect within five years it will be there, but what will be the killer app that drives that demand? I am not really sure. If you take some of the numbers that the labs have put out in terms of revenue that they get per gigawatt, although it sounds like a big number, looking out five years, it is only a couple percentage points of incremental global GDP. If you pose it that way and said to the average investor, “Stocks are baking in that AI will add a couple percentage points within five years to global GDP,” does that sound crazy? I do not think so.Over the past year, people asked about black swan risks. I do not worry about near-term demand or supply. My primary concern has been social and geopolitical backlash—a modern equivalent of people showing up with pitchforks over job displacement or power consumption. That could disrupt political support for AI infrastructure. The industry needs better communication and closer collaboration with government administrations, though the competitive dynamic with China limits how much development can be slowed.I had a debate with a CIO regarding what gets cut first when hyperscaler capex flattens: custom ASICs or merchant GPUs? Most investors assume merchant GPUs get cut first to protect the gross margins of internal custom silicon. My view is that custom ASICs get cut first. ASICs are built for narrow workloads, whereas merchant GPUs provide the architectural flexibility needed during a retrenchment.”

On the final point, we actually think GPUs would see the biggest cut as Nvidia has higher exposure to the neocloud ecosystem, which has the least cash available to actually finance the buildouts. It’s clear that hyperscalers are all in on their ASIC strategy, with both Google and Amazon at the forefront of this trend, and these are the hands with the most cash available.

While we remain constructive on the AI capex outlook, we also continue to look for names that can generate revenues from new AI-powered apps (software-like recurring revenue businesses). Long term, we suspect that this will actually be the most lucrative area for investors. Similar to how the biggest winners from the internet weren’t the picks-and-shovels hardware providers like Sun Microsystems, Dell, Intel, and Cisco, but instead companies that created entirely new markets such as Google with search, Meta with social media, Amazon and Microsoft with the public cloud, Salesforce and ServiceNow with SaaS, etc.

However, also in the short term, this has been an attractive strategy as AI capex names already peaked around May-June of this year—during a real market frenzy in semis—after which we saw a market rotation into other areas in tech such as software, where there are actually attractive companies that are seeing an acceleration in growth from AI. Companies such as Snowflake, Manhattan Associates, Unity, Zscaler, CrowdStrike, etc.

We were fortunate enough to add to a number of these names while some others were panicking, giving us a nice decorrelated alpha from the AI capex cycle. For example, here’s some of what we wrote in early February following a full blown market panic in Unity:

“The latest example of this anxiety is Unity selling off heavily on Friday after the release of Google Genie. We think that Google Genie disrupting Unity is a long way off, if ever. Playing games over Genie would be fairly expensive, as you would have to pay for TPUs/Nvidia GPUs in the cloud per minute. Unity games on the other hand can be played for zero cost on your mobile. Another drawback is that Genie is slow, graphics have to be generated in the cloud on GPUs and then transmitted to your PC or smartphone.Overall, we still like Unity here. While we can’t predict short term market movements, Unity should be a nice asset with attractive growth in the coming years, also as the company is building out an AI driven ad network to compete with Applovin. This is a large market—Applovin has $6 billion in revenue at 70%+ EBIT margins and is still growing strongly. So, if Unity can take a share in this TAM, this would strongly boost profitability.”

Now, we get it, there will be a lot of portfolio managers out there with little understanding of computer science, and so they cut their losses in a position they don’t really understand. Ten months later, no one even remembers Google Genie. But, it dominated news headlines for a few days at the start of the year and did successfully sent Unity’s shares into a tailspin. However, this creates substantial opportunity for people willing to be contrarian, risk some volatility in the short term, and actually dive deep into the underlying tech.

We continue to like Unity here and still have our full position, we’ll be writing about more names like these in the coming months and later in this post.

Next, we’ll discuss our findings on the following topics:

• Cerebras vs Culper’s short report
• The Marvell-Lumentum deal
• And a number of names that will be AI winners in the coming years, and decade, in our view