Inside VMF Research: May 2026, the month the AI trade moved down the stack
Process over narrative: Moving from the visible hardware bottlenecks to the underpriced application layer.
Every great investment theme eventually reaches an uncomfortable point.
At first, the market does not believe it. Then it questions it. Then it slowly accepts it. And finally, after prices have already moved, the same idea that once required imagination starts to feel obvious.
That is when the real work begins.
Because the goal is not to be early forever. Nor is it to remain loyal to the first expression of a thesis simply because it worked. The goal is to keep asking where the mispricing still lives.
That was the question behind almost everything we published in May.
We did not become bearish on AI. Quite the opposite. We continue to believe artificial intelligence is one of the defining investment themes of this cycle. But we also believe the market has become much better at pricing the first layer of that theme: the hardware, memory, data-center, power, and infrastructure bottlenecks that make the AI buildout possible.
Those bottlenecks were once underappreciated. They no longer are.
That does not mean they cannot keep going. It does mean the easy money, if there ever was such a thing, has already been made in some of the most visible parts of the trade.
So May became a month of rotation...
Not a retreat from AI.
A rotation inside AI.
From the part of the trade the market can now see clearly to the parts it still finds harder to underwrite.
That is the thread connecting the May issues of VMF’s Strategic Asset Allocation, VMF’s Security Selection, and Alpha Tier. And it is also the best way to read the free excerpts we published on Substack during the month.
Think of this note as the map.
Begin with the bottlenecks
The first stop is The Visible Side of the AI Boom.
This excerpt came from May’s issue of VMF’s Strategic Asset Allocation, where we revisited one of the biggest winners in our Model Portfolio: South Korea.
When we recommended South Korean equities, the idea was not to make a generic country call. It was to own one of the less obvious beneficiaries of the AI buildout. Everyone understood Nvidia. Fewer investors fully appreciated what the explosion in AI compute would mean for memory: high-bandwidth memory, server DRAM, enterprise SSDs, and the broader supply chain around Samsung and SK Hynix.
AI feels weightless to the end user. Behind the curtain, it is one of the most physical, capital-intensive technology waves the world has ever seen.
That was the opportunity.
The market eventually caught up, and South Korea moved from being an underappreciated AI bottleneck trade to one of the clearest examples of the thesis working. That created the kind of problem every investor wants to have: what do you do when you are right?
The answer is not always “hold forever.”
Sometimes the right answer is to harvest, resize, and redeploy.
That was the first lesson of May. Once the market starts agreeing with you, the job changes. Conviction got you into the trade. Discipline has to manage what happens next.
Then follow the trail to the application layer
The second stop is The Forgotten Application Layer.
If the first AI trade was about building the machine, the next one may be about using it.
That distinction shaped the entire month.
The hardware layer is easier for the market to price. More AI usage means more compute. More compute means more chips, more memory, more data centers, more power, more cooling, and more grid infrastructure. The earnings bridge is visible.
The application layer is less tidy. It is harder to know in advance which companies will turn AI into better economics. Some will use it to improve pricing, logistics, advertising, software development, customer acquisition, fraud detection, personalization, and operating leverage. Others will merely attach themselves to the narrative.
That uncertainty is precisely why the opportunity may still exist.
Markets usually pay first for what is obvious. They pay later for what still requires judgment.
In May’s Strategic Asset Allocation issue, that led us to China’s internet ecosystem. Not because China is clean or comfortable. It is neither. The property bust, regulatory scars, weak confidence, geopolitical tension, VIE structures, governance discounts, and years of foreign-capital exhaustion are all real.
But investing is not about pretending problems do not exist. It is about asking whether the market has already priced them... or over-priced them.
China may not lead the frontier-model race against the United States. But it remains one of the most important markets in the world for AI deployment. Commerce, payments, logistics, advertising, consumer platforms, manufacturing, robotics, mobility, industrial automation... these are not abstract use cases. They are exactly where AI may move from technological promise to economic value.
That was why we increased exposure to China internet in Tier One.
The free excerpt explained the thesis. The full paid issue went further: vehicle, sizing, funding source, target weight, technical setup, catalysts, and the Model Portfolio implications.
From the basket to the business
The third stop is The Index Is Expensive. The Opportunity Set Is Not.
This excerpt came from May’s issue of VMF’s Security Selection, and it addressed a mistake we think many investors are making right now.
They look at the index and conclude that stocks are expensive.
That is partly true. The major U.S. indexes are highly concentrated. The largest companies are expensive relative to the rest of the market. The most obvious AI winners have already attracted enormous amounts of capital.
But the index is not the opportunity set.
A market can be expensive and still contain mispriced businesses. In fact, the more concentrated the market becomes, the more likely it is that capital keeps crowding into what is easy to own while more complicated opportunities are left behind.
That is where security selection matters.
Not because every cheap stock deserves attention. Most cheap stocks are cheap for a reason. But when a business is controversial, cash-rich, strategically positioned, and priced as if too many things will go wrong, the work becomes worth doing.
That brought us to PDD: China Price, Temu Optionality.
PDD is not the kind of company investors instinctively place inside the AI conversation. It does not make chips. It does not own data centers. It is not a frontier-model lab.
But it sits much closer to where AI may actually be monetized: commerce, pricing, product discovery, advertising efficiency, logistics, merchant tools, demand forecasting, and transaction flows.
The market mostly sees a controversial Chinese e-commerce company with Temu risk attached. That concern is understandable. We began the paid issue with the bear case because any thesis that ignores why a stock is cheap is not analysis. It is hope.
But after confronting the risks, we asked the more useful question: what if the market is confusing discomfort with impairment?
Our view is that PDD is better understood as a marketplace intelligence engine. It sits between consumers, merchants, products, prices, advertising, logistics partners, and transactions. That is exactly the kind of system where better intelligence can become better economics.
The public excerpt gave readers the core idea. The paid issue gave subscribers the full underwriting work: bear case, base case, bull case, valuation model, cash haircut, Temu optionality, sensitivity analysis, and position sizing inside the Value Model Portfolio.
That is the difference between seeing an idea and underwriting an investment.
Finally, the Tulip Test
The last part of the month belonged to Alpha Tier.
The public trail begins with The Price of the Obvious and continues with A Pattern Recognition Exercise.
The title of the paid issue was The Tulip Test, but the point was not that AI is tulips. That would be a lazy argument.
Tulips did not change the world. AI probably will.
The better historical lesson is that real revolutions can still become dangerous when investors pay too much for the future. Railroads were real. The internet was real. Broadband was real. Each changed the world. Each also produced moments when capital rushed into the right idea at the wrong price.
That is what we wanted to test.
Not whether AI is important.
Whether parts of the AI trade have become too dependent on uninterrupted acceleration.
The answer was not simple, which is usually a good sign. AI capex expectations are still rising, which means the boom may have further to run. The application layer has not fully participated. The secular bull market may still have another powerful phase ahead.
But the first AI trade has become crowded. Earnings revisions are concentrated. Momentum is concentrated. Leadership is concentrated. And the seasonal window has become less forgiving.
So Alpha Tier did what Alpha Tier is supposed to do. It translated pattern recognition into portfolio action.
The conclusion was rotation, not retreat.
We were not selling the future. We were reducing exposure to the parts of the future the market has already discovered, adding to the application layer, and increasing liquidity because the risk/reward had changed.
The public excerpts walked through the framework. Paid subscribers received the actual tactical changes: what we trimmed, what we added, how much liquidity we increased, and what we will monitor from here.
The May reading order
Read May in this order:
1. The Visible Side of the AI Boom
The market has discovered the cleanest AI bottlenecks.
2. The Forgotten Application Layer
The next opportunity may sit where AI gets used, not where it gets built.
3. The Index Is Expensive. The Opportunity Set Is Not.
A concentrated index can hide mispriced businesses.
4. PDD: China Price, Temu Optionality
Our stock-level expression of the China application-layer thesis.
5. The Price of the Obvious
Real revolutions can still become dangerous when the price gets too obvious.
6. A Pattern Recognition Exercise
The first AI trade is crowded, but the second AI trade may still be forming.
That was May inside VMF Research.
Tier One gave us the map.
Tier Two took us from the basket to the business.
Alpha Tier made the risk-budget decision.
The month was not a collection of disconnected articles. It was a sequence: identify the visible winners, recognize where the market has already paid us, move toward the underpriced application layer, select a direct stock-level expression, and adjust the tactical portfolio.
That is what we are trying to build at VMF Research.
Not more market noise.
A process.
One that connects macro regimes with asset allocation, investment themes with individual securities, and conviction with position sizing.
Because in a market flooded with information, the scarce asset is not another opinion.
It is judgment.
A quick note on accountability.
We don’t publish these theses to be right on paper. We publish them to express edge in the real economy. Our Leaderboard shows the exact scorecard since inception, tracking every position, our compounding outperformance against the market, and the triple-digit winners we’ve captured along the way.
You can view the exact numbers on our Leaderboard.
Disclaimer: This is general information, not personalized investment advice. It’s not a recommendation to buy or sell anything. Investing involves risk, and past performance doesn’t guarantee future results. Do your own research and consider speaking with a licensed/authorized professional who understands your objectives and risk profile.