Listen to the audio version of this article (generated by AI).
On an October morning, Julian Simon opened his mail and found a check for $576.07, signed by Paul Ehrlich, the Stanford biologist whose book “The Population Bomb” warned that humanity was running out of resources. Simon, an economist, was sick of hearing about it. So he made Ehrlich an offer. Pick any raw materials you want, pick any date more than a year out, and we settle this with money.
Ehrlich chose chromium, copper, nickel, tin, and tungsten. Two hundred dollars of each cost $1,000 total, priced in 1980. If that basket was valued more over a decade later, Simon would pay the difference. If it cost less, Ehrlich would be on the hook.
The world added more than 800 million people over those 10 years. Every one of those five metals cheapened in value (after adjusting for inflation). Ehrlich wrote the check.
I keep thinking about that check, because Wall Street channeled its inner Ehrlich on bet concerning artificial intelligence stocks.
In a nutshell, critics rebranded the “scarcity” story as a “peak spending” story. The claim was that hyperscaler capital budgets had topped out, so everyone who bought into the buildout was about to watch as it dried up. That narrative blew up a lot of AI stocks. It took 30%, 40%, and 50% out of some of the strongest businesses in the AI infrastructure complex.
Then earnings season arrived with a bang.
The peak spending thesis is complete hogwash, and the past two weeks prove it emphatically.
So let me walk you through what these companies reported, what the July inflation print does to the rate picture, and why the buy window that just opened does not extend to the rest of the market. You can watch me discuss this in full in the latest episode of Being Exponential With Luke Lango below:
The Neoclouds Settled the Argument
Hyperscalers tell you how much they plan to spend. Chip vendors tell you what is shipping. The Neoclouds tell you something better: whether customers actually rent that infrastructure, what price they pay for it, and how long they commit to it. Utilization, pricing, and forward commitments. That is the read on real demand, and we just heard from the two biggest players in the group.
CoreWeave Inc. (CRWV) grew revenue about 112% year-over-year to $2.6 billion, with adjusted EBITDA of $1.5 billion at a 59% margin. Management guided to an annualized run rate of $18.5 billion to $19.5 billion by year-end. The company finished the second quarter with a $104 billion backlog, up 246%.
Nebius Group N.V. (NBIS) grew faster, which is one reason I recommend it as my favorite name in the space. Revenue jumped 454% to $582 million. Annualized run-rate revenue climbed 56% sequentially to $3 billion. Adjusted EBITDA reached $236 million at a 41% margin. Committed backlog sits above $40 billion, and management said it could sell out all of its planned 2027 capacity today.
Triple-digit revenue growth with 40%-plus margins. You could not dream up better numbers.
Here is the line that matters most. Both companies said newly deployed capacity sells out the moment it comes online. Physical supply is the binding constraint on growth right now. Demand sits there waiting for racks to get plugged in.
That was a rough week to be short Nebius.
Optics Delivered the Knockout
Then we heard from Lumentum Holdings Inc. (LITE), one of my favorite optics names and one of my favorite AI infrastructure names, clean and plain.
Fourth-quarter revenue grew 109% year-over-year. And the first-quarter guide calls for roughly 130% growth.
Sit with that, because it is the most important data point of the past two weeks. A company doing triple-digit revenue growth is guiding for faster triple-digit revenue growth. In a peak spending world, a high-octane supplier like Lumentum decelerates quarter over quarter. Instead, the growth rate accelerates. The pie is getting bigger, and the rate at which it gets bigger is going up too.
The structural story is simple. Every new GPU generation produces more data, runs at faster lane speeds, and demands more bandwidth at lower latency, so AI creates a permanent increase in optical content per system – 800G today, 1.6T and 200G per lane through 2027, then co-packaged optics around 2027 to 2028.
Lumentum is clearly winning dollar share. That does not translate one-for-one to Applied Optoelectronics Inc. (AAOI), Coherent Corp. (COHR), or Fabrinet (FN), though it tells you the whole market is doing extremely well. Fabrinet has the weakest chart in the group, which may make it the most attractively positioned name in it.
Hon Hai, Astera Labs, Arista Networks, Lattice Semiconductor, TTM Technologies, SiTime, and GlobalFoundries all reported the same way. When a narrative runs headfirst into a wall of data that rejects it this emphatically, the narrative loses.
Sentiment and Fundamentals Have to Reattach
In July, sentiment and fundamentals detached. Chicken Little took over the tape while the underlying businesses kept compounding.
Detachments like that resolve one of two ways. Either fundamentals weaken to meet the narrative, or the narrative strengthens to meet the fundamentals. Two weeks of earnings just told us fundamentals are strengthening, so sentiment has to come up. Both the SMH and the SOXX put in V-shaped recoveries off their 100-day moving averages with oversold RSI readings and bullish MACD crossovers. Technical confirmation now sits alongside earnings confirmation.
The macro cooperated too. July CPI came in soft enough to weaken the case for another Fed hike without signaling that the economy is falling apart. Core inflation slowed to 2.5%, its lowest reading in roughly five years, and core CPI is running at 1.6% annualized over the past three months. That real-time trend sits below target while the economy shed about 20,000 jobs in July. Nobody hikes into that. Take the hike off the table, and lower long-term yields feed straight into long-duration assets like AI stocks.
The read for consumer stocks is uglier. Wage growth of 3.2% against 3.4% headline inflation leaves real wages negative again, and that weakness shows up in prints like Dutch Bros and Sweetgreen. The bifurcation that has defined this market since ChatGPT launched persists.
So the buy window is open. And the buy window is open for AI.
Which AI, Though?
That question is where good investors go wrong.
Between 抖阴最新版, 抖阴最新版, and me, InvestorPlace published more than 200 recommendations in the past year, the vast majority connected to AI in some way. You could be right about every theme in this article and still end up with a refrigerator full of ingredients and no meal. Owning the right AI companies is half the job. How much you own of each one, and what you own alongside it, decides your outcome.
So Louis, Eric, and I went through our entire universe of AI research and hand-picked the best ideas in it. Plus, Louis is announcing something Wednesday that he has never done in 47 years in this business.
I cannot go further today. What I can do is make sure you hear it first.
. Once you’re on the list, you’ll get the announcement the moment it goes live Wednesday, no need to do anything further except listen to what we have in store for you.
And for the full breakdown of these earnings, the charts, and the macro picture, watch .
