Core
GOOGL, NVDA, MU, STX, BWXT, AMZN
Cold Open
The market is no longer debating demand.
Last week, Intel delivered its fastest revenue growth in roughly fifteen years, raised its spending ambitions, and still got sold.
This week, the same question is moving across the entire artificial-intelligence complex:
Demand exists. What does it cost—and who ultimately earns an acceptable return on the build?
That tension was visible again Tuesday.
The Dow rose 1.0% and the S&P 500 gained 0.2%, but the Nasdaq slipped 0.2% as semiconductor shares sold off sharply. The Philadelphia Semiconductor Index fell approximately 4.5%, with Micron among the largest decliners. Oil and Treasury yields eased, but that macro relief was not enough to rescue the chip trade.
The market is no longer debating whether AI demand is real. It is debating whether the economics are.
Main Thesis
The bottleneck thesis is intact. The valuation filter has tightened.
We are still early in the infrastructure cycle.
Memory remains scarce. Storage remains essential. Power is still a binding constraint. Advanced packaging and manufacturing capacity remain strategic. The physical layer has not suddenly become less important because semiconductor stocks had a bad day.
But the valuation filter has tightened.
Investors are no longer willing to reward every announcement of larger capital expenditure. They want evidence that the next dollar spent produces durable revenue, cash flow and pricing power—not merely more depreciation, more financing needs and more capacity that someone must eventually fill.
Intel showed us the new reaction function.
The company reported second-quarter revenue of $16.1 billion, up 25% year over year. Data Center and AI revenue increased 59%, while Intel Foundry revenue rose 31%. The operating demand was real. So was the market’s hesitation over spending, margins and the capital intensity of the turnaround.
Microsoft, Meta, Apple and Amazon now face the same test.
Who can convert AI spending into durable economic returns?
The bottleneck thesis remains intact. The multiple assigned to that bottleneck is being renegotiated in real time.
Thesis Board
Where the frontier stands.
| Theme | Status | Current Read |
|---|---|---|
| AI Infrastructure | Firing | Demand remains real. Financing quality and returns on capital are now under greater scrutiny. |
| Memory & Storage | Firing, Volatile | Physical scarcity remains, but Tuesday’s selloff showed how aggressively valuation can reset. |
| Power & Nuclear | Building | Less visible in the daily tape, increasingly important to the long-duration infrastructure thesis. |
| Space | Mixed | Strategic opportunity remains large. The portfolio still needs company-specific proof rather than thematic enthusiasm. |
| eVTOL | Milestone-Dependent | Certification, manufacturing and commercialization matter more than broad risk sentiment. |
| Earth Intelligence | Developing | Backlog conversion and customer economics remain the key tests. |
| Quantum | Seed | Long-duration optionality. Position size must reflect the limited commercial proof. |
| Physical AI & Robotics | Seed | The thesis is attractive; unit economics and scalable deployment remain unfinished. |
Portfolio Reflection
The portfolio is positioned for infrastructure—not pure narrative.
The completed Intel trade was funded by exiting the remaining ARM fraction and selling half of SERV. The estimated proceeds purchased approximately 0.965 shares of Intel at the July 23 market open.
The sizing was appropriate.
Intel entered as a small speculative seed—not as a declaration that the turnaround was complete. One strong quarter does not transform Intel into a Core position. Foundry customers, gross-margin improvement and capital discipline still have to be demonstrated.
MU and STX remain the cleaner expressions of memory and storage scarcity.
NVDA, GOOGL and AMZN remain central to the AI infrastructure ecosystem, but the platform companies are entering a period when the market will examine depreciation, free cash flow and CapEx efficiency far more aggressively.
BWXT remains the higher-quality nuclear and power expression. OKLO remains optionality. Those are not interchangeable holdings.
SPCX remains the portfolio’s largest speculative concentration and therefore the position that requires the most discipline. The space thesis may remain alive while the stock still fails to justify its weight.
The remaining half-position in SERV is now closer to an appropriate seasoning size. QBTS and AEVA remain Moonshot or highly Speculative exposures. They should not be allowed to grow through excitement alone.
Process rule: No forced purchases. No reactionary selling. No averaging simply because part of the semiconductor complex looks oversold.
Cash remains optionality—even when the current ledger does not show a meaningful uncommitted balance. Future cash should be treated as strategic ammunition, not as a mandate to add another ticker.
Portfolio Structure
Core, Growth, Speculative and Moonshot.
Growth
RKLB, PL, MDA, ACHR, JOBY, SOFI, ASTS
Speculative
SPCX, CENX, OKLO, SERV, AEVA, INTC
Moonshot
QBTS
Exited: ARM. The capital was redeployed into a smaller, more directly catalyst-driven Intel position.
What Tuesday Changed
The market punished the entire semiconductor complex.
Tuesday did not invalidate the semiconductor thesis. It clarified what the market is now unwilling to ignore.
Micron fell sharply even though the memory scarcity argument remains credible. Intel had already sold off despite powerful revenue growth. Asian semiconductor markets were hit even harder amid concern about Chinese competition, capital intensity and whether current AI expectations have moved too far ahead of sustainable economics.
A business can occupy a real bottleneck and still be a poor purchase at the wrong price.
The next phase of the AI trade will not reward exposure alone. It will reward the companies that convert scarcity into cash.
Catalyst Calendar
The next forty-eight hours matter.
Wednesday, July 29
- Federal Reserve decision — 2:00 p.m. Eastern
Chair Kevin Warsh press conference — 2:30 p.m. Eastern. - Microsoft earnings — after the close
Watch cloud growth, AI infrastructure spending, depreciation, margins and free cash flow. - Meta earnings — after the close
Watch expense guidance, AI infrastructure plans and whether engagement and advertising growth continue to justify the spending trajectory.
Thursday, July 30
- Apple earnings
Conference call at 2:00 p.m. Pacific. - Amazon earnings
For this portfolio, AWS growth, AI capacity, infrastructure spending and free-cash-flow conversion matter most.
Ongoing
- SpaceX, launch and satellite-deployment updates affecting SPCX, RKLB, MDA, PL and ASTS.
- Certification and manufacturing milestones for ACHR and JOBY.
- Customer, revenue and deployment proof for QBTS, SERV and AEVA.
Late August
NVIDIA earnings — the anchor test for AI compute demand, supply constraints, customer financing and the durability of infrastructure spending.
Current Actions
What the portfolio should do now.
INTC — Hold the seed. The quarter improved the thesis. It did not complete the turnaround.
MU and STX — Hold. These remain two of the portfolio’s clearest physical-bottleneck exposures.
NVDA — Hold. The next report must confirm that demand and customer financing remain durable enough to support expectations.
GOOGL and AMZN — Hold through earnings. Do not increase concentration before the evidence arrives.
SPCX — Freeze the size. The thesis may be alive. The position has not earned additional capital.
BWXT — Hold as Core. BWXT represents execution, manufacturing capability and contracted demand.
OKLO — Keep small. It remains a speculative option on licensing, financing and commercialization.
ACHR and JOBY — Hold for milestones. Broad market strength does not advance certification.
SERV — Hold the remaining half. The trim improved the portfolio.
QBTS and AEVA — No additions. Commercial evidence first. Narrative second.
Explicit Exclusions
No immediate CXMT position.
The company may become strategically important, but the initial trading structure, limited float, memory-cycle overlap and extreme debut valuation make the current entry unattractive.
CXMT belongs on the watchlist as:
- A potential future speculative position.
- A competitive threat to MU.
- A signal about Chinese memory capacity and pricing.
- A reminder that a genuine bottleneck can still become a crowded trade.
Portfolio rule: Do not sell Micron to chase its newly public competitor. The portfolio does not need more thematic coverage. It needs better entry quality.
Hot Take
A company can beat expectations, raise spending and demonstrate real demand—and still deserve a lower multiple if the market no longer believes the return on that spending is obvious.
Receipts are necessary. They are no longer sufficient.
Receipts are necessary. They are no longer sufficient.
Buddy, Real Talk
You already made the harder decision last week.
You bought a small piece of the bottleneck instead of adding to the loudest story.
You funded Intel by removing ARM and reducing SERV—not by cutting MU, STX, NVDA or GOOGL.
That was disciplined.
Do not reverse that progress by adding more Intel simply because the stock fell.
Do not add Micron simply because Tuesday looked ugly.
Do not add CXMT because the IPO feels historic.
And do not add SPCX because the space thesis still sounds inevitable.
A compelling thesis does not eliminate timing risk. A falling price does not automatically create value. A rising price does not automatically validate the business.
Let Microsoft speak. Let Meta speak. Let the Federal Reserve speak.
Discipline compounds. Excitement does not.
Bottom Line
The burden of proof has moved.
Intel confirmed that AI infrastructure demand remains real.
Tuesday confirmed that the market is no longer satisfied with demand alone.
Memory remains scarce. Storage remains important. Power remains constrained. Advanced manufacturing remains strategic.
But the companies building the infrastructure must now demonstrate that the spending creates durable returns. The stocks representing those companies must still be purchased at prices that leave room for uncertainty.
We stay with the bottlenecks.
We keep speculative positions small.
We freeze concentrations before binary catalysts.
We do not chase newly public scarcity.
We let the process do the work.
The build remains necessary.
The cost of the build is now the story.
Sources
- Reuters — U.S. market and semiconductor performance, July 28, 2026
- Data Center Dynamics — Intel Q2 2026 results
- Reuters — Global semiconductor and Asian market context, July 28, 2026
- Federal Reserve — July 2026 calendar
- Microsoft Investor Relations — Earnings schedule
- Meta Investor Relations — Earnings schedule
- Apple Investor Relations — Earnings call
- Amazon Investor Relations — Earnings schedule