To The Moon

A Contradiction Hiding in Plain Sight

On June 10, the Bureau of Labor Statistics released May CPI data. Core CPI came in at 3.8% year-over-year and 0.4% month-over-month—both above consensus expectations of 3.6% and 0.3%. Housing, transportation services, and electricity continued to drive the upside surprise. The Nasdaq Composite closed down 4.18%, its largest single-day decline in 14 months.

The market's prior trading thesis was linear: ceasefire → oil price decline → inflation eases → Fed stays accommodative → tech stocks continue higher.

But Brent crude had already fallen from $120 to $88. Energy's contribution to CPI was actively contracting. Inflation still printed above expectations, which means the force pushing prices higher is no longer in the energy complex.

So what is?

One answer: AI infrastructure itself.

The Physical Cost of $690 Billion in CAPEX

In 2026, the five major North American hyperscalers (Microsoft, Alphabet, Amazon, Meta, Oracle) have revised their full-year capital expenditure expectations upward to $660–690 billion. $谷歌(GOOG)$ alone guided $180–190 billion. $微软(MSFT)$ reported single-quarter CAPEX of $18.7 billion, up 62% year-over-year.

The physical destination of this spending is highly concentrated: data center construction, power system expansion, semiconductor capacity procurement, and high-density storage and cooling infrastructure.

The result is a systematic run on specific physical resources:

Power: A single large AI data center consumes approximately 300MW. The five hyperscalers have collectively locked in over 10GW of power capacity—equivalent to 10 large nuclear plants. Power Purchase Agreement (PPA) pricing has been pushed into the $100–150/MWh range.

Grid equipment: Delivery lead times for 400kV ultra-high-voltage transformers have stretched to 48 months, driven by global shortages in grain-oriented electrical steel (GOES) and copper. Order backlogs at $伊顿(ETN)$, ABB, and Schneider Electric have locked in multi-year revenue visibility.

Labor: Data center construction requires large numbers of skilled electricians and pipefitters. June nonfarm payrolls printed at 172,000 versus a consensus expectation of approximately 85,000. The labor market remains tight.

Semiconductor equipment: $阿斯麦(ASML)$ Holding NV $阿斯麦(ASML)$ High-NA EUV lithography systems now cost close to €400 million per unit, with delivery and calibration cycles exceeding one year.

These costs transmit to final prices through electricity rates, construction services, industrial goods pricing, and equipment depreciation pass-throughs.

It must be noted that AI infrastructure's direct contribution to headline CPI remains limited—estimated at 0.1–0.3 percentage points. It is not the sole reason CPI exceeded expectations. Housing inflation and services wage stickiness remain the primary drivers. But AI infrastructure inflation is a persistent, structural additive pressure that will not dissipate with a ceasefire. And the additive alone is sufficient to make the Fed's decision calculus materially more difficult.

The FOMC: Warsh's First Meeting as Chair

June 16–17 marks the fourth FOMC meeting of 2026. Kevin Warsh presides for the first time as Chair.

The rate decision itself is not in question. Futures markets price 99%+ probability of holding at 3.5%–3.75%.

The risk lies in forward guidance.

May meeting minutes already revealed 4 dissenting votes—the most since 1992—with several members explicitly favoring a rate hike. Ed Yardeni has publicly stated "the Fed will have to raise rates in July to appease bond vigilantes." Wall Street has abandoned rate-cut expectations for 2026 but has not yet priced in a hike.

If Warsh incorporates language such as "upside risks to inflation remain elevated" or signals openness to tightening, markets will be forced to begin pricing July or September hike probabilities. This expectation shift transmits directly to equity valuations through discount rate adjustments—particularly for long-duration assets.

A straightforward calculation: Assume the risk-free rate rises from 3.75% to 4.25% (+50bp). For a high-growth company where 80% of DCF value derives from years 5–10 (typical Forward P/E of 80x), a 50bp discount rate increase reduces theoretical valuation by approximately 12–15%. For a company whose value concentrates in near-term, high-certainty cash flows (Forward P/E of 15x), the same rate move produces only 3–4% valuation compression.

This is mathematics. Not opinion.

Stress-Testing the Paradox: Can It Be Broken?

The logic chain so far reads: AI infrastructure drives up costs → inflation remains elevated → Fed tightens → tech valuations compress.

A rigorous analysis must test whether this loop is genuinely inescapable.

Break path one: AI productivity output eventually exceeds its physical consumption. If enterprises widely adopt AI to reduce operating costs and compress labor requirements, AI becomes deflationary over time. But this effect requires 12–18 months to manifest in macro data. The current phase remains one where input costs run ahead of output efficiencies. In the near term, the paradox holds.

Break path two: Earnings growth at the top directly overwhelms valuation compression. $英伟达(NVDA)$ Q1 FY27 revenue was $81.6 billion (YoY +85%), EPS $1.87. Alphabet Q1 net income was $62.6 billion (YoY +81%). $美光科技(MU)$ Fiscal Q2 revenue was $23.9 billion (YoY +196%), with 75% gross margin. These growth rates can fully absorb 50–100bp of rate-driven P/E compression. For these companies, the paradox is not a threat.

Break path three: The Fed chooses to look through it. If the Fed judges AI infrastructure as long-term supply-side investment—similar to post-war industrial reconstruction—it may choose tolerance. But Warsh's hawkish orientation and the political pressure of elevated CPI make this path increasingly unlikely.

Conclusion: The paradox is systemically solvable, but it will not be solved within 6–12 months. It persists as a selection pressure, sorting AI assets by quality.

The Sorting Criteria: Three Destinies Under Rising Rates

Category One: Companies Whose Earnings Growth Outpaces Discount Rate Compression

Condition: EPS growth sustained above 50% YoY, with Forward P/E in reasonable territory (below 35x).

Validation:

• NVIDIA: Q1 EPS growth +68% YoY. Q2 revenue guidance of $91 billion (5% above consensus). Forward P/E approximately 30x. Earnings growth capacity far exceeds rate sensitivity.

• Alphabet: Q1 net income growth +81%. Cloud backlog exceeding $460 billion. Forward P/E approximately 30x.

• Micron: Fiscal Q2 revenue growth +196%. Forward P/E approximately 12x (based on Fiscal Q3 guidance EPS of $19.15 annualized). A textbook case of a structurally growing asset priced with cyclical-stock multiples.

These companies can sustain in a 3.75% or 4.25% rate environment. Their primary risk is not rates. It is the growth inflection—if any single quarter shows EPS growth decelerating below 30%, the market will reassess their classification.

Category Two: Companies Whose Valuations Depend on Low-Rate Assumptions

Condition: Forward P/E above 60x, EPS growth below 30%, FCF Yield below 1%.

This category concentrates in the AI application layer—companies building products by calling foundation model APIs but owning neither infrastructure nor models. The sector median Forward P/E currently sits around 55–70x, while most companies have seen revenue growth decelerate to the 20–30% range.

They face dual compression: rising rates reduce the present value of distant cash flows, while generational leaps in foundation models continuously erode the functional moat of the application layer. When base models serve end users directly, the economic value of the intermediary layer is physically compressed.

Category Three: Physical Asset Owners Who Profit Directly from AI Inflation

Condition: Revenue growth derives precisely from the physical costs that AI infrastructure is driving up. Forward P/E below 25x, with order backlogs providing multi-year visibility.

This includes: power equipment and grid integration (transformer price increases = revenue growth), semiconductor manufacturing equipment (capacity scarcity = pricing power), storage hardware (NAND prices up 234% = gross margin expansion), and data center thermal management and liquid cooling systems (mandatory physical upgrades = order surges).

Consider Eaton as a specific case: current Forward P/E approximately 22x, 2026 revenue growth expectation of roughly +15% YoY. More critically, its electrical systems segment backlog has grown over 30% year-over-year, and extended lead times mean these orders carry extremely high certainty and are effectively non-cancellable. Its valuation has never enjoyed an "AI narrative premium," yet its cash flows directly benefit from AI infrastructure's physical demands.

These companies are the hedge within the AI inflation paradox. The more aggressively AI builds, the more constrained physical resources become, and the more certain their revenue streams grow. They do not depend on low-rate assumptions.

Applying This Framework

This is not a directional call. It does not claim tech stocks will collapse or that the Fed will certainly hike.

It is a diagnostic tool.

When the next CPI prints hot, when Warsh signals hawkish intent, when the 10-year Treasury yield approaches 5%—different assets will respond in entirely different ways. Category One stabilizes or briefly dips before recovering. Category Two undergoes systematic valuation resets. Category Three may rise against the broader decline.

For any investor holding AI-related positions, the central question is singular:

Can your company's earnings growth rate sustain the valuation pressure of rates remaining at 3.75%—or higher—through year-end?

If the answer is yes, the rate environment is a secondary variable.

If the answer is uncertain or no, what you hold is not a position based on enterprise value. It is a rate bet.

Final Observation

$690 billion is hitting the ground. Transformers are queued for four years. CPI beat expectations. Oil fell. Inflation did not.

AI is constructing the ceiling above its own valuations.

But that ceiling does not collapse on everyone equally.

It collapses on those who cannot push it up with earnings growth fast enough.