La défense a chuté de 20 %. Les semi-conducteurs ont bondi de 40 %. Même guerre.

BREAKING (May 5, 2026): Trump announced that Project Freedom, the US Navy operation launched Sunday evening to guide stranded ships out of the Strait of Hormuz, would be paused so a deal with Iran could be finalized. The blockade itself stays in place. The semis ETF closed at a fresh high in the same session, and the structural setup behind that close is the subject of this piece.

Key Takeaways

• The textbook war trade just inverted: Defense primes Lockheed and Raytheon fell 20% and 13% over the 30 days from April 6 to May 5, the same window the semis ETF SOXX gained 40.3% and SMH gained 32.0%.
• The Strait stayed closed the entire time: Brent ranged $86 to $126 intraday and ended roughly flat at the early-May settlement, while site lore documented the Strait of Hormuz operating at a trickle relative to its 360-ship/day baseline.
• Memory led the rally most aggressively: Micron rose roughly 70% over the full 30-day window, closing at $640.20 on May 5 from $377.76 on April 6. Lam Research rose roughly 25% over the same window. Defense-exposed analog chip names lagged with single-digit gains.
• Microsoft’s capex pullback failed to break the rally: SMH dipped one session on the announcement, then ripped to a fresh high for the move within five sessions. Someone else is bidding.

The Trade That Inverted

The textbook war trade is simple. War breaks out: buy defense, sell tech. Pentagon contracts flow into Lockheed and Raytheon. Inflation crushes long-duration tech multiples. Repeat for every Middle East crisis since 1973.

That trade just inverted in real time. From April 6 through May 5, with the Strait of Hormuz still effectively closed, Lockheed Martin (LMT) closed down 20.2%. RTX Corporation, the parent of Raytheon, closed down 12.9%. Over the same 30-day window, the iShares Semiconductor ETF (SOXX) closed up 40.3%, and the VanEck Semiconductor ETF (SMH) closed up 32.0%.

That is roughly a sixty-point spread between the asset class war was supposed to feed and the asset class war was supposed to crush. The spread did not happen during a ceasefire. It happened during a war that produced a Trump-announced ceasefire that failed inside eleven days, a Microsoft data-center capex cut of 3.5 gigawatts, and an Iranian drone strike on the UAE port that justified leaving OPEC. Defense kept losing. Semis kept ripping.

What the Strait Actually Did

The phrase “Strait of Hormuz closed” is doing a lot of work in news headlines. The reality is more specific. Pre-war Kpler vessel traffic through the Strait ran at roughly 360 transits per day. During the brief Trump-announced ceasefire window of April 8 through April 10, vessel trackers counted twelve transits across three days. Iran is operating a parallel toll system charging ships $1 to $2 million per passage in yuan and crypto. The Strait is technically navigable, practically locked.

Brent crude reflected the chaos. Front-month futures opened the 30-day window at $109.77 on April 6, traded as low as $86.08 intraday on April 17, traded as high as $126.10 intraday on April 30, and settled near $108 in the early-May session. Net move: roughly flat, with a 47% intraday range across the window. The “oil shock” never went away. It just stopped behaving like a single-direction shock and started behaving like a permanent fog.

For the textbook war trade, this is an unusually adverse setup. Persistent oil tail risk, no resolution catalyst, zero visibility on when shipping resumes. The setup should have suppressed cyclical tech and bid up defense. It did the opposite.

Why Defense Lost Anyway

You would think a war that has burned through 850-plus Tomahawk cruise missiles in 30 days against an annual procurement run-rate of 57 would be a screaming buy signal for the prime contractors who build them. Backlog at Lockheed and Raytheon is at record levels. Order books are stacked.

The market is pricing something else. Three plausible explanations, with one of them sourced and the other two analytical:

Procurement timing (analytical). Defense procurement is widely understood to be slow. A new munitions replenishment contract awarded mid-cycle typically does not convert to recognized revenue at the prime contractors for several years, given contracting, supply-chain spin-up, workforce qualification, and program review. If equity markets price roughly twelve months of forward earnings, a backlog measured in years is a future-tense asset, not a present-tense one. That logic sits behind the underperformance, but the article is inferring it from the price action rather than pointing to a specific contracting timeline.

Deficit risk on out-year contracts (analytical). US debt held by the public crossed 100% of GDP in March, and foreign sovereigns have walked their share of Treasury holdings down from 49% to 31%. A war financed on top of that fiscal stack does not, by itself, guarantee defense sequestration. But it does make every out-year defense dollar more politically contested. The longer the procurement cycle, the more sensitive the revenue line is to a future austerity Congress. The market may be repricing that exposure; the article cannot prove it is.

Valuation absorption (sourced). Seeking Alpha’s defense desk noted in early May that “major U.S. defense stocks failed to outperform during the Iran conflict, tracking the S&P 500’s negative performance instead,” because “current valuations in the defense sector reflect already priced-in expectations of significant long-term earnings growth.” Translation: defense already had the war priced in before the war started. There was no incremental multiple to expand. This is the only one of the three explanations that has a published market analyst on the record.

Why Semis Won Anyway

The semiconductor rally is not a single rally. It is a divergence inside a divergence. Memory ran the hardest. Micron Technology (MU) closed April 6 at $377.76 and May 5 at $640.20, a roughly 70% gain across the 30-day window, with the steepest leg coming in the final week. Lam Research (LRCX), which sells the deposition and etch tools that DRAM and HBM fabs depend on, rose roughly 25% over the same window from $220.65 to $275.80. NVIDIA, the headline AI accelerator name, lagged badly: it rallied to a closing high of $216.61 on April 27 before fading to $196.50 on May 5, leaving the GPU bellwether up only about 11% across the full window. Defense-exposed analog names like Texas Instruments and Analog Devices logged single-digit gains. Within semis, the closer a name sat to defense end-markets, the worse it traded.

The mechanics behind that rotation are the inverse of the defense story:

Revenue is hitting now. Hyperscaler AI capex committed in 2025 is being deployed in 2026. Servers are being installed. HBM3e modules are shipping. Lithography tools are being delivered to TSMC and Samsung in real time. The capex cycle that semis are pricing is happening this quarter, not in 2028. The market discounts forward earnings, and forward earnings for memory and equipment are a 2026 event, not a 2030 event.

Near-zero Hormuz shipping exposure. The chip supply chain runs Asia-to-US and Asia-to-Europe over Pacific routes, plus air freight for finished wafers. None of it transits Hormuz. Raw-material exposures that do touch the Persian Gulf, particularly Qatari LNG feeding Korean and Taiwanese fab electricity, have been backstopped by alternative gas supply at a higher price. Higher fab electricity costs barely move chip COGS, since the value of a finished wafer is overwhelmingly intellectual property rather than energy input. Airlines run roughly a third of operating cost on fuel; petrochemicals can run higher. Semis sit at the asset-light end of the spectrum.

Scarcity premium under stress. Memory pricing is set by spot DRAM and NAND markets that respond to inventory tightness in real time. The HBM3e generation that powers AI accelerators has been supply-constrained for three quarters. In a war environment, no buyer wants to be short of inventory. The “just in case” bid that emerges in stressed supply chains lands disproportionately in memory pricing, which lands disproportionately in Micron and SK Hynix margins.

The Microsoft Test

The cleanest stress test of the rally arrived on April 26, when Microsoft quietly killed 1.5 gigawatts of near-term data center builds and walked away from 2 gigawatts more in non-binding leases. Microsoft is the largest single buyer of AI compute on the planet. If any one buyer had the muscle to tell semis where to trade, it was Microsoft.

SMH dipped exactly one session on the news, closing at $491.21 on April 28 from the $506 close on the prior session. Five sessions later, SMH closed at $522.69, a fresh high for the move. The market absorbed the largest hyperscaler walking back capex, and re-bid chips to a new high. That is not a buy-the-dip technical reaction. That is a refusal to accept the premise that Microsoft is the marginal buyer.

Which raises the question the rest of the tape is leaving on the table: if the dominant buyer pulls back and the price goes up, who is bidding?

The Stealth Bid

There are three plausible answers, and they are not mutually exclusive.

Hyperscaler #2 through #5 backfilling. Google, Meta, Amazon, and Oracle have not publicly announced equivalent pullbacks. The aggregate 2026 Big Tech AI capex figure cited in the site’s prior Microsoft coverage tops $660 billion, and Microsoft’s own cut does not, on its face, reduce that aggregate. Whether competitors actively absorb the released capacity or simply do not slow their own programs is a question the article cannot resolve from price action alone.

Sovereign AI demand from outside the US. Several non-US AI infrastructure programs (Saudi Arabia’s PIF-backed compute build-out, UAE’s G42, India’s IndiaAI mission, European national-champion projects) have been publicly announced over the past year and represent a category of chip demand that does not appear on any US hyperscaler’s 10-K. The article does not have a sourced number for what fraction of recent foundry and equipment-maker revenue this category represents.

Foreign capital rotating out of US debt. The site’s prior coverage of the foreign-sovereign Treasury exit is explicit that the immediate replacement bid in Treasuries came from leveraged hedge funds running the basis trade, not from the same foreign capital reappearing in US equities. Whether any of the dollars that left long-duration sovereign debt are also turning up in US-listed semis is plausible but unverified by the cited source. Treat this third explanation as the weakest of the three.

What the rally rules out is cleaner than what it rules in. It is not a one-buyer story; the dominant buyer pulled back and the price went up. Beyond that, the identity of the marginal bid is genuinely an open question.

The Bear Case

The honest bear case is not that the rally is a fake. It is that the rally is a stretch.

Cheddar Flow flagged a $1.2 million SOXX put sweep in late April: an institutional desk bought 1,152 of the May-1 expiry $460 strike puts at the ask, the kind of outsized short-dated hedge that lands on the tape when a desk expects a near-term reversal. The Philadelphia Semiconductor Index posted 17 consecutive winning sessions from March 31 through April 23, the longest streak in the index’s 32-year history. The Motley Fool noted that prior streaks of comparable length over the past three decades typically delivered total gains of 8% to 10%; this one delivered more than 40%.

The rally has also been narrow. The semis index has done the bulk of the index-level work since the March lows; equal-weight S&P measures and the broader market have lagged the headline number meaningfully. A market this concentrated in one subsector is one news cycle away from a violent rebalance.

None of that contradicts the structural thesis. Even a steep pullback would still leave SOXX well above its April low. The trade that inverted is still inverted. The point is that the inversion is durable, not that the rally is linear.

What This Means for the Real Economy

Here is the part the tape is not pricing.

If the war trade has structurally inverted, if oil shocks now feed semis instead of crushing them, then capital concentration in chips is not a short-term rotation. It is a permanent feature of how shocks transmit through the equity market. Every future macro shock now gets bought in the few sectors with no fuel exposure, no Hormuz exposure, and no procurement-cycle drag. Software and silicon take the bid. Industrials, utilities, energy, and discretionary take the hit.

That is the K-shape, but not the income K-shape policy circles talk about. It is a capital K-shape. Chip equity sits overwhelmingly with top-decile US households and foreign sovereign wealth. The energy inflation feeding the divergence hits everyone. The same dynamic that lifts NVDA widens the wealth gap. The inversion is rational at the portfolio level and corrosive at the household level.

The defense industrial base, the boring, slow, contract-driven part of the economy that is supposed to absorb the war bid, is being told by the equity market that its 36-month backlog is not worth what its current price-to-earnings ratio implied a month ago. The primes mostly fund expansion from operating cash flow and debt rather than new equity, so the direct capex-cost effect of a lower share price is limited. The indirect effect (executive compensation, M&A currency, employee retention via stock comp, supplier confidence) is real but slower-moving. Whether any of that meaningfully extends the war’s duration is a chain the article will not stretch beyond hypothesis. What the article does claim is narrower: capital that should structurally have flowed to defense in this environment did not, and the shortfall went disproportionately to chips.

The trade is rational. The economy under it is not.

The Bottom Line

Across this 30-day window, the war kept running. Lockheed lost a fifth of its market value. Raytheon lost an eighth. The semis ETF that supposedly cannot survive an oil shock posted its longest winning streak on record. Microsoft’s 3.5-gigawatt data-center pullback did not break the trend. Brent traded the $86 to $126 range intraday and the chip bid did not flinch.

The conventional war trade is dead. What replaced it is a capital K-shape that funnels every macro shock into a narrower and narrower set of asset-light, IP-rich, dollar-denominated equities. The names that benefit are real. The system that produces the benefit is not stable. Project Freedom itself was the third Hormuz reset Trump pre-announced in eight weeks. The first two failed inside two weeks each. This one paused inside two days. Watch the open: not what oil does on the headline, but whether semis even bother to react.