Wall Street Stopped Believing AI Scarcity

H2 opened with the first sustained rotation out of AI chips — the Philadelphia Semiconductor Index fell roughly 12% in two sessions, momentum dropped 18%, and the Dow hit a record while the Nasdaq fell twice straight because Wall Street stopped pricing infinite GPU scarcity the same week June payrolls missed by half.

What moved on July 2:

• The Philadelphia Semiconductor Index fell as much as 6% Thursday, putting it on pace for a two-session decline of roughly 12% — the steepest since early June.
• The VanEck Semiconductor ETF (SMH) dropped 5.8% and iShares Semiconductor (SOXX) fell 7.0%, per TS2 Tech.
• Micron tracked toward a 12%+ weekly loss. SanDisk's two-day slide topped 20%. Teradyne fell 13.6%. KLA slid 11.5%.
• Meanwhile the Dow Jones Industrial Average closed at a record high — powered by Apple, McDonald's, Disney, Visa, and Walmart.

The scarcity thesis cracked:

• FXStreet's Stephen Innes wrote the market is "changing seats" — not selling everything, but marking down AI, memory, and semiconductors while financials and consumer discretionary find bids.
• Momentum fell more than 18% over two days, its worst two-day decline since November 2020.
• Memory stocks dropped more than 18% in two sessions — their sharpest move in at least 12 years.
• The trigger wasn't a demand collapse. It was a question: Can the AI build-out keep running at this pace without supply catching up?

Meta lit the fuse:

• Bloomberg reported July 1 that Meta will rent spare GPU capacity through a new cloud business — Meta rose 9% Wednesday, then fell 4.6% Thursday as investors digested the contradiction.
• JPMorgan analyst Doug Anmuth told clients he'd "much prefer that Meta develop core AI products" over selling infrastructure access, IBTimes reported.
• Neocloud names CoreWeave and Nebius took collateral damage. D.A. Davidson's Gil Luria said pressure would land "more likely to be on neoclouds than the big hyperscalers."

Jobs data poured gasoline:

• The BLS reported just 57,000 June payroll gains on July 2 — well below the 110,000 economists polled by Reuters expected, with April and May revised down by 74,000 combined.
• The unemployment rate fell to 4.2%, but the civilian labor force shrank by 720,000 — participation hit 61.5%, a more than five-year low.
• Traders trimmed Fed hike odds. Gold jumped 2%+. Treasury yields fell. Defensive ETFs — health care, staples, utilities — all rallied.

The split nobody wants to price:

• Enterprises are still spending: hyperscaler capex is tracking above $700 billion in 2026. AI infrastructure remains one of the strongest earnings stories heading into Q2.
• Workers are thinning out: roughly 700,000 of the labor-force decline came from workers aged 25 to 34, per FXStreet's read of the household survey.
• Investors want proof, not narrative: bottom-up consensus still expects ~22% S&P 500 earnings growth. The easy trade was owning the story. The next trade is picking who can still deliver.

What to watch:

• Q2 earnings from Nvidia, Microsoft, Alphabet, Amazon, and Meta — the burden of proof just went up.
• Whether the rotation broadens into a sustained de-rating or stays a crowded-trade unwind.
• July 7 White House voluntary frontier-model standards — another governance overhang for the sector.

Convina's view: The AI boom isn't dead. The monopoly trade is. Markets spent two years pricing a world where every GPU finds a buyer and every hyperscaler stays capacity-constrained forever. Meta's surplus-compute headline didn't prove oversupply — it proved investors were desperate for any excuse to take profits after an 88% quarterly run in semis. The June jobs miss added macro cover. Hyperscalers are still spending hundreds of billions. The question for H2 isn't whether AI works. It's whether the stocks already priced perfection before the workloads showed up.