Why Is the AI Bubble Selling Off? (June 23, 2026)
Why Is the AI Bubble Selling Off? (June 23, 2026)
On Tuesday June 23, 2026, AI-exposed tech stocks fell sharply. Micron led the decline at -13.18%, Nvidia closed -4.13%, AMD dropped, and Alphabet was down for a second consecutive session. NPR ran a piece titled “Is AI ‘one big bubble’?” Wikipedia’s AI bubble article was actively updated. Business Insider published a “Exactly how the dot-com bubble burst” warning. Here’s what’s actually happening, why it matters, and what changes from here.
Last verified: June 24, 2026.
TL;DR
- Date: Tuesday, June 23, 2026 (second consecutive down day for major AI names)
- Worst hit: Micron Technology (MU) -13.18%
- Mega-caps: Nvidia -4.13%, Alphabet -0.98%, AMD also down
- Earlier: SPCX (SpaceX) fell 10% on June 22 after Colossus 2 contract-clause concerns
- What it is: Broad re-rating of AI-capex beneficiaries, not a single-trigger crash
- What it isn’t: AI revenue collapsing — the underlying AI-product businesses are still growing roughly as guided
- Bear case: Forward multiples vs realistic 2027–2029 cash flows; the ~$1.1T projected US mega-cap AI capex isn’t yet generating proportionate revenue
What actually fell, and how much
From Yahoo Finance and Interactive Crypto reporting close-of-day June 23:
| Ticker | Move | Notes |
|---|---|---|
| Micron (MU) | -13.18% | HBM memory concentration risk |
| Nvidia (NVDA) | -4.13% | Day 2 of decline |
| AMD (AMD) | down (multiple percent) | AI-trade beta |
| Alphabet (GOOGL) | -0.98% | Day 2 of decline |
| Broadcom (AVGO) | down | Networking + custom silicon exposure |
| SPCX (June 22) | -10% | 90-day exit clauses on Colossus 2 contracts |
The pattern is unusual. AI-leadership names (Nvidia, Alphabet) declined modestly, and the AI-capex-leverage names (memory, networking, AI infrastructure pure-plays) declined sharply. That tells you it’s not panic — it’s a recalculation of who actually benefits if capex slows.
What’s driving the move
Four converging concerns:
1. Capex math vs revenue math
US mega-cap AI capex is now projected at roughly $1.1 trillion from 2026 to 2029. That’s an enormous number relative to AI-product revenue in 2026, which (excluding ad-network and search-driven AI revenue) is still measured in tens of billions per company at the top end.
Investors are starting to ask: do the cash flows from AI products plausibly cover that capex over a reasonable depreciation horizon? At current run-rates, the answer is uncomfortable.
2. Adoption-rate doubts
A Business Insider piece from earlier this week, citing market research firm Apollo, drew direct parallels to dot-com 2000:
If AI adoption and effectiveness fall short of expectations, as investors appear to be worried about, it could signal a pullback in investment, dramatically hurting sales for firms tied to the AI buildout.
The enterprise AI revenue cycle (POCs → pilots → production → expansion) is taking longer than the 2025 hype implied. POCs are happening at huge volume, but production expansions are slower than the chip orders required to support them.
3. The Colossus 2 / SPCX signal
SpaceX’s Colossus 2 facility went from internal cluster to $2.3B+/month committed third-party AI infrastructure in roughly two months — but on June 22, SPCX fell 10% after the market focused on the 90-day exit clauses in those Anthropic, Google, and Reflection contracts.
That clause means only ~$7B of the headline ~$97B is contractually firm. The rest is rolling quarterly customer decisions. If even one of those three customers exits, the entire revenue story unwinds quickly.
That’s a more general worry. A lot of the projected AI infrastructure revenue is “soft” in the same way — committed but cancellable, or contingent on future tranches.
4. The political / regulatory overhang
Separately, NPR and the Guardian reported this week that AI-aligned super PACs have now spent over $185 million on the 2026 midterms, with groups linked to OpenAI and Anthropic spending the largest amounts on rival initiatives. The political fight over the RAISE Act (and similar state AI bills) is now real and expensive.
Investors increasingly see AI regulation as a non-trivial cost. That feeds into the bubble narrative because it raises the discount rate.
Why Micron took the worst hit
Micron’s -13.18% is the most analytically interesting move. Micron is one of three global suppliers (with Samsung and SK Hynix) of the HBM (High-Bandwidth Memory) that ships on Nvidia B200/B300 GPUs.
HBM is the most concentrated point of leverage in the AI capex chain:
- Every Nvidia GPU at scale needs HBM
- HBM is a small piece of total bill-of-materials but a huge piece of supplier revenue
- Micron has the highest HBM revenue concentration of the three major memory suppliers
If AI capex slows even modestly, HBM demand softens immediately, and Micron earns less than Samsung or SK Hynix. So a -13.18% move on a “what if AI capex slows” day is internally consistent with that leverage.
What this is, and what it isn’t
What it is
- A valuation re-rating of AI-leverage stocks against more skeptical 2027–2029 forecasts
- A timing reset — the market is re-pricing how long the capex cycle lasts and how steep the eventual revenue curve becomes
- A political overhang — regulation is now a real cost line item
- A structural signal that committed-but-cancellable revenue (90-day exit clauses, multi-year-but-renegotiable deals) is being discounted
What it isn’t
- A revenue collapse. AI revenue at Microsoft, Oracle, ServiceNow, Anthropic, and OpenAI continues to grow roughly as guided.
- A bubble pop. Forward P/E ratios remain elevated but not extreme. Nvidia’s at roughly 30x forward earnings — high, but nothing like 2000-era multiples.
- A regime change. AI infrastructure spend is still real, still growing, and still essential. The question is whether it’s growing fast enough to justify the current price of the leverage names.
What to watch from here
- Q2 2026 earnings (late July through August) — will Microsoft, Google, Meta, and Amazon hold guided AI capex? Or trim?
- Anthropic and OpenAI revenue updates — are Claude Sonnet 5 / GPT-5.5 driving the revenue acceleration the market is pricing in?
- HBM order normalization — Micron’s next earnings report is the most important single data point for the memory leverage trade
- The RAISE Act and state AI bill outcomes — regulation is now a real input to the discount rate
- Colossus 2 customer renewals at the 90-day mark — first decision point on the SPCX revenue thesis