The AI Hype Cycle Short

Capex cycle peak. The four largest hyperscalers are guiding ~$700B combined capex in 2026 — almost entirely into AI infrastructure. NVIDIA data center revenue hit $75.2B in Q1 FY2027 (+77% YoY), but growth is decelerating: the prior quarter was +93%. Deceleration at this scale historically precedes a multi-year digestion cycle.

Merchant GPU displacement is confirmed. Meta signed a 6-gigawatt deal with AMD for custom AI chips (MI450-based), with first GW shipments in H2 2026. Google TPU, Amazon Trainium, and Meta MTIA are all in production. Every hyperscaler now has an alternative to NVIDIA — the CUDA moat is narrowing each cycle. Datacenter lease cancellations are also accelerating — a signal that moves 6–12 months before it shows in earnings.

Enterprise ROI reckoning — now confirmed at the supply side. Microsoft Copilot study (UK, 2026): no statistically significant productivity gains. Uber burned a full year's AI budget in 4 months, capped spend at $1,500/month. Microsoft built its own models to cut OpenAI costs. Goldman Sachs: "AI profit problem getting bigger." Then on June 6, 2026, Broadcom missed its AI revenue forecast — the first AI infrastructure earnings miss of this cycle, wiping $300B in minutes. The demand-side slowdown has propagated to the supply layer. This is where Cisco was in June 2000.

SpaceX IPO lock-up — a structural pressure event. Listed June 12 2026 at $1.75T (~90–115× revenue — 2–3× Cisco's dot-com peak). NASDAQ/Russell ETF inflows absorbed at listing (S&P 500 blocked — rule change rejected Jun 2026). The lock-up is staggered: insider selling begins August 2026 (20% unlock post-Q2 earnings), with 7% tranches every 15 days through October, and full unlock December 9 2026. Elon Musk is locked for one year. Every other insider can sell from August onwards into a market with no remaining forced buyer.

Signal accumulation. The scanner monitors 8 categories of bubble-peak signals daily: enterprise pullbacks, capex guidance cuts, datacenter cancellations, ROI skepticism reports, valuation warnings, insider behavior, regulation risk, and demand softness. Each scored 0–10 by AI.

Universe scoring. AI chipmakers, hyperscalers, datacenter infrastructure, and neoclouds are audited every 2 days — forward P/E, P/S, short interest, options open interest, IV percentile, and ATH drawdown — to rank which names offer the best put entry at any given moment.

Defined-risk entries. Positions are long puts only. Max loss = premium paid per contract. No margin, no undefined risk. Entry timing targets low IV percentile (cheap vol) on green days to minimise premium.

Catalyst-anchored exits. Core positions target Jan 2028 expiry — 13 months past the Dec 2026 SpaceX lock-up event — giving the cycle time to turn without fighting short-dated theta decay.

Government equity stake in OpenAI / AI companies — creates political floor on valuations. If US sovereign wealth fund buys in, pension funds and ETFs follow. Could temporarily re-rate AI stocks up on the announcement, squeezing puts before the correction. Watch closely: this does not fix the underlying cost problem, but it can delay it 6–12 months. Raised as active risk Jun 2026 (Trump/White House discussions confirmed).

AI inference demand exceeds supply into 2027 (volume compensates for efficiency gains); custom silicon fails to displace GPU demand; SpaceX lock-up absorbed without broader pressure; hyperscaler capex guidance reaccelerates post-Q2 2026; breakthrough ROI evidence in enterprise software.