When the music stops: the unravelling of AI companies’ flawed valuations

Public and private markets are pricing large premiums into AI companies amid signs of a valuation bubble and unhealthy circular investing (e.g., Nvidia investing in OpenAI and Coreweave; cloud providers investing in OpenAI/Anthropic). Notable frothy private rounds include Lovable's $330m Series B at a $6.6bn valuation and Mistral AI's €1.7bn Series C at a €11.7bn valuation; structural concerns include overstated ARR metrics (one‑off, volume/performance/value‑based contracts) and true gross margins in the low teens-to-twenties versus 70–80% for traditional software, implying heightened downside risk for speculative AI plays while presenting selective opportunities in durable B2B and agentic AI companies with sticky recurring revenue.

Analysis

Market structure: Winners are GPU suppliers (NVDA) and large cloud providers (AMZN, MSFT, GOOGL) that can monetize compute scale; losers are speculative private AI plays and GPU-resellers with thin margins (CRWV risk indicated). Tight GPU supply + sticky model-inference costs sustain pricing power for semiconductor suppliers and generate higher implied vol in NVDA options; energy and specialty semiconductor materials see demand pressure, while credit spreads on AI‑heavy private comps will widen.

Risk assessment: Tail risks include rapid regulatory intervention (EU/US AI safety/antitrust within 3–12 months), a TSMC/TS supply shock, or a clustered private-market markdown that forces forced selling. Immediate (days) risk = elevated volatility around NVDA earnings; short-term (weeks–months) = private valuation repricing and churn; long-term (12–36 months) = potential margin recovery if inference costs fall 40–60% or durable enterprise adoption materializes. Hidden dependency: circular investing (suppliers invest in customers) can amplify valuation shocks and liquidity cascades.

Trade implications: Tactically favor capital allocators to durable enterprise AI with multi-stakeholder sales (B2B automation, cybersecurity) and underweight hype-driven startups. Implement hedged NVDA exposure (small core long + protective tail hedges), use directional shorts/puts on high‑valuation, low-margin AI infra resellers (CRWV), and rotate 5–10% of tech exposure into AMZN/MSFT for recurring cloud revenue. Time trades around NVDA earnings, GPU pricing signals, and any AI regulatory announcements in next 30–90 days.

Contrarian angles: Consensus underestimates value of slow, gruelling enterprise sales — companies with >45% gross margins and sales cycles >6 months are underpriced versus one-off ARR proxies. The market may be over-penalizing hardware (NVDA) despite durable moats; set tactical buy triggers (NVDA down >20% from today) and opportunistic shorts if CRWV rallies >30% vs NVDA within 60 days. Historical dot‑com parallels hold for sentiment but differ because AI requires tangible compute capital and long-term contracts, creating asymmetric opportunities for well-capitalized incumbents.

Key Decisions for Investors

• Establish a hedged core long in NVDA equal to 2–3% of portfolio: buy NVDA shares and simultaneously purchase 3‑month 5% OTM puts equal to 30% of notional to protect against a >15% drawdown around next earnings (monitor GPU ASPs and TSMC capacity weekly).
• Initiate a 1–2% short position in CRWV (or buy 3‑month ATM puts if borrow is constrained) expecting a 20–40% downside over 3–6 months driven by margin compression and client concentration; trim if CRWV relative performance to NVDA exceeds +30% in 60 days.
• Rotate 5–10% of high‑multiple AI exposure into AMZN or MSFT over the next 4–8 weeks (scale in on any tech weakness >10%) to capture durable cloud monetization; exit if combined cloud margin guidance falls >200bps YoY over two consecutive quarters.
• Allocate 1% to directional options on a ‘AI pullback’ trade: buy a 6‑month NVDA put spread (e.g., 15%/30% OTM) funded by selling a small number of 6‑month NVDA call spreads (10%/25% OTM) to monetize elevated IV while capping risk; reassess after 90 days or post regulatory announcements.