Does AI already have human-level intelligence? The evidence is clear

The authors argue that current large language models have reached a level consistent with artificial general intelligence, citing GPT-4.5 being judged human 73% of the time in a March 2025 Turing-style test and capabilities including gold-medal International Mathematical Olympiad performance, solving PhD-level problems, generating experimentally validated scientific hypotheses, and producing literary output preferred by readers. They dismiss common objections (e.g., 'stochastic parrots', lack of world models, embodiment, autonomy) and urge policymakers and investors to reassess risk, regulation and strategic allocation in light of materially advanced AI capabilities.

Analysis

Market structure: AGI-as-fact crystallizes durable winners — hyperscaler cloud providers (MSFT, AMZN, GOOGL) and GPU-leader NVDA — plus niche AI software platforms and data‑center REITs. Expect pricing power in datacenter GPUs and cloud AI services to support gross‑margin expansion of +200–500bps for leading vendors over 12–24 months; low-cost incumbents (INTC, legacy on‑prem software like ORCL) face margin compression. Content, low‑skill service providers and certain education incumbents (e.g., CHGG) are direct downside candidates as automation replaces repeatable labor.

Risk assessment: Tail risks include (1) regulatory shocks — EU/US AI mandates or model‑sharing rules within 3–12 months that could force revenue dilution; (2) export controls halting advanced GPU shipments to China; (3) model safety incidents causing rapid de‑rating. Near term (days–weeks) sentiment spikes are likely; medium term (3–12 months) sees re‑rating and capex cycles; long term (2–5 years) productivity gains could lower labor costs but raise corporate capex and energy demand. Hidden dependency: concentration in NVDA for advanced accelerators and in a few cloud providers for distribution creates single‑point systemic risk.

Trade implications: Direct plays — overweight NVDA (capture tight GPU supply/pricing) and MSFT/AMZN (cloud/OpenAI exposure); hedge with short INTC and short education/content disruptible names (CHGG). Use 6–18 month LEAPS to express convexity, adding on pullbacks of 8–12%; take profits on +30–50% rallies. Rotate out of consumer discretionary into IT hardware/software and utilities that can hedge rising datacenter power demand over next 12–36 months.

Contrarian angles: Consensus underestimates regulatory and energy constraints and overestimates absolute winner‑take‑all economics — verticalized specialist AI firms can capture lucrative niches, limiting mega‑cap upside beyond the next 12 months. Historical parallel: 2010–2015 cloud ramp where incumbents rerated but many niche players also outperformed; expect dispersion, not uniform alpha. Unintended consequences include downward pressure on wages (supporting bonds) or spikes in industrial power prices (benefiting utilities/energy).

Key Decisions for Investors

• Establish a 2–3% portfolio long position in NVDA (ticker NVDA) via 9–12 month LEAPS or equivalent notional; accumulate on any pullbacks of 8–12% and target trimming at +40% (take‑profit) or stop at -20%.
• Add 2% long in MSFT (ticker MSFT) and 1% long in AMZN (ticker AMZN) to capture OpenAI/cloud exposure; express with 9–12 month calls (LEAPS) sized to 1–2% notional each and reduce holdings by 50% if regulatory developments impose >5% estimated cloud revenue hit for either within 90 days.
• Implement a 2% long NVDA / 2% short INTC (ticker INTC) pair trade to exploit expected GPU vs CPU spread widening over 6–12 months; unwind if the NVDA/INTC spread narrows by >30% or if INTC reports >10% sequential improvement in datacenter design wins.
• Establish a 1% short position in Chegg (ticker CHGG) as a high‑conviction disruption candidate over 6–12 months; maintain stop‑loss at +25% and target downside of -40% based on substitution risk from LLM tutoring and homework tools.
• Hedge macro tail risk by reducing portfolio duration by ~0.5 year (trim core bond exposure by 1–2%) and add 1% allocation to utility/energy names (e.g., XLU or selective power generators) to insulate against higher datacenter energy demand over the next 12–36 months.