‘We will be outcompeted’: Leading AI expert warns world may be running out of time to prepare for AI risks

David Dalrymple, a programme director and AI safety expert at the UK-funded ARIA, warns that cutting-edge AI is advancing rapidly and may outpace safety measures, risking destabilisation of security and the economy. He projects that within five years most economically valuable tasks could be performed better and cheaper by machines and that AI could automate a full day of R&D by late 2026, urging governments to not assume system reliability and to prioritize technical controls and mitigation strategies.

Analysis

Market structure: Winners will be GPU/AI-infrastructure leaders (NVDA, ASML, TSM), hyperscalers (MSFT, AMZN, GOOGL) and cybersecurity/OT-protection vendors (CRWD, PANW) as demand for datacenter compute and secure control systems rises materially over 12–36 months. Losers are labour-intensive service providers (staffing, routine BPO) and non-AI-enabled legacy software whose margins face compression; expect pricing power concentration among top-3 model providers and a 200–400bps potential gross-margin tailwind for hyperscalers from AI services over 12–24 months.

Risk assessment: Tail risks include rapid regulatory clampdowns (export controls, model caps) or a high-profile safety incident triggering bans — each could cut adoption growth by >30% in 6–12 months. Immediate (days) risks are sentiment swings and supply chokepoint headlines (NVDA/TSMC); short term (3–12 months) are earnings/margin shifts and layoffs; long term (2–5 years) are structural productivity gains and geopolitical industrial policy reshaping supply chains.

Trade implications: Priority direct longs: NVDA (core infra), ASML (lithography), CRWD/PANW (security) and selective hyperscalers (MSFT, GOOGL) sized 1.5–4% each, scaled over 3–12 months. Pair idea: long NVDA (2–3%), short MAN (ManpowerGroup, 1%) to play tech substitution vs staffing. Use LEAPS to capture 18–30 month convexity (buy NVDA Jan 2027 calls or 350/550 call spreads) and sell covered calls on matured positions after >20% move.

Contrarian angles: Consensus underestimates government procurement and safety-compliance winners — defense primes (LMT, GD) and specialist control-security vendors could re-rate on contracts; also market may be over-allocating to ‘app’ AI names while underweighting semicap/energy names that enable scaling. Historical parallel: dot-com (apps) vs semiconductor/cloud infrastructure winners; unintended consequence — concentrated winner-take-most could invite antitrust action that temporarily derates the largest beneficiaries.

Key Decisions for Investors

• Establish a 2.5% long position in NVDA (ticker: NVDA) over the next 6–12 weeks and hedge with a Jan 2027 350/550 call spread (cost-paid or small debit) to capture expected 18–30 month upside from AI compute demand while limiting downside.
• Add 1.5% positions in ASML (ASML) and TSM (TSM) to play semicap supply tightness; scale into weakness >10% and target a combined hold period of 12–36 months given multi-year capex cycles.
• Rotate 3–5% of consumer discretionary/white‑collar staffing exposure into cybersecurity: buy CRWD (1.5%) and PANW (1.5%) now, and increase by another 1% if there is a safety incident or regulatory procurement announcement within 90 days.
• Implement a relative-value pair: long NVDA (2%) / short MAN (1%) to express automation winners vs staffing losers; tighten stop on MAN if it rallies >15% on buyout rumors, and trim NVDA if it rallies >25% from average cost.
• Monitor regulatory catalysts: if EU AI Act enforcement or a major US federal procurement program is announced within 30–90 days, increase infrastructure/security exposure by 50% of the planned allocations; conversely, reduce AI infra exposure by 30% if an international export ban materially restricts chip shipments (assess within 7 trading days of announcement).