The real promise of AI isn’t fewer jobs, it’s cheaper thinking

The author — a serial founder who sold Freshly for about $1.5 billion and now leads Petfolk, a veterinary clinic platform backed by over $150 million and operating 36 clinics — argues that AI’s low measured returns reflect leadership and operational framing, not technology failure. He cites a PwC CEO survey (56% say AI has yet to deliver cost savings or revenue; ~12% report gains on both) and an MIT finding that only ~5% of integrated GenAI pilots deliver measurable financial value, and proposes reframing AI as ‘Synthetic Human Intelligence Hours’ (SHIH): a capacity-expanding input that compounds over time and requires redesigning workflows and governance. For investors, the implication is to favor companies that embed AI to expand analytical capacity and long-term compounding advantage rather than those pursuing short-term efficiency cuts.

Analysis

Market structure: Winners are platform and infrastructure providers that supply cheap SHIH at scale — think GPU/data-center owners and cloud stacks (NVIDIA, Microsoft, Google, AWS) and data-operational layers (Snowflake, Palantir). Losers are labor-heavy, per-seat analytics and legacy consulting/BPO models where high‑quality human hours are the product; pricing power shifts toward vertically integrated software/platform players. Supply/demand: expect sustained demand for GPUs, memory, copper and electricity over 12–36 months, with cyclical semiconductor supply risk; software demand concentrates on a smaller set of hyperscalers and data vendors. Cross-asset: equity premiums for AI-exposed names should rise, modest downward pressure on long-term inflation (helpful for duration) but near-term higher energy/commodity prices and semiconductor capex can raise cyclical inflation and corporate capex spend.

Risk assessment: Tail risks include accelerated regulation (EU AI Act enforcement, U.S. FTC/SEC actions) and liability from model failures or data breaches that could trigger multi‑quarter revenue hits; estimate 5–15% downside to valuations in stressed scenarios. Timeframes matter: immediate (days) = heightened volatility around earnings and GPU supply signals; short (3–12 months) = winners prove enterprise integrations; long (2–5 years) = compounding SHIH drives durable ROIC divergence. Hidden dependencies: clean labelled data, change‑management, power grid constraints, and semiconductor cycle timing—each can bottleneck ROI. Catalysts: major enterprise rollouts, hyperscaler pricing moves, or a patent/partnership (next 3–9 months) will accelerate adoption; adverse regulation or high-profile hallucination incidents could reverse sentiment quickly.

Trade implications: Direct plays — establish 3–5% core long in NVIDIA (NVDA) via 9–12 month call spreads (to capture GPU secular tailwind) and 2–3% long in Snowflake (SNOW) or Palantir (PLTR) for data ops exposure, dollar‑cost average over 90 days. Pair trade — long SNOW (2%) / short Accenture (ACN) (1.5%) as a relative‑value between platform monetization and labor‑intensive services over 6–18 months. Options — buy 6–9 month call spreads on MSFT/GOOGL to play cloud AI monetization with defined risk; add tail hedge (long 9–12 month puts) on consulting/BPO ETFs if regulatory headlines spike. Sector rotation — overweight semis, cloud infra, enterprise data; underweight traditional consulting, staffing and legacy per‑seat software for the next 12 months. Entry/exit — deploy initial tranches into any 5–10% pullback, add on positive earnings/partnership announcements, cut positions by 50% if market moves >20% adverse or if EU/US regulation materially restricts cross‑border data flows.

Contrarian angles: Consensus underrates the multi‑year compounding of SHIH — early P&L moves will look small but translate to persistent margin expansion for platform owners; this is analogous to cloud adoption 2010–2018 where multiples expanded before broad ROIC visibility. The crowd overprices quick efficiency wins and underprices durable capacity expansion; current mispricings create alpha in selected infrastructure and data‑ops names. Unintended consequences include labor pushback, unionization, and fragmented regulation that could localize data and create winner‑takes‑most national champions; monitor union/legislative activity and major jurisdictional data rules as contrarian signals over 6–18 months.

Key Decisions for Investors

• Establish a 3–5% portfolio long position in NVIDIA (NVDA) via 9–12 month call spreads (buy ITM or near‑ATM, sell higher strike to fund) to capture GPU/dc demand; scale in on any 5–10% pullback and target 18–36 month hold for secular thesis.
• Allocate 2–3% long to Snowflake (SNOW) or Palantir (PLTR) (choose based on valuation) to own the data‑ops layer that enables SHIH; accumulate on any >10% drop and expect asymmetric returns over 12–24 months as enterprise integrations materialize.
• Implement a pair trade: long SNOW (2%) and short Accenture (ACN) (1.5%) to express software/platform monetization vs. labor‑intensive services; target relative outperformance of +8–15% over 6–18 months and trim if spread narrows by 50%.
• Buy 6–9 month call spreads on MSFT and GOOGL (0.5–1% each) for cloud AI monetization exposure with defined downside; simultaneously purchase a 9–12 month 2% portfolio tail hedge (puts on a consulting/BPO ETF or ACN) to protect against regulatory/hallucination shocks.
• Establish 1–2% exposure to commodity/energy beneficiaries: long copper via FCX or copper futures and 1% long in a power/renewables ETF to hedge rising data‑center energy demand over the next 12–36 months; reduce exposure by 50% if GPU supply normalizes and datacenter build slows.