When Electricity Confused Everyone
In 1890, electricity was the most exciting technology on earth. Investors poured fortunes into “central station” power plants—giant facilities generating direct current (DC) for a few city blocks. By 1900, over 3,000 electric utilities had been founded in the United States alone. By 1910, most had gone bankrupt.
Why? Because they misunderstood the deployment curve. Early adopters simply replaced steam engines with electric motors in the same factory layout. That yielded zero productivity gain. It wasn’t until factory managers realized they could reorganize production around the motor—putting workstations in a straight line instead of clustered around a central shaft—that productivity exploded. Between 1910 and 1920, manufacturing output per worker nearly doubled.
The lesson: General Purpose Technologies (GPTs) take 20–30 years to show up in productivity statistics. The first decade is “installation” (hype, infrastructure, disappointment). The second decade is “deployment” (reorganization, process innovation, value creation).
The Modern Parallel: Generative AI in 2025
Today, the market is fixated on Large Language Models (LLMs). Every board presentation mentions AI. But as of 2025, less than 10% of mid-market Asian companies have successfully integrated generative AI into core operations beyond basic chatbots and document summarization.
The hot topic among LPs is “AI-driven EBITDA expansion.” The historical reality is that most companies will see costs increase before they see savings. Data hygiene, legacy system integration, and talent acquisition are expensive. In our portfolio, we have seen a consistent “AI J-curve”: year one costs rise 8–12%, year two productivity flatlines, year three margins expand 15–20%.
Three Historical Lessons for Private Equity
1. Avoid the “Pick and Shovel” trap after the peak
In the California Gold Rush (1849), only two fortunes were made sustainably: Levi Strauss (clothing) and the shovel manufacturers. The miners themselves lost everything. In AI, the shovel sellers are the cloud compute providers and chip designers. Their multiples already reflect perfection. For a PE firm, the real alpha lies in applying AI to fragmented, analog industries—logistics, legal services, specialty manufacturing, healthcare administration. These sectors have low tech adoption, high labor costs, and enormous data silos.
A concrete example: freight brokerages in Hong Kong’s Pearl River Delta region. Most still use manual rate sheets and phone calls. A targeted LLM that ingests real-time shipping rates, customs delays, and fuel prices can reduce bid-to-booking time from 4 hours to 4 minutes. That is not an AI company. That is a logistics company using AI. The valuation multiple is half, and the exit pathway is clear.
2. The productivity paradox demands operational patience
Economist Robert Solow famously said in 1987: “You can see the computer age everywhere except in the productivity statistics.” That was 17 years after the first microprocessor. AI is following the same timeline. For a PE firm with a 5–7 year hold period, this is critical: you cannot deploy AI in year 4 and expect an exit lift. You must implement in year 1 or 2, absorb the pain, and harvest in year 5.
We use a three-layer deployment framework:
-
Layer 1 (Year 1): Data standardization. Kill spreadsheets. Move to unified cloud data lakes.
-
Layer 2 (Years 2–3): Process re-engineering. Remove middle-management layers that AI can replace. This is the painful political phase.
-
Layer 3 (Years 4–5): Autonomous decision-making. Inventory reordering, dynamic pricing, customer service triage.
3. The labor arbitrage window is closing fast
Historically, technology-driven labor arbitrage lasts about 7–10 years before wages adjust. When the mechanical loom arrived in 1810s England, textile wages fell 30% over a decade, then rose again as labor shifted to higher-skilled roles. The same will happen with AI. The PE opportunity is to capture that productivity spread before competitors enter.
Hong Kong has a unique advantage: it sits between mainland China’s AI engineering talent (Shenzhen’s tech corridor) and ASEAN’s labor-intensive industries. A HK-based PE firm can deploy AI solutions built in Nansha into Vietnamese garment factories or Thai auto parts plants, with legal and financial structuring done in the SAR. No other jurisdiction offers that triple access.
Like electricity in 1895, AI in 2025 is real, inevitable, and currently overhyped on price. The winning PE firms will not be those who buy the “AI pure plays” at 30x revenue. They will be those who buy boring, analog businesses and spend 36 months turning them into digital-first platforms. History is unambiguous: the deployment phase creates 5x more wealth than the installation phase. We are just entering the deployment phase.
