Andrew Ng on AI as the New Electricity and Why Most Companies Are Doing It Wrong
Andrew Ng's 'AI is the new electricity' thesis is three years old now — here's what held up, what didn't, and what he's still getting right about adoption that most executives ignore.
Top Claims — Verdict Check
AI will transform every industry the way electricity did 100 years ago
🟢 Real“Representative of his position: just as electricity transformed every major industry, AI is a general-purpose technology that will reshape how every sector operates.”
The biggest AI opportunity is not in tech companies but in traditional industries
🟢 Real“Representative of his position: the largest untapped AI value lies in manufacturing, agriculture, healthcare, and logistics — not in Silicon Valley startups.”
Small data approaches and transfer learning make AI accessible to smaller companies
🟡 Partially True“Representative of his position: you do not need Google-scale data to build useful AI — transfer learning and data-centric approaches let small teams ship production models.”
AI Fund's venture model proves AI startups can be systematically created
🔴 Hype“Representative of his position: AI Fund builds companies by identifying industry problems first and applying AI solutions systematically, rather than waiting for founders to appear.”
AI education and literacy will be the bottleneck, not the technology itself
🟢 Real“Representative of his position: the constraint on AI adoption is not better algorithms — it is getting enough people trained to apply existing AI effectively.”
What's Real
The 'AI is the new electricity' analogy has aged well. McKinsey's 2023 report estimated AI could add $2.6-4.4 trillion annually to the global economy across industries. Manufacturing AI adoption grew 35% year-over-year through 2024 (Deloitte), and the biggest deployments — predictive maintenance at Siemens, quality inspection at Foxconn, crop yield optimization at John Deere — are in exactly the 'boring' industries Ng highlighted. His education bottleneck thesis is also confirmed: LinkedIn's 2024 workforce report showed AI-related job postings grew 3.5x faster than the supply of qualified candidates. DeepLearning.AI (Ng's own platform) and Coursera's AI courses enrolled over 10 million learners. The 'traditional industries' call was contrarian when he made it and has been proven directionally correct.
What's Hype
The 'small data' accessibility claim needs significant qualification. Transfer learning works well for computer vision and NLP classification tasks. It does not work well for domain-specific problems with unique data distributions — which is exactly what most traditional industry applications require. A manufacturing defect detection model still needs thousands of labeled defect images specific to that production line. Ng's AI Fund model — systematically creating AI startups — has a mixed track record. Landing AI, the most prominent AI Fund company, pivoted multiple times and has not achieved breakaway scale. The venture studio model for AI has not outperformed traditional VC on returns. The framing of AI as electricity also obscures a critical difference: electricity is a commodity input with no lock-in. AI models create deep vendor dependency, data moats, and switching costs that electricity never did.
What They Missed
The regulatory dimension is almost entirely absent. The EU AI Act (passed March 2024), the US Executive Order on AI (October 2023), and China's Interim Measures for Generative AI all impose compliance costs that disproportionately affect the small companies and traditional industries Ng wants to empower. A 50-person manufacturing firm cannot navigate AI compliance the way Google can. The energy cost of AI is also unaddressed — training and running large models requires massive electricity consumption, creating an ironic tension with the 'AI is electricity' analogy. Data centers consumed an estimated 1-2% of global electricity in 2023, and AI workloads are projected to double that by 2026 (IEA). The labor displacement question — what happens to the workers in those traditional industries when AI arrives — gets a brief mention about retraining but no honest engagement with the timeline mismatch between job loss (fast) and reskilling (slow).
The One Thing
The biggest AI returns are in industries that have never heard of Hugging Face — manufacturing floors, hospital supply chains, agricultural operations — and the bottleneck is trained people, not better models.
So What?
- If you sell AI tools to enterprises, Ng's framing is your sales deck: 'electricity for your industry' is the pitch that resonates with non-technical executives who control budgets
- AI literacy programs for your existing team will generate more ROI than hiring expensive ML engineers — the bottleneck is people who understand both the domain and the tool
- Regulatory compliance costs for AI are real and growing — factor EU AI Act and sector-specific regulations into your AI adoption business case, especially if you're in healthcare, finance, or manufacturing
Action Items
- 1Take Andrew Ng's free 'AI For Everyone' course on Coursera (6 hours total) and assign it to 3 non-technical leaders on your team — it is the single best resource for building organizational AI literacy without the hype.
- 2Identify the 3 most repetitive, data-rich processes in your business that no one has tried to automate with AI. Write a one-page brief for each: what data exists, what the current manual cost is, and what 'good enough' AI accuracy would look like. That's your starting backlog.
- 3Read the EU AI Act's risk classification summary (EUR-Lex, 20-minute read) and classify your current or planned AI features against it — even if you are not in the EU, this framework is becoming the global baseline for AI compliance.
Tools Mentioned
DeepLearning.AI
Ng's education platform — free and paid AI courses, strong on practical ML fundamentals
Landing AI
AI Fund's flagship company — data-centric AI platform for manufacturing visual inspection
Coursera
Hosts the original Machine Learning Specialization — still the most-enrolled AI course globally
Workflow Idea
Run an internal 'AI opportunity audit' quarterly. Gather one person from each department (ops, finance, marketing, product, support). Each person brings their top 3 most time-consuming repetitive tasks. Score each on: data availability (1-5), current manual cost per month, and estimated accuracy threshold needed. Rank by score. Pick the top 1-2 and prototype with existing tools (ChatGPT, Claude, simple Python scripts) before committing to a vendor or ML team. Most companies discover their best AI opportunities were invisible to the technical team because they live in operations, not engineering.
Context & Connections
Agrees With
- Fei-Fei Li on democratizing AI access beyond big tech
- Satya Nadella on AI as a platform shift comparable to cloud and mobile
Contradicts
- AI doomers who argue we should slow down all AI development and deployment
- Venture investors who claim AI value accrues primarily to foundation model companies, not application-layer builders
Further Reading
- McKinsey Global Institute — 'The Economic Potential of Generative AI' (June 2023)
- EU AI Act full text — EUR-Lex (search 'Artificial Intelligence Act')