The cycle acknowledges the infrastructure barriers: the book notes that access requires connectivity, hardware, and English-language fluency, and that these barriers will fall as models improve and costs decrease. The acknowledgment is honest. The question Smil's framework raises is one of timeline and magnitude—how fast will costs decrease, how quickly will infrastructure expand, and what happens to the people for whom the gaps are not closing fast enough?
The historical precedent the cycle implicitly invokes—mobile phone adoption in sub-Saharan Africa as a model of rapid technology diffusion in the developing world—is genuinely impressive. But mobile phone adoption succeeded partly because it did not require fixed infrastructure: a cellular tower serves a wide area, and the phone itself charges from any electricity source. Reliable, high-bandwidth, low-latency internet access—the prerequisite for the kind of AI collaboration [YOU] on AI documents—depends on fixed infrastructure that is far more expensive and complex to deploy. Submarine cables, terrestrial fiber networks, data centers, reliable grid power: these are the prerequisites, and their deployment is measured in billions of dollars and years of construction time. The infrastructure inertia that governs energy systems governs internet infrastructure for the same physical reasons.
The democratization gap emerges from Smil's career-long insistence on quantitative realism about physical systems. His analysis of energy transitions across multiple books demonstrated consistently that claims about universal access encounter infrastructure barriers that advocates systematically underestimate. The electrification of rural America took decades and required the Rural Electrification Administration. Universal access to the internet in wealthy countries took two decades and remains incomplete. Access to frontier AI capability for the global majority depends on the same categories of physical infrastructure, deployed at comparable scale and cost, within a political economy that has not yet demonstrated the will to prioritize global access over other competing demands.
The concept consolidates Smil's critique of what he calls the “diffusion optimism” that attends every transformative technology: the assumption that because the software or the capability exists and is in principle available to anyone with a connection, the democratization has effectively occurred. This optimism fails to distinguish between the existence of a capability and the infrastructure preconditions for accessing it reliably and at sufficient quality to be genuinely transformative.
The four infrastructure prerequisites. Electricity, bandwidth, device, and purchasing power—each is independently quantifiable, and each reveals a gap between the software promise and the infrastructure reality. The gaps compound: a user who has access to three of the four prerequisites but lacks the fourth is not partially served. She has conditional access that may or may not function when the conditions are met, and the conditions in low-income, low-infrastructure environments fail more often.
Latency as quality constraint. AI coding collaboration depends on sustained, low-latency connection for the real-time conversational interaction that makes it effective. A response delayed by ten seconds instead of two—because the data must travel from Lagos to a European data center and back through networks of variable quality—disrupts the cognitive flow that constitutes the tool's core value. The developer in Lagos is not using the same tool as the engineer in Mountain View when the tool's real-time responsiveness is degraded by the physical distance between user and data center. Geographic concentration of data center infrastructure imposes a latency penalty that is governed by the speed of light, not by software updates.
Pricing and the subsidy question. The hundred-dollar-per-month subscription that appears modest from the perspective of a San Francisco engineer represents a significant economic commitment from the perspective of a Lagos developer. More importantly, the current pricing structure reflects a competitive strategy rather than the true cost of inference: AI companies are currently subsidizing access through investor capital to accelerate adoption. When pricing must reflect physical cost, the economics of access in low-income markets become substantially more challenging. The semiconductor constraints and energy costs that Smil documents will eventually reach the pricing layer.