Technology Is Becoming More Decentralized. Or Is It?

Technology is becoming more distributed, personal, and local. Buildings generate their own power. Microbes brew food. Vaccines are tailored to a single patient's tumor. After a decade defined by platform centralization — giant clouds, social networks, app stores — the frontier of commercially relevant innovation has shifted toward systems that work without intermediaries.

Or has it? The same week the World Economic Forum published its annual Top 10 Emerging Technologies report in Dalian, highlighting a pattern of decentralization across energy, medicine, and manufacturing, three of the world's largest companies announced data-center expansions that will each consume more electricity than a small country. The contradiction is not a bug in the analysis. It is the analysis.

What follows is a debate between two readings of the same evidence. One sees a century-long arc toward distributed systems. The other sees centralization masquerading as progress.

The case for decentralization

The WEF report identifies 10 technologies at the inflection point between research and scale. Seven of them operate on physical matter — extracting lithium, cooling buildings, destroying forever chemicals, brewing proteins, delivering medicine inside cellular particles. Only two are purely digital. This matters because physical infrastructure has historically been the most centralized domain of all: power plants, refineries, pharmaceutical factories, chemical facilities. The shift toward smaller, local, personalized physical systems is structurally different from anything the last century produced.

Consider everything-to-grid energy. Electric vehicles sit idle 95% of the day. Home batteries charge during solar peaks and discharge at night. Data centers have backup generators that run minutes per year. Everything-to-grid technology turns each of these from a passive consumer into an active grid participant. A single building becomes a power plant; a fleet of EVs becomes a virtual baseload generator. The economic logic is inexorable: distributed storage and generation already undercut new gas peaker plants in 12 US states.

Direct lithium extraction replaces evaporation ponds that take 18 months with engineered systems that produce battery-grade lithium in hours. Precision fermentation turns microorganisms into miniature factories that produce the same proteins as a dairy cow using a fraction of the land, water, and emissions. Personalized mRNA cancer vaccines, synthesized from a patient's own tumor mutations, train the immune system to recognize cells it had previously missed — Moderna's mRNA-4157 cut melanoma recurrence by 49 percent in Phase 3. Each of these technologies moves production closer to the point of consumption, reducing dependence on centralized infrastructure and long supply chains.

The pattern is not limited to the WEF list. Lattice-based cryptography, also featured, is designed to protect data against quantum computers — but its deeper logic is cryptographic sovereignty. Instead of trusting a certificate authority or a cloud provider, security is embedded in mathematical hardness that works anywhere, without a central validator. World models, another entry, are AI systems that build internal representations of physical environments, enabling robots and autonomous systems to operate without constant connection to a cloud brain. Local intelligence, local energy, local manufacturing, local medicine. The vector is unmistakable.

If the CAGR of these distributed technologies holds, by 2060 a significant share of basic energy, food ingredients, and diagnostic medicine could be produced at or near the point of use. That is a structural shift comparable to the move from mainframes to personal computers.

The case for concentration

The decentralization thesis tells a compelling story, but it selects evidence that fits the narrative while ignoring the forces pulling in the opposite direction. The same technological trends that enable a building to generate its own power also enable a handful of companies to control the software, trading algorithms, and grid interfaces that make that power valuable. Decentralization of hardware does not imply decentralization of control.

AI compute is the clearest example. Training a frontier model requires 10,000-plus GPUs running for months in a single facility. The capital cost of such a cluster — upwards of $4 billion — means that only a handful of organizations on the planet can participate. The cost of compute, as McKinsey estimates, will require $6.7 trillion in data-center build-out by 2030. That is centralization of a different kind: not of ownership, but of access. The individual running a small model on a laptop is not competing with the organization training GPT-7.

The cloud market tells a similar story. AWS, Azure, and Google Cloud together control 67 percent of global cloud infrastructure. The everything-to-grid energy systems described in the decentralization case above will depend on software platforms to coordinate when buildings charge and discharge, which markets they participate in, and how they respond to grid signals. Who writes that software? Who owns the data it generates? Decentralized energy nodes running on centralized cloud platforms is not decentralization — it is layered architecture. The physical layer disperses; the logical layer consolidates.

Data itself is undergoing a centralization wave. By 2026, estimates suggest that over 50 percent of online content is AI-generated. Training data for the next generation of models is increasingly scarce, pushing AI developers toward proprietary data sources — user interactions on platforms, sensor data from device ecosystems, clinical data from hospital partnerships. The organizations that control these data streams gain compounding advantages that smaller players cannot match. Decentralized production of physical goods does not counteract centralized ownership of the information those goods depend on.

The mRNA vaccine story illustrates the tension perfectly. Yes, personalized cancer vaccines represent a triumph of individualization. But the manufacturing infrastructure to produce them at scale is being built by exactly two companies — Moderna and BioNTech — in a handful of facilities. An mRNA vaccine costs roughly $200,000 per course. The sequencing and AI analysis required to design each patient's vaccine depends on computational resources concentrated in the same handful of cloud providers. Personalized treatment running on centralized infrastructure: the physical product is bespoke; the platform that enables it is anything but.

What the data shows

Both positions rest on real evidence. The question is which trend dominates over a 100-year horizon.

The table reveals something useful: in every domain, both forces are operating simultaneously. The question is not whether technology is becoming decentralized or centralized — it is whether the rate of decentralization in physical systems will outpace the rate of centralization in the software and capital layers that govern them.

One hundred years ago, electrification followed a similar pattern. Early adopters installed isolated generators in individual factories and homes — decentralized. Then utilities consolidated generation, built grids, and centralized control for half a century. Only in the last two decades has distributed generation begun to reemerge, enabled by solar, batteries, and smart inverters. The arc bent toward centralization first, then bent back. The same rhythm may be playing out now across a much wider set of domains — faster this time, because the enabling technologies are digital and compounding.

The WEF report is not wrong about the direction of physical innovation. But it tells only half the story. The other half is that every decentralized physical node creates a new centralized control point — the software that coordinates it, the data it generates, the capital that financed it. The technology is becoming distributed. The power over it is not.

Sources