COLUMN – In this column, three lean and technology experts respond to the same pressing question shaping today’s tech/AI debate. This month, we ask whether it is possible for countries to achieve digital sovereignty without destroying the environment.

Words: Erasto Meneses, Marie-Pia Ignace, and Theodor Panayotov

THE QUESTION

As countries pursue greater digital sovereignty, they face pressure to invest in data centers and AI infrastructure that consume enormous amounts of energy, water, land, and other resources. Is there an inevitable trade-off between technological independence and environmental responsibility, or can Lean Thinking help us find a different path that builds the digital capabilities countries need while minimizing their material and environmental footprint?

THE ANSWERS

The debate around digital sovereignty is becoming increasingly important. Countries want greater control over their data, AI capabilities, and critical digital infrastructure. But building that independence comes with a very physical cost: energy, water, land, hardware, capital, and human effort.

So are technological sovereignty and environmental responsibility inevitably in conflict? They don't have to be, but I do think we're asking the wrong question.

Throughout my career, from Toyota to technology companies and, more recently, working with organizations on their lean digital transformations, I've learned something simple: starting with the solution is a bad idea. Before asking how many data centers, GPUs, or AI models we need, I'd start with a more fundamental one: what problem are we trying to solve?

AI may appear intangible, but its infrastructure is anything but. Behind every AI interaction sits a physical value stream: chips, servers, electricity, water, cooling, networks, buildings, people, capital. Once you see that flow, a lean practitioner starts asking different questions. Where is the value? Where is the waste? What's underutilized? What are we scaling simply because the technology makes scaling possible?

This is where Lean can bring something useful to the sovereignty debate. Take Europe and Brazil — two very different starting points.

Europe is building at real scale. Brussels has opened tenders for up to seven AI "gigafactories," each expected to house at least 100,000 state-of-the-art processors — roughly four times as many processors as the largest AI centers currently operating in the EU — on top of the EU's existing network of 19 AI Factories. Combined investment is expected to exceed €30 billion. And the physical footprint is real: each gigafactory is expected to consume roughly as much electricity as a medium-sized town. That's the sovereignty bet in a nutshell: build the capacity first, and hope European demand shows up to fill it.

Brazil faces a different reality. As of early 2026, roughly 85% of Brazil's installed large-scale generation capacity came from renewable sources, mostly hydropower — well above the global average. Add to that land availability, natural resources, a large domestic market, and a fast-growing digital ecosystem, and it's easy to see why the country has become one of the most attractive destinations in the world for AI infrastructure money: projects currently underway could draw energy demand as high as 9,400 megawatts, equivalent to the consumption of 16 million homes.

But there's a danger here: abundance can hide waste. Renewable electricity is still a resource. Water still has alternative uses. Land still has value. Attracting billions of dollars in data-center investment doesn't automatically create Brazilian digital sovereignty — it can just as easily mean exporting cheap, clean power to serve someone else's compute needs.

One lesson from Toyota is still the right instinct here: go and see. In the AI era, our gemba has to expand beyond the data center itself. We need to see the whole digital value stream — energy, water, computing capacity, data, applications, human effort — and understand how it actually converts into customer and societal value.

Maybe the real opportunity isn't choosing between sovereignty and sustainability. It's designing a better system: improve utilization, cut unnecessary processing, experiment before you scale, and keep asking whether the next unit of resource is creating the next unit of value.

That's the real contribution Lean can make to this debate. The goal shouldn't be building the most infrastructure. It should be creating the most value with the least resource. That's not a limit on ambition; it's a sharper definition of it.

The theme of this column left me perplexed, because I remain of the view that Lean has little to contribute to it. For three reasons.

1. Sovereignty is not addressed through operational practices

The Fable 5 episode provided the proof of this mechanism. On 12 June 2026, access to the Fable 5 and Mythos 5 models was suspended outside the United States. The measure was lifted on 30 June, and access restored the following day.

This is what sovereignty risk is: a foreign country can withdraw or restrict a capability that has become critical, thereby creating political dependencies.

Companies suffer from this, of course, but they have a geographical freedom that countries do not.

Large companies naturally protect themselves against sovereignty risk. Toyota takes part in the Rapidus R&D program, created in 2022 as a direct response to the semiconductor shortage, with eight conglomerates — Toyota, Sony, NTT, NEC, SoftBank, Denso, Kioxia, MUFG — each having recognized that the national interest required a domestic alternative to Taiwan and Korea.

In parallel, it is deepening its partnership with NVIDIA to extend AI hardware and software beyond autonomous driving.

In short, Toyota handles sovereignty pragmatically, as a matter of physical supply chain (chips, network, embedded compute), through capital, R&D and the choice of alliances. Never through what happens at the workstation. Lean has no grip on that register.

2. The TPS has only two pillars, and that is not an oversight

Lean draws largely on the Toyota Production System, which has only two pillars: just-in-time and quality at every step. Respect for the environment is not directly translated into the operational mechanics of Lean.

Some will answer that hunting muda is frugal by construction: less inventory, less transport, less overproduction, therefore less environmental impact. But that is a side effect, not an objective.

The real reason for this absence is structural, and it deserves to be named. What JIT and jidoka have in common is that they are revelation mechanisms. Lowering inventory levels brings to the surface the problems that inventory was hiding. Stopping the line at the first defect makes that defect visible immediately. In both cases, the system turns an invisible, deferred quantity into a local, instantaneous signal. And both serve operational relevance and economic impact at once.

Neither the environment nor AI fits that mould: their impact is delocalized and deferred, so no andon can trigger and no gemba can reveal it. That's why the Prius could carry frugality in its specifications without the TPS ever absorbing it — engineering can design for invisible constraints; kaizen cannot act on them.

3. That leaves Hoshin

Hoshin leads executives, managers and employees to build “imposed” subjects into their thinking, and thereby steers both shopfloor kaizen and engineering. It is the only place in the system where the environment has genuinely taken hold.

“Toyota 2050” carries a very high ambition for reducing environmental impact. The group has also published  its “AI Principles,” attached to its ESG section. Under point 2, Sustainable Development, it states that beyond using AI to reduce environmental impact and improve resource efficiency, the group attaches importance to environmental considerations in the development and operation of the AI technologies themselves.

It is a statement of principle, with no figure and no method: no footprint measurement, no target. In other words: the subject exists in the direction-setting, not in the system.

_***

We are going to have to move past our indecision about AI — past the "digital magic" framing that says it will disrupt the world and stops there. What we need instead is a secure framework, built around sovereignty, economic competition and systems security.

In parallel, we will have to absorb this new capability into our operational practices. We are far from it. AI reproduces exactly the structural flaw that kept the environment outside the TPS: the cost of an inference is invisible at the point where it is called; the training cost was paid upstream, elsewhere, long before.

Hence what I expect from a third pillar — not an "environment" pillar, not an "AI" pillar, but the mechanism that makes consumption locally visible at the moment of use, the way kanban made inventory visible. Until that mechanism exists, we don't have a pillar. We have a policy.

The trade-off is real, but not inevitable. It becomes a problem when countries pursue sovereignty by copying the hyperscalers: the same chips, the same giant data centers, the same dependence on a handful of suppliers. Copy the architecture and you inherit the footprint.

Lean gives us a useful test to run here: capacity built ahead of demonstrated value is waste, whatever its size.

When Brussels first tested demand for AI gigafactories, 76 expressions of interest came back, imagining at least three million GPUs. The program itself moved from four planned gigafactories to five, and then to seven. By this summer, Bloomberg was counting roughly 10 likely bidders. When interest follows the subsidy faster than proven demand, a push system is born.

So the more useful question is: sovereign over what?

At Ethermind, we work with nuclear organizations where sensitive documentation cannot leave the site. The required compute fits in a rack in a room they already own, and much of the daily work can run on models fine-tuned on hardware that fits under a desk.

That matters because most organizations do not need frontier training. They need to use AI safely on their own data and preserve something much harder to buy than compute: the judgment of their own experts.

Chips can be purchased, but always through someone else's supply chain and on someone else's terms. Europe's gigafactories will still run largely on American-designed silicon. Institutional memory, engineering practice, and expert judgment are different. They are genuinely local capabilities, and they are worth protecting.

There is also a timing mismatch. GPUs are written off in a handful of years. Substations and buildings last for decades. At the same time, quantization, distillation and better architectures keep reducing the compute needed for a given task.

The sensible response is not to avoid infrastructure. It is to separate the slow layer from the fast one: build the grid, power and facilities that cannot be pulled just in time, then add silicon in increments as demand proves itself.

And Lean does not mean small. A rack sitting idle most of the day is waste, too. Sometimes the right answer is shared sovereign capacity. Europe already has 19 AI Factories.

The principle is simpler: put compute close to data that cannot move, share capacity where the data can move, reuse heat where it makes sense, and reserve the biggest facilities for workloads that genuinely need them.

If sovereignty means scale, environmental cost becomes part of the ticket price. If it means capability — running what matters, where it matters, with infrastructure sized to actual demand — sovereignty and sustainability can point in the same direction.

Sometimes that means a rack. Sometimes it means a gigafactory. What matters is that scale follows strategy, not the other way around.