This report is largely written by people who build, fund, power or staff data centres. Martín Ramírez Riquelme is the first contributor who buys the outcome. He runs critical infrastructure and data centre at Agrosuper, a food producer where a delayed response on a production line is not a service-level breach but a quality failure. Asked whether production reality or available technology decides where digital infrastructure lives, he scores it at 80 out of 100 toward what the production process actually needs, and then spends eleven files on the sequence that follows from it.
01Contributed
Start at the wrong end and the process gets adapted to the solution.
“We want AI”, “we want to move to the cloud”, “we need edge”, and only then asking where it fits.
He calls that one of the most common mistakes. The right path runs the other way: understand the business value chain, examine what each production stage genuinely requires, and derive the operational requirements from that. Criticality, response time, continuity, availability, data sovereignty, cybersecurity and resilience. Only then does the architecture take shape, and only then the technology.
02Contributed
Access stopped being the advantage.
Access to technology is far more democratised than a decade ago. The competitive difference is no longer in reaching cloud, edge or AI platforms. It is in translating operational needs into a coherent architecture, while still recognising when a new capability creates a genuine improvement in the process.
03Contributed
Not every process needs milliseconds.
Reporting, planning and model training can tolerate seconds, minutes or hours. Others cannot: a delayed response, even where it is not an outage, can already degrade performance, quality or continuity. A machine-vision system doing quality control on a line may need local processing to avoid latency incompatible with the speed of the process.
Skip the question and decisions get made on cost, trends or preference. The result is workloads placed too far from the process that needs them, or edge infrastructure deployed where no operational requirement ever justified it.
04Contributed
Edge, core and cloud are one architecture, not three choices.
From the end user’s side the layers are not alternatives; they answer different operational needs. Strict requirements on response time, continuity or autonomy tend toward the edge. Value that comes from aggregation, training, planning or analysis sits better centrally or in cloud.
Mature architectures are hybrid in practice, not because hybrid is a trend but because production processes rarely share identical requirements. The common mistake is trying to solve everything with a single layer.
05Contributed
Both sides are right. They just start in different places.
One builds infrastructure to deliver a service; the other designs infrastructure to sustain a process.
He is explicit that this is not a knowledge gap. The supply side opens on availability, capacity, power, cooling, density and platforms. The industrial user opens on the production process and the operational consequence of that process failing. Both are valid, and they look at different parts of the same value chain.
What changes is the order. First understand the continuity the process requires, then size capacity. First define how data will be governed and protected, then select the AI model. First ensure the infrastructure can be maintained across its life, then pursue density.
06Contributed
Resilience is not a formula.
The reference point should not be the manufacturer’s catalogue. It should be the operational consequence of a failure.
Every process faces a different consequence when interrupted, and that consequence should set the redundancy. Oversizing ties up resources the process never required. Undersizing means discovering the true criticality of a process at the worst possible moment.
07Contributed
The hard part is outside the rack.
In a conventional facility the environment is a controlled condition. In an industrial operation it is a permanent design variable: humidity, corrosion, dust, vibration, washdown processes, thermal variation and operational constraints.
That reaches material selection, maintenance strategy and which technologies are appropriate at all. Equipment behaviour and useful life vary considerably by environment. Design cannot begin with the data sheet.
08Contributed
Six questions before AI is a technology discussion.
What continuity does the process require. What response time does it demand. What operational risk can it accept. What data exists and what is its quality. How will that data be governed and protected. How will the IT and OT environments coexist.
Answering those defines the architecture. Only then does it make sense to ask how AI creates value. In industry, sequence matters.
09Contributed
When it works, you cannot see it.
As long as infrastructure keeps pace with the process, it is almost invisible and production simply happens.
It becomes a constraint when capacity stops keeping pace with growth, when connectivity becomes a single point of failure, when maintenance requires the operation to stop, or when the environment accelerates degradation. In most cases the problem does not originate in the technology. It originates in an architectural decision not aligned with governance, made without sufficient understanding of the process and its effect on performance, quality, OEE or interruptions.
10Contributed
Two disciplines that need each other.
Colocation has developed extraordinary discipline around availability, standardisation, operations and continuous improvement, and he says that rigour is enormously valuable to industry. Industry brings the other half: availability is fundamental but it is a means, not the objective. Infrastructure decisions are ultimately judged by their effect on operations.
He declines to rank them. As architectures become hybrid and AI enters industrial operations, he argues the two perspectives need to converge. The best infrastructure is not the one with the most technology, but the one that strengthens the production process.
11Contributed
Chile: not a misalignment, an integration gap.
Chilean industry needs more than digital capacity. It needs an ecosystem able to support production processes that are increasingly dependent on data, automation and connectivity, and he says the investment arriving is aligned with a genuine need.
The gap he sees is in translation. Infrastructure is designed broadly and to standards; each production process has different requirements on criticality, response time, continuity and autonomy. And the growth does not rest on data centres alone: connectivity has to evolve at the same pace, and the power system has to carry demand from an economy that is electrifying as well as digitalising.
Contributed by Martín Ramírez Riquelme, Head of Critical Infrastructure and Data Center, Agrosuper, 10 August 2026, named with firm and title at the contributor’s request · passages inside quotation marks are reproduced from his written submission; his other positions are reported in the desk’s house style · framing and editorial selection are Entelligencia’s
