Why AI Is an Infrastructure Story
Artificial intelligence is often discussed in terms of models, algorithms, and software breakthroughs. But beneath every AI advance lies physical infrastructure: semiconductor plants, cooling systems, cables, logistics chains, substations, transformers, and power plants.
The countries and companies that understand this will design accordingly. The age of AI is also the age of energy realism. Nuclear belongs in that conversation not as a niche technology, but as one of the few systems capable of supporting the scale and reliability that the AI era demands.
Why AI Needs Energy Density
Data centers do not run on marketing language, and inference does not care about political narratives. AI requires power, large quantities of it, delivered continuously, with high reliability and strong power quality.
Intermittent electricity is not the same as dependable infrastructure. Compute clusters, industrial-scale cloud operations, and sovereign AI programs need power systems that can support them twenty-four hours a day. Nuclear energy fits that requirement better than most alternatives.
Kistora: Governing What Agents Are Allowed To Do
Jonas Helwig is co-founder of Kistora, an AI security company built around a specific shift: agents have moved from producing answers to taking actions. Once an agent can run a deploy, move money or read a production database, the interesting question is no longer output quality. It is permission.
Kistora sits in front of that moment. Every proposed action is evaluated against policy before execution and is then allowed, blocked or escalated to a human, with a tamper-evident receipt written to a signed, hash-chained ledger for each decision. The platform covers secret-boundary protection, credential use without possession, deterministic verification, workflow control and encrypted storage. The principle behind it is short: hold the proof, not the secret.
AI Inside Operational Systems
The other half of the AI work is unglamorous and much more concrete: putting AI into the daily machinery of an energy business. Sales pipelines, customer records, contract and metering data, billing questions, telephony, dispatch. These are the systems people use every day and complain about the moment they are slow.
The interesting problems there are rarely about the model. They are about latency, data shape, permission boundaries and what happens when a suggestion is wrong. An AI feature that saves ten seconds and is right 95% of the time is worse than useless if the remaining 5% costs an hour to unpick, which is why the governance work and the product work end up being the same work.
Measuring Models Instead of Guessing
In July 2026 he published a blind, peer-judged coding benchmark: six frontier models solved the same six problems, then graded each other anonymously, with no model allowed to score itself. Opus 5 swept all six domains. The method, the full heatmap and the caveats are published openly, including the finding that the model orchestrating the benchmark finished last.
AI Workflows and Product Thinking
The most interesting AI applications are not the flashiest, they are the ones that integrate deeply into real workflows, solve concrete problems, and make measurable differences in speed, quality, or cost. Jonas's perspective on AI is shaped by this product-level thinking.
What matters is not whether a system uses AI, but whether it delivers value reliably, at scale, in production environments. This pragmatic approach informs both his commentary and his building efforts.
Nuclear + AI Convergence
The convergence of AI and nuclear energy is not a distant hypothetical, it is happening now. AI needs power. Nuclear provides it. AI can optimize energy systems. Nuclear benefits from that optimization. The relationship is symbiotic and increasingly strategic.
Jonas Helwig writes and thinks at this convergence point, exploring how computation, capital, regulation, and electricity supply are becoming tightly coupled, and what that means for founders, investors, and policymakers.