The Infrastructural Cost
2026-07-15 · 5 min read · Janaina Maia
Three stories this week that don't seem connected. They are.
New York became the first US state to halt all new data center construction. Governor Kathy Hochul signed an executive order blocking permits for data centers 50 megawatts or larger. The moratorium lasts until the state completes an environmental review — roughly a year. Her words: 'Progress shouldn't arrive with a higher utility bill, deleted water supply, or noise pollution.' Two-thirds of Americans are concerned about data centers driving up electricity prices. One survey found people would rather have an Amazon warehouse in their backyard than a data center.
Apple shipped its redesigned Siri AI to the iOS 27 public beta. With 2.5 billion active devices, even a fraction of users installing the beta makes this the largest test of an AI assistant in history. The new Siri can access your emails, photos, messages, and whatever is on your screen. It has its own standalone app for the first time. It was built by distilling Google's Gemini model into smaller, efficient models that run on Apple Silicon. Every iPhone is now an AI device.
And ThoughtWorks published a piece on the zero-cost fallacy of open source in the agentic era. The thesis: we've confused permissive licensing with a license to exploit. Maintainers of load-bearing open-source packages are burning out, drowning in AI-generated pull requests, and facing harassment from the billion-dollar companies that consume their labor for free. The trust model of open source has been degraded. Libraries skyrocket to tens of thousands of GitHub stars within weeks based on AI-agent hype, despite having a three-week commit history.
Three stories. One question: who pays the infrastructural cost of AI?
Data centers are the new factories
We spent the last decade believing that software eats the world by being weightless. No warehouses, no supply chains, no smokestacks. Cloud computing was supposed to be the clean alternative to industrial production. That story was always a half-truth, and the other half is arriving now.
A data center is a factory. It consumes electricity on an industrial scale — the new ones exceed 500 megawatts, which is more than many small cities use. It draws water for cooling. It occupies land. It generates noise. And the products it manufactures — AI predictions, generated text, image synthesis — are consumed by people who never see the factory, never pay for its construction, and never deal with its waste.
New York's moratorium is the first time a government has said: we are not going to let this industry build faster than we can regulate. That's significant. But what's more significant is the Pew Research finding that only 10% of Americans are more excited than concerned about AI. Less than a quarter think it will have a positive impact on their jobs. The people who use AI products are not the same people who live next to data centers, and the gap between those two groups is where the politics gets explosive.
2.5 billion devices, 2.5 billion experiments
Apple's Siri launch is the biggest product design story of the week, and most of the coverage missed why. The story isn't that Siri got better. The story is that 2.5 billion devices just became AI instruments.
When you design a product for 2.5 billion people, you are not designing for power users. You are designing for the person who has never thought about AI, doesn't know what a large language model is, and just wants their phone to help them add a calendar appointment without five taps. That person is now using AI. They didn't opt in to an experiment. The experiment came pre-installed.
From a product design perspective, this is where the real work begins. Siri AI can read your emails, see your photos, and access your messages. That's an extraordinary surface area for helpfulness, and an equally extraordinary surface area for harm. Every incorrect suggestion, every hallucinated fact, every time it reads a text aloud that you didn't want read — those are design failures that happen at a scale no lab can test for.
The product question isn't whether Siri can handle questions. It's what happens when Siri gets it wrong in front of your child, your boss, your partner.
Open source's extraction problem
The ThoughtWorks piece is the most important one for product designers who work with AI, because it names something most of us would rather not think about.
When you use an open-source library, you are consuming the unpaid labor of someone who maintains it. That was always true. What's new is that AI agents have industrialized the extraction. They flood repositories with low-quality pull requests that maintainers have to review for free. They make it cheap to create the appearance of trustworthiness — viral stars, rapid commits — without the substance of earned credibility. And they enable the largest companies to wrap open-source code in proprietary products and capture the economic value while returning nothing.
This matters for product designers because every AI product you build sits on a stack of open-source dependencies maintained by people who may be one bad week away from quitting. When that maintainer walks away, your dependency becomes a vulnerability. Not a hypothetical one — a real one, with real security implications, and no one paid to fix it.
The ThoughtWorks article asks a question that product teams should be asking themselves: are we importing a 20,000-line library to solve a 200-line problem? And if so, are we prepared to own the security and maintenance lifecycle of that dependency? That's not a technical question. It's a product question. It's a design question. It's a sustainability question.
Why this matters for product design
- Infrastructure is a design surface. When New York halts data center construction, it's not just a regulatory story. It's a signal that the physical cost of AI is becoming politically visible. Product teams that ignore this — that treat compute, energy, and water as someone else's problem — will find that regulators have designed their constraints for them. The alternative is to design for efficiency now: smaller models, local inference, compressed prompts, and architectures that don't require a 500-megawatt building to serve a single user.
- Scale is a design discipline. Apple is about to learn what happens when you put AI in front of 2.5 billion people who never asked for it. The design failures will not be subtle. They will be failures of trust, failures of context, failures of vulnerability. The teams that succeed at this scale are the ones designing for the worst case — the confused parent, the elderly user, the person in crisis — not the best case.
- Sustainability is a dependency. Your AI product depends on open-source software maintained by people who are burning out. The ThoughtWorks piece describes a system where permissive licensing enables extraction and punishes correction. If you're building on top of that system, you have a dependency on its sustainability. The design question is: are you accounting for that dependency, or are you treating it as free and infinite?
My take
Every new technology goes through a phase where the infrastructure is invisible. The roads were someone else's problem. The power plants were over there. The maintainers were volunteers you never met. AI is exiting that phase.
New York just made the infrastructure visible. Apple just made the user base visible. ThoughtWorks just made the extraction visible. When all three become visible at the same time, the question changes. It's no longer can we build this? It's can we sustain this?
The answer depends on whether the people paying the infrastructural cost are the same people capturing the value. Right now, they're not. The people living next to data centers don't use the AI products. The maintainers propping up the software stack don't share in the revenue. The 2.5 billion people about to get Siri AI didn't ask for it, and they don't get a say in how it's designed.
Infrastructure is a design surface. Sustainability is a dependency. Scale is a discipline. The products that last will be the ones designed by people who understand all three.