“I had one of those chance airplane conversations recently—the kind that sticks in your mind longer than the flight itself.”
My seatmate was reading a book about artificial intelligence, and at one point, they described the idea of an “infinitely growing AI.” I couldn’t help but giggle a bit. Not at them, but at the premise.
An AI cannot be infinite. Computers are not infinite. We don’t live in a world where matter and energy are limitless. There aren’t enough chips, fabs, minerals, power plants, or trained engineers to sustain an infinite anything.
This isn’t just a nitpicky detail about science fiction. It gets at something I’ve written about before:
- In Who Really Pays When AI Agents Run Wild? I noted that scaling AI systems isn’t just about clever protocols or smarter algorithms. Every prompt, every model run, every inference carries a cost in water, energy, and hardware cycles.
- In The End of the Global Internet, I argued that we are already moving toward a fractured network where national and regional policies shape what’s possible online.
The “infinite AI” conversation is an example that ties both threads together. We may dream about global systems that grow without end, but the reality is that technology is built on finite supply chains. It’s those supply chains that are turning out to be the real bottleneck for the future of the Internet.
You can Subscribe and Listen to the Podcast on Apple Podcasts, or wherever you listen to Podcasts.
And be sure to leave me a Rating and Review!
The real limits aren’t protocols
When people in the identity and Internet standards space talk about limits, we often point to protocols. Can the protocol scale? Will a new protocol successfully replace cookies? Can we use existing protocols to manage delegation across ecosystems?
These are important questions, but they are not the limiting factor. Protocols, after all, are words in documents and lines of code. They can be revised, extended, and reinvented. The hard limits come from the physical world.
- Chips and fabs. Advanced semiconductors require fabrication plants that cost tens of billions of dollars and take years to build. Extreme ultraviolet lithography machines (say that five times, fast) are produced (as of 2023) by exactly one company in the Netherlands—ASML—and delivery schedules are measured in years.
- Minerals and materials. Every computer depends on a handful of rare inputs: lithium for batteries, cobalt for electrodes, rare earth elements for magnets, neon for chipmaking lasers, high-purity quartz for wafers. These are not evenly distributed across the globe. China dominates rare earth refining, while Ukraine has been a critical source of neon. And there is no substitute for water in semiconductor production.
- Power and cooling. Training a frontier AI model consumes gigawatt-hours of electricity. Running hyperscale data centers requires water for cooling that rivals the consumption of entire towns. When power grids are strained, there’s no protocol that can fix it.
- People. None of this runs itself. Chip designers, process engineers, cleanroom technicians, miners, metallurgists—these are highly specialized roles. Many countries are facing demographic changes that include aging workforces and immigration restrictions for the current tech giants and uneven education where the populations are booming.
You can’t standardize your way out of these shortages. You can only manage, redistribute, or adapt to them.
Geopolitics and demographics
The Internet was often described as “borderless,” but the hardware that makes it run is anything but. Supply chains for semiconductors, network equipment, and the minerals that feed them are deeply entangled with geopolitics and demographics.
No region has a fully independent pipeline:
- The US leads in chip design but depends on the Indo-Pacific region for chip manufacturing.
- China dominates rare earth refining but relies on imports of high-end chipmaking tools it cannot yet build domestically.
- Europe has niche strengths in lithography and specialty equipment but lacks the scale for end-to-end independence.
- Countries like Japan, India, and Australia supply critical inputs—from silicon wafers to rare earth ores—but not the whole stack.
This interdependence is not an accident. Globalization optimized supply chains for efficiency, not resilience. Each region specialized in the step where it had a comparative advantage, creating a finely tuned but fragile web.
Demographics add another layer. Many of the most skilled engineers in chip design and manufacturing are reaching retirement age. The same is true for technical standards architects; they are an aging group. Training replacements takes years, not months. Immigration restrictions in key economies further shrink the talent pool. Even if we had the minerals and the fabs, we might not have the people to keep the pipelines running.
The illusion of global resilience
For decades, efficiency reigned supreme. Tech companies embraced just-in-time supply chains. Manufacturers outsourced to the cheapest reliable suppliers. Investors punished redundancy as waste.
That efficiency gave us cheap smartphones, affordable cloud services, and rapid AI innovation. But it also created a brittle system. When one link in the chain breaks, the effects cascade:
- A tsunami in Japan or a drought in Taiwan can disrupt global chip supply.
- A geopolitical dispute can halt exports of critical minerals overnight.
- A labor strike at a port can ripple through shipping networks for months.
We saw this during the 2020–2023 global chip shortage. A pandemic-driven demand spike collided with supply chain shocks: a fire at a Japanese chip plant, drought in Taiwan, and war in Ukraine cutting off neon supplies. Automakers idled plants. Consumer electronics prices rose. Lead times stretched into years.
AI at scale only magnifies the problem. Training one large model requires thousands of specialized GPUs. If one upstream material is constrained—say, the gallium used in semiconductors—it doesn’t matter how advanced your algorithms are. The model doesn’t get trained.
Cross-border dependencies never vanish
This is where the conversation loops back to the idea of a “global Internet.” Even if the Internet fragments into national or regional spheres—the “splinternet” scenario—supply chains remain irreducibly cross-border.
You can build your own national identity system. You can wall off your data flows. But you cannot build advanced technology entirely within your own borders without enormous tradeoffs.
- A U.S. data center may run on American-designed chips, but those chips likely contain rare earths refined in China.
- A Chinese smartphone may use domestically assembled components, but the photolithography machine that patterned its chips came from Europe.
- An EU-based AI startup may host its models on European servers, but the GPUs were packaged and tested in Southeast Asia.
Fragmentation at the protocol and governance level doesn’t erase these dependencies. It only adds new layers of complexity as governments try to manage who trades with whom, under what terms, and with what safeguards.
The myth of “digital sovereignty” often ignores the material foundations of technology. Sovereignty over protocols does not equal sovereignty over minerals, fabs, or skilled labor.
Opportunities in regional diversity
If infinite AI is impossible and total independence is unrealistic, what’s left? One answer is regional diversity.
Instead of assuming we can build one perfectly resilient global supply chain, we can design multiple overlapping regional ones. Each may not be fully independent, but together they reduce the risk of “one failure breaks all.”
Examples already in motion:
- United States. The CHIPS and Science Act is pouring billions into domestic semiconductor manufacturing (though how long that act will be in place is in question). The U.S. is also investing in rare earth mining and processing though environmental and permitting challenges remain.
- European Union. The EU Raw Materials Alliance is working to secure critical mineral supply and recycling. European firms already lead in certain high-end equipment niches.
- Japan and South Korea. Both countries are investing in duplicating supply chain segments currently dominated by China, such as battery materials.
- India. This country has ambitious plans to build local chip fabs and become a global assembly hub.
- Australia and Canada. Positioned as suppliers of critical minerals, Australia and Canada are working to move beyond extraction to refining.
Regional chains come with tradeoffs: higher costs, slower rollout, and sometimes redundant investments. But they create buffers. If one region falters, others can pick up slack.
They also open the door to more design diversity. Different regions may approach problems in distinct ways, leading to innovation not just in technology but in governance, regulation, and labor practices.
Reframing the narrative
So let’s come back to that airplane conversation. The myth of infinite AI (or infinite cloud computing, for that matter) isn’t just bad science fiction. It’s a misunderstanding of how technology actually grows.
AI, like the Internet itself, is bounded by the real world. Protocols matter, but they are only the top layer. Beneath them are the chips, the minerals, the power, and the people. Those are the constraints that will shape the next decade.
Which leads us to the current irony in all of this: even as the Internet fragments along political and regulatory lines, the supply chains that support it remain irreducibly global. We can argue about governance models and sovereignty all we like and target tariffs at a whim, but a smartphone or a GPU is still a planetary collaboration.
The challenge, then, isn’t to pretend we can achieve total independence. It’s to design supply chains—local, regional, and global—that acknowledge these limits and build resilience into them.
Looking ahead
When I wrote about The End of the Global Internet, I wanted to show that fragmentation is not just possible, but already happening. But fragmentation doesn’t erase interdependence. It just makes it messier.
When I wrote about Who Pays When AI Agents Run Wild? I wanted to point out that scaling computation is not a free lunch. It comes with bills measured in electricity, water, and silicon.
This post ties both threads together: the real bottlenecks in technology are not the protocols we argue about in standards meetings. They are the supply chains that determine whether the chips, power, minerals, and people exist in the first place.
AI is a vivid example because its appetite is so enormous. But the lesson applies more broadly. The Internet is fracturing into spheres of influence, but those spheres will remain bound by the physical pipelines that crisscross borders.
So the next time someone suggests an infinite AI, or a fully sovereign domestic Internet, remember: computers aren’t infinite. Supply chains aren’t sovereign. The real question isn’t whether we can break free of those facts, it’s how we design systems that can thrive within them.
📩 If you’d rather track the blog than the podcast, I have an option for you! Subscribe to get a notification when new blog posts go live. No spam, just announcements of new posts. [Subscribe here]
Transcript
[00:00:29] Welcome back to The Digital Identity Digest. I’m Heather Flanagan, and today, we’re going to dig into one of those invisible but very real limits on our digital future — supply chains.
[00:00:42] Now, I know supply chains don’t sound nearly as exciting as AI agents or new Internet protocols. But stay with me — because without the physical stuff (chips, minerals, power, and people), all of those clever protocols and powerful algorithms don’t amount to much.
[00:01:00] This episode builds on two earlier posts:
- Who Really Pays for AI? — exploring how AI comes with a bill in water, electricity, and silicon.
- The End of the Global Internet — examining how fragmentation is reshaping the network itself.
Both lead us here: the supply chain is one of the biggest constraints on how far both AI and the Internet can actually go.
[00:01:27] So, if you really want to understand the future of technology, you can’t just look at the code or the protocols.
[00:01:35] You have to look at the supply chains.
The Reality Check: Technology Needs Stuff
[00:01:38] Let’s start with a story. On a recent flight, my seatmate was reading a book about artificial intelligence. Go him.
[00:01:49] At one point, he leaned over and described an idea of an infinitely growing AI.
[00:01:56] I couldn’t help but laugh a little — because computers are not infinite.
[00:02:04] There just aren’t enough chips, fabs, minerals, power plants, or trained people on the planet to sustain infinite anything. It’s not imagination — it’s physics, chemistry, and labor.
[00:02:20] That exchange captured something I keep seeing in conversations about AI, identity, and the Internet. We treat protocols as if they’re the bottleneck. But ultimately, it’s the supply chains underneath that constrain everything.
Chips, Fabs, and the Fragility of Progress
[00:02:38] Let’s break that down — starting with chips and fabricators, also known as fabs.
[00:02:44] The most advanced semiconductors come from fabrication plants that cost tens of billions of dollars to build — and take years, even a decade, to come online.
[00:02:56] And the entire process hinges on one company — ASML in the Netherlands.
[00:03:03] They’re the only supplier of extreme ultraviolet lithography machines. Without those, you simply can’t make the latest generation of chips. The backlog? Measured in years.
[00:03:21] Then there’s the issue of minerals and materials:
- Lithium for batteries
- Cobalt for electrodes
- Rare earth elements for magnets
- Neon for chipmaking lasers
- High-purity quartz for wafers
[00:03:44] These resources aren’t evenly distributed. China refines most rare earths. Ukraine supplies much of the world’s neon. And water — another critical input — is also unevenly available.
Power, People, and Production
[00:04:05] A frontier AI model doesn’t just use a lot of electricity — it uses gigawatt-hours of power.
[00:04:26] Running a hyperscale data center can consume as much water as a small city. And when power grids are strained, no clever standard can conjure new electrons out of thin air.
[00:04:26] Then there’s the people. None of this runs itself:
- Chip designers
- Process engineers
- Clean room technicians
- Miners and metallurgists
[00:04:57] These are highly specialized roles — and many experts are nearing retirement. Replacing them takes years, not months. Immigration limits compound the challenge.
[00:05:05] So yes, protocols matter — but the real limits come from the physical world.
Geopolitics and the Global Supply Web
[00:05:16] The Internet may feel borderless, but the hardware that makes it work is not.
[00:05:26] Every link in the supply chain is tangled in geopolitics:
- The U.S. leads in chip design but depends on Taiwan and South Korea for manufacturing.
- China dominates rare earth refining but still relies on imported chipmaking tools.
- Europe has niche strengths in lithography but lacks materials for full independence.
- Japan, India, and Australia provide key raw inputs but not the entire production stack.
[00:06:16] This global interdependence made systems efficient — but also fragile.
Demographics: The Aging Workforce
[00:06:21] There’s also a demographic angle. Skilled engineers and technicians are aging out.
[00:06:35] In about 15 years, we’ll see significant skill gaps. Even if minerals and fabs are available, we might not have the people to keep things running.
[00:06:58] The story isn’t just about where resources are — it’s about who can use them.
The Illusion of Resilience
[00:07:06] For decades, efficiency ruled. Tech companies built “just-in-time” supply chains, outsourcing to low-cost, reliable suppliers.
[00:07:21] That gave us cheap smartphones and rapid innovation — but also brittle systems.
[00:07:38] A few reminders of fragility:
- 2011: Tsunami in Japan disrupts semiconductor production.
- 2021: Drought in Taiwan forces fabs to truck in water.
- 2022: War in Ukraine cuts off neon supplies.
- 2020–2023: Global chip shortage reveals how fragile everything truly is.
[00:08:18] AI at scale only magnifies this fragility. Even one constrained resource, like gallium, can halt model training — regardless of how advanced the algorithms are.
The Splinternet Still Needs a Global Supply Chain
[00:08:48] Even as the Internet fragments into regional “Splinternets,” supply chains remain global.
[00:09:18] You can wall off your data, but you can’t build advanced tech entirely within one nation’s borders.
Examples include:
- A U.S. data center using chips refined with Chinese minerals.
- A Chinese smartphone using European lithography tools.
- An EU startup running on GPUs packaged in Southeast Asia.
[00:09:46] Fragmentation adds complexity, not independence.
The Myth of Digital Sovereignty
[00:09:46] The idea of total “digital sovereignty” sounds empowering — but it’s misleading.
[00:10:07] You can control protocols, standards, and regulations.
But you can’t control:
- Minerals you don’t have
- Fabricators you can’t build
- Workforces you can’t train
Designing Resilient Regional Systems
[00:10:14] So, what’s the alternative? Regional diversity.
Instead of one global, fragile chain, we can build multiple overlapping regional systems:
- U.S.: The CHIPS and Science Act investing in domestic semiconductor manufacturing.
- EU: The Raw Materials Alliance strengthening mineral supply and recycling.
- Japan & South Korea: Building redundancy in battery and material supply.
- India: Launching its “Semiconductor Mission.”
- Australia & Canada: Expanding refining capacity for critical minerals.
[00:11:38] Yes, these efforts are costlier and slower — but they build buffers. If one region falters, another can pick up the slack.
The Takeaway: Infinite AI is a Myth
[00:12:06] That airplane conversation sums it up. The myth of infinite AI isn’t just science fiction — it’s a misunderstanding of how technology works.
[00:12:17] AI, like the Internet, is bounded by the real world — by chips, minerals, power, and people.
[00:12:45] Even as the Internet fragments, its supply chains remain irreducibly global.
[00:13:02] The challenge isn’t escaping these limits — it’s designing systems that thrive within them.
Closing Thoughts
[00:13:27] The real bottleneck in technology isn’t protocols — it’s supply chains.
[00:13:48] AI is just the most visible example of how finite our digital ambitions are.
[00:14:13] So, the next time you hear someone talk about “infinite AI” or a “sovereign Internet,” remember:
- Computers are not infinite.
- Supply chains cannot be sovereign.
[00:14:19] The real question isn’t how to escape those facts — it’s how to build systems that can thrive within them.
Outro
[00:14:19] Thanks for listening to The Digital Identity Digest.
If you enjoyed the episode:
- Share it with a colleague or friend.
- Connect with me on LinkedIn @hlflanagan.
- Subscribe and leave a rating wherever you listen to podcasts.
[00:15:02] You can also find the full written post at sphericalcowconsulting.com.
Stay curious, stay engaged — and let’s keep the conversation going.