Why EVs and AI Are Hitting the Same Infrastructure Wall
Software will only eat the world when the electrical grid allows it
I. Two Revolutions, One Constraint
Electric vehicles and artificial intelligence are often spoken about in wildly different tones. EVs are cast in climate morality tales, AI in dystopian sci-fi. But beneath the narrative gloss, both are engaged in the same battle: scaling past the prototype phase into societal ubiquity. And both are hitting the same wall. It is not consumer appetite. It is not investor enthusiasm. It is not even regulatory inertia. The bottleneck is simple, more primitve, and harder to solve: physical infrastructure.
For EVs, the limiting substrate is the charger network. For AI, it is electricity supply. They are mirror images of each other. Both are being throttled not by demand, but by supply chain and grid-level throughput. Both expose the deep asymmetry between software-scale ambition and physical-scale execution. And in both cases, the delusion persists that success as the early adopter level implies inevitable mass adoption. It does not.
II. Anecdotes vs Systems: The Mirage of Scalability
The most common mistake in both the EV and AI debates is confusing individual experience with system-level viability.
“My Tesla is great. I charge it at night in my garage.”
“My AI app is amazing. I’m running GPT-4 through an API.”
These are not invalid experiences. But they say little about the true scalability of the ecosystem. In both cases, the infrastructure that enables the early adopter experience cannot be linearly extended to the general population without massive, slow-miving, capital-intensive upgrades to real-world systems.
In other words: what works at 10% penetration fails at 90%.
III. EVs and the Charging Ceiling
The EV growth narrative runs into hard material limits:
Home charging is a class privilege. Only ~60% of Americans live in detached homes. Fewer have off-street parking. Urban renters are locked out.
Public chargers are expensive and unreliable. DC fast chargers cost $100,000+ per install. Maintenance is poor. Uptime is unreliable. Location density is sparse.
Grid readiness is uneven. Many cities cannot yet handle the load spikes of mass EV charging. Retrofitting is slow.
Tesla’s Supercharger network is a genuine moat, but it is still insufficient at scale. The company’s adoption of the NACS standard makes it a de facto substrate for the industry, but even that is more a workaround than a solution.
The upshot: EVs are fantastic if you are rich, suburban, and infrastructurally advantaged. The market ceiling is much lower than evangelists believe, unless massive charger rollout catches up, which is unlikely without new economic models for ROI on charger deployment.
IV. AI and the Electricity Wall
Now turn to AI.
Training frontier models and serving them at scale are not “just software” problems. They are thermodynamic problems:
Training a single GPT-4 class model requires megawatt-months of electricity. Multiple that across experiments, redundancy, and data center cooling.
Inference at scale (millions of users) requires sustained power draw and low-latency interconnects.
The US grid is not prepared. Interconnection queues for new power capacity are years long. Susbtations and transmission lines are backlogged. Most AI data centers need dedicated buildouts.
The AI revolution is colliding the the slow-motion reality of American energy infrastructure. Like EV chargers, AI datacenters need copper, concrete, permits, and grid cooperation. Unlike code, these things cannot be agile.
V. Capital-Intensity as Destiny
Both revolutions were sold as software stories. In fact, they are infrastructure plays.
EVs were imagined as consumer products with software-like margins. They are now revealed as grid-integrated machines reliant on real estate, power availability, and maintenance logistics.
AI wa imagined as a pure SaaS model. It is becoming a utility. GPU supply chains, colocation agreements, electrical engineering.
This is why Tesla and OpenAI are outliers: they are vertically integrated in a way that aligns software ambition with physical control. Tesla owns its Superchargers. OpenAI is partnering on its own datacenter infrastructure. Everyone else is a wrapper on top of someone else’s constraint.
VI. Where the Alpha Is: Own the Bottleneck
In both domains, the profits will not go to the wrappers. They will go to those who own the bottlenecks.
In EVs, that means Tesla (with chargers), utility-scale fleet charging platforms, and perhaps the companies building low-cost, high-throughput urban charging.
In AI, that means power providers, GPU landlords, and those with long-term access to low-cost electricity and grid interconnects.
This mirrors classic infrastructure arbitrage: the road is worth more than the car if you control the tolls.
VII. Market Mispricing and Strategic Errror
Investors are still mispricing these revolutions:
Valuations assume unconstrained adoption curves.
Charger companies are capital-starved because their returns are misunderstood.
AI apps are overfunded because their infrastructure dependencies are underappreciated.
This also explains why the smartest actors in the AI race—OpenAI, Amazon, CoreWeave, and increasingly, Microsoft—are building vertically into power infrastructure. They are not betting on software. They are betting on substrate control.
VIII. The False Hope of Modularity
A seductive but flawed idea pervades both spaces: that modularity will solve the constraint.
“Someone else will build the chargers.”
“Someone else will scale the grid.”
But no one wants to hold the bag on CapEx without guaranteed throughput. The ROI is ambiguous. This is a coordination failure, and it cripples both markets. The right analogy is not AWS. It’s railroads. Or ports. Or pipelines. These things don’t get built on the margin. They require vision, patience, and long-term capital.
IX. The Realist Forecast
Here’s what to expect:
EV adoption will plateau in urban areas without public charger breakthroughs.
AI model training will consolidate among those who can negotiate power at scale.
Electricity becomes the scarce commodity that limits AI advancement. The gating factor is not just GPUs, but megawatts and megahertz.
Most AI startups will die not because their business model is bad, but because they are built on top of infrastructural assumptions that no longer hold.
X. Conclusion: Software Eats the World only when the Grid Allows It
The 2010s taught us that software scales faster than institutions. The 2020s will teach us that infrastructure scales slower than ambition.
Electric vehicls and AI are not software businesses. They are grid businesses. Real estate and rights of way businesses. If you want to predict the adoption curve, follow the copper, not the code.
Software eats the world only if the world is plugged in.
The current infrastructure problem mirrors the cabling problem the IT industry had 20 years ago. They needed fiber everywhere, and no one had fiber anywhere. So the software companies bought and laid their own fiber networks, charging everyone else to use the highways they built. That problem is mostly solved, at least in principle. Pretty much all copper comms lines will be completely replaced with fiber by 2030, but it took twenty years to build that out.
Now the problem has changed from moving the copper comms grid over to a fiber comms grid into the problem of re-building the electrical grid, which, ironically, requires more copper.
But here's the thing - the people who subsidized rebuilding the comms grid are going to be the same people who subsidize rebuilding the electrical grid. And these are also the same people who will control the AI that these two grids combine to create. Finishing this process will take about 20 years, but not much more than that.
The IT industry is vertically integrating the whole of modern society.
All of it.
Power, comms, information, transportation, and the automated AI process to control it all, all of it.
This is the end of the nation-state.
In twenty years, citizenship will not be the coin of the realm, stockholding will be.
National governments will be, like churches, vestigial organizations that some people will continue to believe in and bow before, but that won't be the locus of power.