What was then wrong with the golden age of internet wasn’t processing power at all; it was bandwidth. Then, it was mobile connectivity. Today, a far more existential shortage is looming in the United States, at least according to former Google CEO Eric Schmidt. It isn’t oil (though that’s a perennial worry), or even talent.
Schmidt’s admonition is stark and plain: America is running out of the infrastructure needed to fuel the Artificial Intelligence revolution. We are, specifically, running out of specialized chips (GPUs) in the same way that we’ve run out of miners to go dig for precious metals; the huge electrical grids required to juice them; and the physical land stretched by fiberoptics or otherwise optimized for data storage.
If artificial intelligence is the “new electricity,” as some tech luminaries say, Schmidt is for all intents and purposes warning that Oroville dam is already pushing the red line, and we haven’t even started bathing under the lights in the guest house.
The GPU Hunger: There’s More Than Just Silicon
When Eric Schmidt speaks of “running out,” he is referring to at the heart of the machine: H100s and Blackwell chips that Nvidia manufactures. The supply chain on these chips is more robust than during the post-pandemic crunch, but demand has grown inordinately.
We’re also no longer in the world of an era where a startup needs a few dozen servers to get an app running. In order to train a Frontier Model (such as the lineage of GPT-4 and Gemini) companies now need clusters of 100,000–300,000 GPUs that operate in perfect, liquid-cooled unison.
The Concentration of Power
Schmidt warns that this is leading to a “Compute Aristocracy”. If only three or four corporations in the entire world have the kind of money and access to these chips, the democratic promise of the internet starts to get rickety. We are not simply running out of chips; we’re running out of chips that the next generation of innovators can use. Without substantial new domestic manufacturing (through the CHIPS Act and beyond), America faces a bottleneck in which only incumbents — the “Big Tech” players that can afford such preposterously high stakes, outrageous as they are — will be able to participate.
Nvidia CEO Says Big Tech AI Spending Is Appropriate and Sustainable
The Power Grid Problem: Our 1950s Infrastructure Meets 2026 Standards
One of the most visceral points in Schmidt’s warning is that hum on the electrical grid. Training a single large language model can take as much energy as it does to power the typical American home for 20 years, according to a new estimate.
The “NIMBY” Obstacle
Data centers are not built with money alone — they need gobs of high-voltage electricity. The upshot, says Schmidt: the American electrical grid is old and in disrepair. It’s also balkanized, hamstrung by “Not in My Backyard” (NIMBY) red tape.
- Allowing for Purgatory: It can take almost a decade just to receive the necessary approvals for new high-capacity transmission lines.
- Baseload Demand: Solar and wind are wonderful, but AI requires around-the-clock “baseload” electricity. That’s why we’re seeing tech giants like Microsoft and Amazon desperate to bring back the dead nuke plants (e.g. Three Mile Island).
- The message from Schmidt is clear: unless we change how and where we allow and build out our energy infrastructure, the “AI Miracle” will remain literally in the dark. We are getting short of the patience to construct the backbone of the 21st century.
The “Intelligence Poverty” Gap
What TODO when a country runs out of compute? We move into Intelligence Poverty, again in Schmidt’s parlance. So in that case, the U.S. ends up leading on research but losing it for application. If a medical startup in Boston can’t get the compute power to simulate a new protein because some tech giant has hoovered up all of the local data center capacity for yet another gigaproject—say, training an AI to chatbot — that is a national security failure.
Schmidt asserts that AI is not an industry but a “force multiplier” across every single other industry — from defense to biology. And if we ever run out of the raw materials (chips and power) required to sustain this expansion, our global competitors, frequently far less inconvenienced by environmental regulations or private land rights, will just outbuild us.
I am a versatile content writer from the MP region, covering politics, business, crime, current affairs, entertainment, video games, and sports with clear insights, engaging analysis, and timely, reader-focused updates.
