Who would expect a sex scandal, with a spy-story plot twist, in humanity’s quest for artificial superintelligence? Yet here we are.
This time it was not a rogue swarm of agents or another extinction scare. It may have been one of the oldest tricks in the book. A honeypot.
NVIDIA’s Jensen Huang has acknowledged an extramarital affair with a business executive whom U.S. intelligence reportedly suspects of passing information to Beijing. The company’s board has launched an independent investigation, saying it has no reason to believe national security or company secrets were compromised. The espionage allegations remain unproven.
The irony is hard to miss. We are building machines meant to surpass human intelligence, and the leader of perhaps the most strategically important company in that race may have been exposed through one of humanity’s oldest vulnerabilities.
Desire.
If the leadership of an AI powerhouse cannot hold to its own scruples, how do we trust it with decisions of far greater consequence? Climate change. Zero hunger. Public health.
The objective function of the AI factory
As a former chip designer, I worked inside the Power, Performance and Area (PPA) trade-off.
NVIDIA has historically pushed performance and scale, accepting higher power when throughput justified it. Energy efficiency matters, but increasingly it serves a bigger goal, getting the most AI computation and revenue out of a facility’s power budget.
NVIDIA’s own technical communications describe performance per watt as the defining metric of AI infrastructure, the conversion of electricity into revenue-generating intelligence.
These are different objectives:
Computational efficiency: tokens per joule.
Economic efficiency: revenue per megawatt.
Environmental sustainability: real societal value for minimal energy use and environmental impact.
In contrast:
Agriculture measures productivity in calories per hectare.
Medicine measures progress in healthy years of life.
Shouldn’t agentic AI be measured by its real-world impact per token, not just tokens per joule?
Time to challenge the AI industry’s power
NVIDIA acknowledges the rebound effect. Cheaper tokens unlock new applications whose demand can more than offset the energy efficiency gains.
The International Energy Agency projects global data center electricity use to nearly double, from 485 TWh in 2025 to 950 TWh by 2030. AI-focused data centers could triple theirs.
Chip-level power optimization is essential to curb the AI factory’s footprint. But efficiency cannot guarantee sustainability if every watt saved becomes a chance to produce more tokens.
Efficiency is not sustainability when the objective function rewards ever-expanding consumption.
Who decides what intelligence is worth?
Superintelligence promises breakthroughs in medicine, agriculture, energy and science. As an entrepreneur working with AI in the physical world, I have every reason to believe in that promise.
But look at the economics:
An AI application that keeps thousands of tons of food from spoiling can create enormous social value while consuming relatively few tokens. One producing synthetic entertainment at industrial scale may generate far more computational demand, and far more revenue.
Which contributes more to humanity?
Markets reflect purchasing power, not necessarily societal value. The same holds for healthcare in underserved communities, climate adaptation and other public goods whose benefits are hard to monetize.
Some of humanity’s most pressing challenges simply cannot be co-optimized with token revenue.
We need accountability for absolute environmental impact, not just performance per watt. And we need institutions that can weigh economic returns against societal priorities, whatever the personal virtues of the people in charge.
Alignment starts in the boardroom
A sex scandal reminds us that the architects of superintelligence remain human. Brilliant, and fallible.
The AI factory reminds us that brilliant engineering can still optimize a misguided objective.
Perhaps the first serious test of AI alignment has nothing to do with machines. It is whether we can align the people, companies and institutions building them with the future we claim to be pursuing.
That challenge is as old as humanity itself.
If you see it differently, write me at sotiris at centaur dot ag. I’d be glad to compare notes.


