The Efficiency Escape Velocity
If we view civilization through the lens of thermodynamics and information theory, it has always been a relentless search for efficiency: the ability to reduce entropy and create order with less energy, less time, and less human toil.
With the current drive for recursive self-improvement in AI systems, an optimistic view of the future is one of Universal Abundance. I realize that with wars raging globally and daily reminders of how efficiently we’ve learned to destroy one another, it is hard to stay focused on a "positive" future. But as I often tell my friends, I am a software engineer by vocation—and software engineers are optimists at heart. We wouldn’t survive the day-to-day debugging challenges otherwise.
We can map the progress of civilization by the efficiency of its primary inputs. Each "leap" was a jump in how much value we could extract per unit of effort:
Agricultural Revolution: Efficiency in calorie capture.
Industrial Revolution: Efficiency in kinetic energy (steam/oil replacing muscle).
Digital Revolution: Efficiency in information transmission.
The Intelligence Revolution (AI): Efficiency in reasoning.
As AI begins to automate the "OODA loop" (Observe, Orient, Decide, Act), the time between a human need and its fulfillment shrinks. When efficiency reaches a certain threshold—where the robot can build the robot, and the solar panel can power the mine—we hit what is known as the Abundance Threshold.
The Abundance Threshold is the theoretical tipping point where the cost of producing a core resource falls below the cost of measuring, billing, or restricting it. It is "mathematical escape velocity," where a resource transitions from a scarce commodity to a utility that is "too cheap to meter."
We’ve seen the prologue to this already. Just a few centuries ago, 75% of the population was engaged in agriculture; today, in the US, it is roughly 1.2%. Consider the Viking era: a 100m² sail took 600 days of labor to shear the sheep and prepare the wool, and another 285 days to spin the thread. That’s nearly three years of human life for one sail. It’s no wonder they turned to plundering; they were trying to escape the sheer drudgery of their own supply chain.
The "Abundance" movement argues that most social and political conflicts are "scarcity games." When the pie is fixed, we fight over slices. Efficiency increases the size of the pie so dramatically that the "cost of conflict" eventually becomes higher than the "cost of provision." Universal High Income (UHI) is a forward-looking concept, proposing a permanent unconditional income stream high enough to support a comfortable, middle-class lifestyle for all citizens. Driven by AI and robotic automation, this system aims to provide abundance and eliminate the need for work. It differs from Universal Basic Income (UBI) by targeting a much higher standard of living, potentially funded by taxing AI productivity or a social dividend.
However, the transition is the hard part. The tension between those building this vision—the ones pushing for data centers, nuclear acceleration, and Mars colonies—and those who can barely make ends meet today will play out in distressing ways.
I’ll be sharing more on the mechanics of efficiency in future notes. I’m currently deep in Brian Potter’s The Origins of Efficiency, which has led me down some fascinating rabbit holes. Looking back, so much of my career—from my early software engineer roles to the FAA—has been a search for these marginal gains. My former colleagues are probably just glad they no longer have to listen to my harangues about process improvements!
In AI News
NVIDIA’s GTC 2026 was interesting: from "AI as a research tool" to "AI as an industrial economy", NVIDIA is trying to deepen the moats. Beyond the hype, a few items of note:
The core theme was that the industry has moved past the training phase and into the inference phase, where models are deployed at scale to "run the world."
Vera Rubin Architecture: The successor to Blackwell, is scheduled for late 2026/2027. It isn't just a GPU; it is a "seven-chip" platform including the new Vera CPU, designed specifically for the sequential reasoning required by agentic AI. So NVIDIA is coopting the success of OpenClaw in full measure.
The Groq Licensing Deal: In a surprise move, NVIDIA integrated Groq’s LPU (Language Processing Unit) technology into its Rubin racks. This "specialized turbocharger" allows for ultra-low-latency token generation, addressing the speed bottleneck in real-time AI conversations. In another era the anti-trust folks would have taken a hard look at deals such as this.
NemoClaw & OpenClaw: NVIDIA is positioning itself as the "OS of AI Agents." NemoClaw is a secure enterprise framework that allows companies to deploy autonomous agents with built-in guardrails (using OpenShell sandboxing) to access databases and execute code safely. You throw in the word “secure” and your valuation goes up, I guess.
Physical AI & Robotics: The Isaac GR00T N1.7 model for humanoid robots is now commercially viable. Partnerships with Uber (for robotaxis in 28 cities) and industrial giants like ABB and KUKA show AI moving into physical manufacturing and transport. There will be more “robots” than humans in the not too distant future.
Space-1 Vera Rubin: NVIDIA announced plans for orbital data centers, extending their "AI Factory" vision literally beyond Earth.
I read NVIDIA’s GTC 2026 more as "aggressive paranoia," signaling a shift from selling individual chips to locking customers into a proprietary "AI Factory" ecosystem. By cannibalizing their own GPU-only roadmap—integrating specialized Groq technology for speed and launching the Vera CPU for reasoning—they are preemptively filling the gaps where competitors might have found a foothold. As they transition from hardware vendor to the foundational "OS of AI Agents," they are making the switching costs for enterprises astronomical. For the strategic leader, the takeaway is clear: NVIDIA isn't just leading the AI economy; they are architecting a physical and logical monopoly designed to make themselves indispensable as the world's primary AI utility.