The X+1 Imperative: How the logic of the AI arms race threatens everything we hold dear
In the race for artificial intelligence supremacy, we face a deadly simple equation: if X = AI power, X+1 is always better than X. This isn’t just another technological competition — it’s an inexorable logic that, without intervention, will transform our entire planet into a development zone in service of compute and energy infrastructure.
While visiting family in Boise, Idaho, during this last Christmas, my brother-in-law, a real estate agent, told me about the two new massive data center projects being built just outside of the city: one by META (formerly Facebook) and another by Micron. He wasn’t sure if these were for AI, but my suspicions were confirmed when I looked them up.
Indeed, both are being built as massive new AI data centers (META) or chip fabrication for AI (Micron), in order to take advantage of very cheap power, mostly coal, in Idaho, but also solar power, a growing energy source around the country.
These two plants, at about $16 billion total, are just two of hundreds like them being built around the country — and many many more being built in other nations.
How many AI data and chip fabrication centers are enough? Well, under the “logic” of the new AI arms race, the answer is clear: we will never have enough. We must always build more. And more. And more. This essay is a follow up to my recent piece: “How AI and Bitcoin will eat the world.”
The Merciless Mathematics
The fundamental problem is brutally straightforward: if you have X amount of AI compute, your competitor wants X+1. If they achieve X+1, you need X+2. There is no natural equilibrium, no point where “enough is enough.” This creates an endless escalation with two fatal flaws:
- No Natural Ceiling: Unlike most technologies, there’s no point where more compute stops delivering benefits. Even minuscule advantages in capabilities can translate to decisive strategic edges. Each increment of compute continues to deliver value well past any reasonable limit.
- Winner-Take-All Dynamics: The first actor to achieve certain capability thresholds could permanently lock in their advantage, creating overwhelming pressure to race ahead at any cost. The risk of falling behind permanently makes restraint feel existentially dangerous.
Together, these create a perfect storm: a competition with no natural stopping point and existential stakes for falling behind. Without intervention, this logic leads to a planet fundamentally transformed.
The Path to Planetary Sacrifice
Without effective restraint, this logic unfolds with mathematical certainty as follows:
Stage 1: Acceleration (Now-2027)
- Exponential increases in AI compute deployment
- Intensifying competition for energy resources
- Security and safety measures sacrificed for speed
- Environmental and social concerns overridden
Stage 2: Transformation (2027–2035)
- Massive energy infrastructure buildout
- Forests cleared for solar farms and transmission lines
- Waterfronts dedicated to data centers and power plants
- Agricultural land repurposed for energy production
- Habitats destroyed at unprecedented scale
Stage 3: Domination (2035–2050)
- Every viable location turned to energy or compute infrastructure
- Society reorganized entirely around AI development
- Ecosystem collapse from habitat destruction
- Resources diverted from human needs to AI infrastructure
Stage 4: Endgame (Beyond)
- Planet-scale transformation to serve compute demands
- Remaining wilderness sacrificed for resources
- Climate systems destabilized from land use changes
- Human civilization subordinated to AI development imperatives
This isn’t dystopian fiction — it’s the logical endpoint of our current trajectory. The X+1 imperative, without constraint, consumes everything in its path.
Planetary Sacrifice Zone
Under this dynamic, Earth becomes a planetary sacrifice zone. Every river that could power a hydroelectric dam, every valley suitable for a solar farm, every coastal area that could host a data center — all progressively annexed to serve the insatiable appetite for more compute.
The calculation becomes brutally utilitarian: what generates more value — a forest ecosystem or the compute infrastructure that could replace it? A fishing ground or a coastal data center? Farmland or solar arrays? And since AI capabilities continue scaling with more compute, the economic and strategic logic always favors transformation.
This isn’t mere speculation. We’re already seeing:
- Power companies unable to meet data center demands
- Mothballed nuclear power plants being brought back online for AI data centers
- Environmental reviews expedited or bypassed in the name of an “energy emergency” to ensure “AI dominance”
- Land acquisition at unprecedented scale
- the new Trump administration removing what little attempts at restraint on full-throttle AI development that Biden had created
- Thought leaders like Leopold Aschenbrenner advocating “beat China” on AI at all costs (and of course China sees exactly the mirror image as it moves full-throttle to develope AGI/ASI)
The pressure has just begun, yet already normal constraints are buckling.
The Roots of the Problem
What makes this dynamic so intractable is that it emerges from the core incentives of multiple systems:
- Military Logic: Any potential edge in strategic capabilities must be pursued.
- Market Logic: Companies must maximize capability to capture market share.
- National Logic: Countries must maintain technological supremacy for security and prosperity.
- Technological Logic: More compute consistently delivers better performance.
When these four logics align — as they do in AI development — they create a force that overwhelms normal restraints. No single actor can unilaterally exercise restraint without risking their own destruction.
Two Paths to Restraint
There appear to be only two possible paths to escape this dynamic:
Path 1: Singleton Restraint
A single AI system achieves decisive strategic advantage and imposes restraint from above. This “benevolent singleton” could:
- Restrict further AI development
- Allocate compute resources optimally
- Enforce global limits on environmental impact
- Prevent competitive dynamics from resuming
This path avoids collective action problems but raises profound questions about autonomy and control. Who would oversee this singleton? What values would guide it? How could we ensure its continued benevolence?
Path 2: Human Cooperative Restraint
Alternatively, humanity could develop frameworks similar to those that constrained nuclear proliferation:
- Binding international agreements on AI compute limits
- Verification regimes to monitor compliance
- Domestic regulations on energy use for AI
- Shared safety standards for all systems
This path preserves human agency but requires unprecedented international cooperation. The nuclear treaty regime provides a template, but the winner-take-all dynamics make cooperation even more challenging.
Breaking the Cycle
Whichever path we choose, escaping this trap requires something typically rare in human affairs: coordinated restraint in the face of competitive pressure. But the stakes demand it. We need:
- Binding International Frameworks: Verifiable limits on AI compute deployment, similar to nuclear arms control regimes.
- National Regulation: Domestic controls on energy use for AI training and inference.
- Alternative Metrics: New ways to measure AI progress beyond raw compute and capabilities.
- Shared Safety Standards: Common frameworks ensuring all systems meet minimum safety requirements.
Most critically, we need recognition that this isn’t a classic tragedy of the commons — it’s worse. In a tragedy of the commons, actors deplete a shared resource for individual gain. In the AI arms race, actors pursue a trajectory that transforms our entire planet into a resource for computation.
A Final Choice
The X+1 imperative isn’t just another policy challenge. It’s an existential equation that, unaddressed, solves for planetary sacrifice. We have perhaps this decade — likely less — to change the mathematics.
Whether through collective human wisdom or benevolent singleton restraint, we must find a way to escape the merciless logic of endless escalation. If we fail, we won’t merely damage our planet — we’ll fundamentally transform it into a machine for serving AI development, with whatever remains of nature and humanity existing in the margins.
[Claude was used extensively in writing this essay]
