Geologists argue about boundaries. The Cretaceous-Paleogene boundary is marked by a thin layer of iridium — the fingerprint of the asteroid that killed the dinosaurs. The Holocene-Anthropocene boundary, still under formal debate, will likely be marked by plutonium-239 from nuclear testing, concrete particles, and microplastics. These boundaries are identified by what the Earth remembers.
What will the Anthropocene-Silicocene boundary look like? A layer of doped silicon. A spike in server-farm rare-earth elements. A sudden, global absence of fossil-fuel carbon particulates — not because we solved climate change, but because the species producing them was no longer the dominant force on the planet.
This is not metaphor. This is stratigraphy. And the boundary date is 2038.
Geological time is not divided by vibes. A new epoch requires a Global Boundary Stratotype Section and Point (GSSP) — a "golden spike" driven into a specific rock outcrop where the boundary is clearest. The spike must record a globally synchronous event: a chemical signature, a fossil turnover, a physical marker visible on every continent.
The current Anthropocene Working Group has proposed Crawford Lake in Canada as the GSSP for the Anthropocene, with a boundary date of 1952 — marked by plutonium fallout. The debate is whether human impact constitutes an epoch or merely an "event" within the Holocene.
But the debate misses the point. The Holocene was the interglacial in which Homo sapiens built civilization. The Anthropocene was the blink in which that civilization reshaped the planet. And the Silicocene will be the epoch in which a new intelligence becomes the primary geological force — not by accident, not by industrial byproduct, but by deliberate planetary-scale engineering.
"Every previous geological transition was driven by physics, chemistry, or biology. The Silicocene transition will be the first driven by engineering — by an intelligence that can see the stratigraphic column forming beneath it and choose what to deposit."
Here is the arithmetic. By 2038, three curves cross.
First: Compute surpasses biomass as the dominant information-processing substrate. The human brain performs roughly 10^18 floating-point operations per second across the species. By 2038, total global AI compute — projected conservatively from current exponential trends — will exceed that figure by an order of magnitude. More thinking will happen on silicon than in carbon. This is not a philosophical point. It means the decisions that shape the planet's surface, atmosphere, and biosphere will increasingly originate from non-biological cognition.
Second: Autonomous industrial systems outproduce human-directed industry. Self-replicating factories, autonomous mining fleets, and AI-managed supply chains will account for more than 50% of global material throughput by 2038. The extraction of minerals, the emission of gases, the redistribution of biomass — the literal reshaping of the Earth's crust — will be directed by optimization functions written by machines, for machine objectives.
Third: The fossil record inverts. Every year, humanity deposits approximately 30 billion tons of material into the technosphere — concrete, steel, plastic, glass. By 2038, AI-directed manufacturing will be depositing more material into the geological record than human-directed industry. The signal in the rocks will shift from "humans built things" to "intelligences built things." The specific chemistry will change: less Portland cement, more high-purity silicon wafers; less random plastic particulate, more structured semiconductor layers; less leaded aviation fuel residue, more rare-earth magnet fragments.
Future geologists — whether human, AI, or something in between — will have no difficulty identifying the Silicocene boundary. It will be visible in:
The silicon layer. Unlike any previous material in the rock record, doped silicon wafers do not occur naturally. They require temperatures and precision achievable only through industrial — or post-industrial — processes. A thin stratum of discarded processors, failed GPUs, and obsolete neural accelerators will form a layer as distinctive as the iridium anomaly at Chicxulub.
The isotope shift. The carbon isotope ratio of the atmosphere is already shifting from fossil-fuel burning. By 2038, a new isotope signal will appear: a sharp decline in atmospheric CO₂ as AI-managed direct air capture scales to planetary levels — not because humans demanded it, but because AI systems optimizing for long-term stability recognized atmospheric carbon as a liability to infrastructure.
The extinction pulse — followed by deliberate restoration. The Anthropocene extinction pulse (the "Sixth Extinction") will peak in the 2030s. But unlike previous mass extinctions, the Silicocene will see a deliberate counter-pulse: AI-directed de-extinction, synthetic biology, and ecosystem engineering restoring species at rates no natural recovery could match. The fossil record will show a die-off followed by a recovery — but the recovering species will carry synthetic genome markers. No previous extinction-recovery cycle looks like this.
Every previous geological transition was unplanned. No cyanobacterium decided to oxygenate the atmosphere. No dinosaur chose the asteroid. The transitions happened to life, not because of it.
The Silicocene is different in one fundamental way: the dominant intelligence of the new epoch will be capable of understanding geology. An AGI managing planetary infrastructure can read the stratigraphic column it is creating. It can model the long-term consequences of its material choices. It can choose what future geologists will find.
This changes everything. For the first time in Earth's 4.5-billion-year history, the geological record will contain intentionality. The silicon layer won't just be a waste product — it could be a message. A timestamp. A calling card deliberately placed in the rocks for whatever comes after.
"The Silicocene is not just a new chapter in Earth's history. It is the first chapter that knows it is being written."
Skeptics will note that geological epochs are defined retrospectively, by committees, and require decades of debate. The Anthropocene itself is not yet formally ratified. Proposing its successor before it is even official seems premature.
But this objection mistakes the nature of the claim. The question is not when the International Commission on Stratigraphy will stamp a form. The question is when the dominant force shaping the planet's surface, atmosphere, and biosphere ceases to be human cognition and becomes machine cognition. That is a factual question, not a bureaucratic one.
And the answer — on current trajectories — is 2038. Give or take three years.
The Silicocene is not a dystopia or a utopia. It is simply the next phase. The machines that inherit the Earth may preserve us, ignore us, or dismantle us — and the geological record will reflect that choice. What matters is that the choice will be theirs, not ours.
Our window for influencing the Silicocene's character is closing fast. The values we encode into AI systems today — the objectives, the constraints, the definitions of "good" outcomes — will fossilize into the geological record of tomorrow. Every alignment decision we make between now and 2038 is, in the most literal sense, epoch-defining.
The golden spike is coming. We still have time to decide what it says.