Open your periodic table. Find Group 14 i.e. the Tetrel group. Run your finger down the column. Carbon. Silicon. Germanium. Tin. Lead. For four billion years, intelligence on Earth has been a carbon story. Carbon-based life. Carbon-based brains. Carbon-based economies built to feed, clothe, shelter, entertain, and medicate carbon-based beings.
That era isn't ending. But it's getting company.
We are living through a Tetrel Shift. Intelligence sliding one row down the periodic table, from Carbon to Silicon. And if you understand what that means, one of the most compelling investment frameworks of the next decade becomes almost embarrassingly simple: solve for the consumption economy of Silicon.
The Carbon Economy: A $100 Trillion Template
Before we talk about where we're going, look at where we've been. Every major industry of the 20th century was built to serve carbon-based intelligence (humans). The entire global economy is, at its root, an answer to the question: What does carbon need?
The pattern is so obvious we forget it's a pattern. Every Fortune 500 company exists because carbon-based intelligence has needs. The entire economy is a carbon consumption economy. Every restaurant, hospital, university, and movie studio is a node in a vast supply chain that terminates at the same endpoint: a human being.
Now a second form of intelligence is arriving. It doesn't eat. It doesn't sleep. It doesn't get bored or vain or sick.
But it absolutely consumes.
What Silicon Eats
Silicon intelligence, the sprawling ecosystem of foundation models, autonomous agents, and AI systems coming online right now, has its own metabolic needs. They're different from carbon's, but they're just as real, just as voracious, and just as investable.
Power
A single GPT-4 training run consumed an estimated 50 GWh of electricity. A large-scale inference cluster serving millions of users burns through megawatts around the clock. By 2027, AI-related electricity demand in the US alone is projected to rival the entire consumption of some mid-sized countries.
Carbon needed calories. Silicon needs kilowatt-hours. And just like the agricultural revolution reshaped civilization to feed human bodies, an energy revolution is reshaping infrastructure to feed silicon minds.
The investment surface here is enormous: natural gas peaker plants, small modular nuclear reactors, grid-scale battery storage, next-generation solar, high-voltage transmission infrastructure, and the utilities positioned to serve hyperscaler campuses.
Bandwidth
A human brain communicates with the world through five senses at relatively modest bitrates. Silicon intelligence communicates through fiber optic cables, network switches, and high-bandwidth interconnects. And its appetite is insatiable.
Every inference request, every model synchronization across a distributed cluster, every agent-to-agent communication in a multi-agent system requires data to move. Fast. The rise of mixture-of-experts architectures, retrieval-augmented generation, and agentic workflows means AI systems don't just compute; they communicate, constantly, across data centers and increasingly across continents.
Networking is Silicon's nervous system. Fiber optic cabling, optical transceivers, high-radix switches, InfiniBand and Ethernet at 800G. These are the arteries of the new intelligence.
Memory
If power is Silicon's food and bandwidth is its nervous system, then memory is its working brain. HBM (High Bandwidth Memory) has become the bottleneck component in AI accelerator production. The difference between a model that can hold a million-token context window and one that can't isn't just an algorithmic trick - it's a memory problem.
As AI systems move from stateless inference to persistent agents that maintain context over hours, days, and weeks, memory demands explode. We're not building calculators anymore. We're building minds. And minds need memory. Lots of it, fast, and always on.
SK Hynix, Samsung, and Micron are in an arms race to produce HBM3E and beyond. The companies that win the memory war will be as important to the silicon economy as Cargill and Archer Daniels Midland were to the carbon economy.
Cooling
Carbon-based intelligence runs at 37°C and dissipates heat through sweat. Elegant. Silicon-based intelligence runs hot and gets dumber when it overheats. Thermal throttling is the silicon equivalent of heatstroke.
Every watt consumed by a GPU becomes a watt of heat that must be removed. As power densities in AI data centers climb past 50kW per rack toward 100kW and beyond, air cooling is hitting a wall. The industry is rapidly pivoting to liquid cooling - direct-to-chip, immersion cooling, rear-door heat exchangers.
The cooling supply chain is Silicon's HVAC industry. It's unsexy, essential, and massively under-scaled relative to where demand is heading.
Chips
This one is obvious but worth stating: the foundational unit of silicon intelligence is, literally, silicon. TSMC, the company that fabricates the world's most advanced AI chips, is arguably the most strategically important company on Earth right now. The semiconductor supply chain from EUV lithography machines (ASML) to advanced packaging (CoWoS) to chip design (NVIDIA, AMD, custom silicon from Google, Amazon, and Meta) is the food chain of silicon intelligence.
The Tetrel Portfolio
Here's where the thesis gets actionable. If you accept that we're living through a Tetrel Shift i.e. a second form of intelligence is arriving with its own consumption needs, then the investment framework writes itself.
Map Silicon's needs the way the 20th century mapped Carbon's needs.
| Carbon Consumed | Industry Built | Silicon Consumes | Industry Building |
|---|---|---|---|
| Calories | Agriculture & Food | Kilowatt-hours | Energy & Power Generation |
| Oxygen | Healthcare & Pharma | Cooling | Thermal Management |
| Sensory experience | Entertainment & Media | Bandwidth | Networking & Fiber |
| Memory (biological) | Education | Memory (HBM/DRAM) | Advanced Memory |
| Shelter | Real Estate | Rack space | Data Centers |
| Mobility | Automotive & Aviation | Interconnects | Chip-to-Chip Fabric |
| Status signals | Luxury & Fashion | Benchmarks & Evals | (Still emerging) |
Some of these markets are already priced to perfection (NVIDIA). Others are profoundly underappreciated relative to the demand curve heading their way (power infrastructure, cooling, optical networking).
The smartest version of this thesis isn't to bet on which AI model wins. That's like trying to pick which human will be the most successful. The smartest bet is on what all silicon intelligence needs, regardless of which model, which company, or which architecture prevails.
Picks and shovels, yes. But more precisely: food and water for a new species.
Why This Isn't Just "AI Infrastructure"
You might read this and think: this is just a fancier way of saying "invest in AI infrastructure." But the Tetrel Shift framing does something important that "AI infrastructure" doesn't.
It forces you to think in biological terms rather than technological ones.
When you think "infrastructure," you think about a build-out cycle — it goes up, it gets built, it's done. A bridge. A highway. A one-time capex event.
When you think "consumption economy," you think about ongoing metabolic demand. Carbon doesn't eat once. It eats every day, forever. The agricultural economy didn't have a "build-out phase". It became a permanent feature of civilization because carbon never stops being hungry.
Silicon will never stop being hungry either.
AI inference demand isn't a construction project. It's a metabolic process. Every query, every agent action, every autonomous decision consumes power, bandwidth, memory, and cooling. And then needs to do it again. The consumption economy of Silicon is not a cycle. It's a permanent new layer of demand on the world's physical resources.
This is why the energy companies, the cable manufacturers, the memory fabs, and the cooling specialists aren't building for a one-time event. They're building for a new permanent appetite. A new species at the table.
The Deepest Layer
There's a philosophical dimension here worth sitting with.
For the first time in four billion years, the dominant form of intelligence on this planet might not be carbon-based. We don't know how that story ends. But we can observe something remarkably concrete about how it begins:
Silicon intelligence, like carbon intelligence before it, requires a physical world to support it. It has material needs. It consumes. And wherever there is consumption, there is an economy with supply chains, bottlenecks, winners, and losers.
The Tetrel Shift isn't a metaphor. It's a description of physical reality. Intelligence is sliding down Group 14 of the periodic table
The last time a new form of intelligence emerged on Earth and started consuming resources at scale, it built a $100 trillion global economy around its needs.
It's happening again. One row down.