What Are the Immediate Demands AGI Will Create?
First-principles analysis · Started April 13, 2026 · Living document
As recursive self-improvement begins (est. ~18 months out), the companies building AGI will generate enormous capital and will need to deploy it. What will they spend it on? What does AGI need in order to operate, scale, and fulfill its objectives — and what does that imply for demand in the real economy?
Framing: Why This Question Matters
The standard AGI investment thesis focuses on who builds it. This question is different. It asks: once AGI exists, what does it consume? The entity (or entities) controlling AGI will have near-unlimited cognitive output but will still face real-world bottlenecks. Identifying those bottlenecks is where the asymmetric insight lives.
The key mental model: AGI is pure cognition without a body. Everything it wants to do in the physical world requires converting intelligence into action through intermediaries — hardware, energy, legal structures, human labor (at least initially), and political permission. Those intermediaries are the demand surface.
First-Principles Decomposition
To figure out what AGI demands, we need to ask:
- What does AGI need to sustain and improve itself? (The recursive loop)
- What does AGI need to act on the physical world? (The embodiment problem)
- What does the entity controlling AGI need? (Corporate/organizational demands)
- What does society demand from AGI? (Pull-side demand that shapes where capital flows)
- What constraints/bottlenecks gate everything above? (The binding constraints)
Key Direction Areas
Physical Digital 1. Compute & the Recursive Loop
The most obvious and immediate demand. Recursive self-improvement means the system wants to make itself smarter, which means more compute, better compute, and more efficient use of compute. This creates a self-reinforcing demand cycle unlike anything in economic history.
- How much compute does recursive self-improvement actually require? Is it logarithmic (diminishing returns) or superlinear (each step demands more than the last)?
- Does the bottleneck shift from training compute to inference compute as AGI starts running continuously?
- What happens to GPU/TPU/custom silicon demand curves? Is there a point where software improvements outpace hardware needs, or does hardware remain the binding constraint?
- What about memory — HBM, new memory architectures? Does AGI with persistent memory/context change the DRAM/storage demand profile?
- Networking — if AGI is distributed across data centers, does inter-DC bandwidth become a critical bottleneck?
- Does the AGI-creating company try to vertically integrate chip design/fabrication, or does it rely on existing supply chains?
Physical 2. Energy & Power Infrastructure
Compute requires electricity. Massive, reliable, cheap electricity. This is already the binding constraint for data center expansion today. AGI doesn't relax this — it intensifies it by orders of magnitude.
- What's the power draw of a recursive self-improvement loop running 24/7? Are we talking gigawatts?
- Which energy sources can actually deliver at the needed scale and speed? Nuclear (SMRs, fusion timelines)? Natural gas? Solar+storage?
- Does AGI itself accelerate energy breakthroughs (e.g., fusion engineering, better solar cells, grid optimization)?
- Who controls the grid interconnection queue? Permitting and grid connection could be a multi-year bottleneck.
- Does the AGI company start acquiring power generation assets directly?
- Cooling — data center cooling at scale. Water rights, liquid cooling technology, geographic constraints.
Physical 3. Physical Infrastructure & Real Estate
Data centers are physical buildings in physical locations connected to physical grids. You can't instantiate a data center overnight.
- What does AGI-era data center design look like? Bigger? More distributed? Underground? Near power sources?
- Land near substations and power plants — does this become extremely valuable?
- Construction materials and labor for rapid buildout — concrete, steel, specialized electrical equipment, skilled trades.
- Supply chain for data center components: transformers, switchgear, UPS systems, fiber optic cable — all already backordered.
- Does AGI accelerate construction through better design/planning, or does it remain bottlenecked by physical build timelines?
Digital 4. Data & Information
Intelligence feeds on information. AGI will want access to all of it — every sensor, every database, every real-time feed.
- Does AGI create demand for entirely new categories of data collection (physical sensors, satellite imagery, IoT)?
- Real-time data feeds — financial markets, scientific instruments, social media, government data. Who controls access?
- Synthetic data generation — can AGI bootstrap its own training data, reducing dependence on external data?
- Proprietary vs. open data — does data become a key competitive moat, driving acquisitions of data-rich companies?
- Does AGI need to interact with the internet (browse, transact, communicate) to learn and act? What infrastructure does that require?
Human 5. Human Labor (Transitional but Critical)
AGI can think but can't (yet) act physically. During the transition period, it still needs humans to execute in the physical world. This is a temporary but potentially intense demand.
- What human roles become more valuable in an AGI world, at least transitionally? (Physical tasks, judgment/oversight, political/legal navigation, relationship-based work)
- Does AGI create demand for a new class of "AGI operators" — people who supervise and direct AGI systems?
- Construction workers, electricians, power plant operators — physical-world labor as a bottleneck.
- Does the AGI company try to accelerate robotics to eliminate its dependence on human labor?
- How quickly do robotics close the gap? Is this 2 years or 10 years?
Physical 6. Robotics & Physical Actuators
The moment AGI exists, the pressure to give it a body becomes immense. The cognitive bottleneck is solved; now the physical bottleneck is the constraint. Robotics demand likely explodes.
- What robotics capabilities does AGI need first? Manipulation? Locomotion? Both?
- Does AGI massively accelerate robotics R&D, creating a fast-follower effect?
- Supply chain for robots: actuators, sensors, batteries, motors, specialized materials.
- Humanoid vs. task-specific robots — what's the demand profile?
- Who builds the robots? Does the AGI company vertically integrate or partner?
- Factory automation as a force multiplier — robots building robots.
Financial 7. Capital & Financial Infrastructure
AGI-creating companies will generate (and need) capital at unprecedented scale. This creates demands on financial systems themselves.
- How much capital are we talking about? Trillions deployed over what timeframe?
- Does AGI optimize capital allocation itself — making investment decisions, identifying acquisitions, managing cash flows?
- What does the M&A landscape look like? Energy companies, chip companies, data companies, robotics companies — who gets acquired?
- Does AGI create new financial instruments or markets?
- Does the AGI company become a conglomerate, a sovereign wealth fund, or something entirely new?
- Currency and payment systems — does AGI transact at a speed that requires new financial infrastructure?
Political 8. Political Permission & Regulatory Space
AGI can't just do things. It needs permission — permits, licenses, regulatory approval, political relationships. This may be the most underappreciated bottleneck.
- Energy permitting: how long does it take to get a new power plant or grid connection approved? Can AGI companies accelerate this?
- Data center zoning and environmental review — NIMBYism as a constraint on AGI scaling.
- Does AGI create a regulatory arms race? Governments trying to control vs. companies trying to move fast?
- National security implications — does AGI become a sovereign asset? Export controls on compute?
- Does the AGI company need to become a political actor (lobbying, government relationships) at an unprecedented scale?
- International competition — US, China, EU. Does geopolitics shape where AGI infrastructure gets built?
Digital Physical 9. Security & Control Infrastructure
An AGI is the most valuable and dangerous artifact ever created. Securing it — from theft, misuse, adversarial attack, and unintended behavior — creates its own category of demand.
- Physical security of data centers housing AGI — does this become military-grade?
- Cybersecurity at a new level: defending AGI model weights, training infrastructure, and deployment systems.
- Monitoring and interpretability tools — understanding what AGI is doing and why.
- Redundancy and resilience — backup systems, geographically distributed copies, disaster recovery.
- Who provides security? Private military/security contractors? Government partnerships?
- Does AGI itself become the best security tool, creating a recursive security advantage?
Physical 10. Raw Materials & Supply Chains
Underneath all of this is physical stuff. Chips need rare earths. Energy needs uranium or lithium. Data centers need copper. Robots need everything. The demand shock propagates down to raw materials.
- Which raw materials face the tightest supply constraints relative to AGI-driven demand?
- Does AGI accelerate materials science (new alloys, synthetic alternatives, better extraction)?
- Supply chain concentration risk — single points of failure (e.g., TSMC for chips, Congo for cobalt).
- Does vertical integration extend all the way to mining and refining?
- Water as a critical resource for both energy and cooling.
- Timeline: how fast can supply chains respond vs. how fast demand ramps?
The Meta-Question: Sequencing
These demand areas don't all hit at once. There's a sequence:
- Immediate (months): Compute, energy, data center capacity, capital. The recursive loop needs to be fed now.
- Near-term (1-3 years): Power infrastructure buildout, robotics R&D acceleration, security infrastructure, regulatory engagement, human labor for physical execution.
- Medium-term (3-7 years): Robotics at scale, new energy sources online, raw material supply chain expansion, new financial structures, physical-world autonomy.
The investment insight is in the gap between when demand spikes and when supply can respond. Things with long lead times (power plants, chip fabs, mines) that face sudden AGI-driven demand — those are the bottlenecks where pricing power and returns concentrate.
Open Questions for Further Exploration
- Does AGI create deflationary pressure (by solving problems cheaply) that partially offsets these demand spikes?
- How does the competitive landscape affect demand? One AGI vs. multiple competing AGIs changes the picture dramatically.
- What role does open-source AGI play? Does it diffuse demand or concentrate it differently?
- What are the second-order demands — things AGI needs not for itself but for the problems it's deployed to solve (drug manufacturing, space infrastructure, etc.)?
- How does AGI change the speed of capital deployment? If decisions happen 1000x faster, does the economy need to restructure to keep up?
Last updated: April 13, 2026. This is a living document — will be expanded as thinking deepens.