China is killing the US on energy. Does that mean they’ll win AGI? – Casey Handmer

The race for AI dominance is fundamentally an industrial race, hinging on the ability to manufacture and deploy GPUs, batteries, and energy infrastructure at an unprecedented scale. While China appears to have a significant lead in manufacturing, particularly with a 20x advantage in solar panel production, the United States is not destined to lose. The notion that China’s business environment is superior is questionable, given issues with capital allocation (e.g., high-speed rail), lack of rule of law, and mandatory CCP oversight. The US possesses critical advantages in cheaper natural gas, advanced automation, and superior financial capacity.

If sufficiently motivated, akin to a wartime effort, the US could onshore and scale its own solar manufacturing pipeline—from raw materials to finished modules—in approximately two years. The key bottleneck for AI development is not energy infrastructure but the availability of high-end GPUs. Therefore, even if Chinese solar panels were hit with a 200% tariff, they would still be a bargain for data centers, as energy costs are a small fraction of the total expenditure.

Why Solar Will Power the AI Revolution

While current hyperscale data centers are opting for natural gas, this is a short-term strategy driven by the immediate need for speed. Companies like xAI are tapping into existing gas pipelines because it’s the fastest way to bring massive, gigawatt-scale data centers online. However, this approach will soon hit hard constraints:

• Supply Chain Bottlenecks: The manufacturing capacity for essential components like gas turbines and large transformers is limited, with production slots already booked out to nearly 2030.
• Grid Competition: These new data centers are competing for power with legacy residential and industrial users, causing price spikes in regional grids like PJM.
• Learning Rates: The most crucial factor is the economic learning curve. Natural gas technology is mature, with minimal cost reductions over time. Solar, in contrast, is on an astonishing trajectory governed by Wright’s Law.

“The Wright’s Law coefficient is 43%. So every time we double cumulative production, we get a 43% reduction in cost.”

This means solar production is not only growing but accelerating, consistently defying predictions of saturation. The demand elasticity is so high that every price drop unlocks a massively larger market. For a bank deciding whether to finance a new gas turbine factory or a solar factory, the choice is clear. The long-term viability of gas turbines is highly uncertain in a world where solar and battery costs are in freefall. It is predicted that by 2027, the majority of new data centers breaking ground will be predominantly solar-powered.

The Blueprint for Off-Grid, Solar-Powered AI

The future of AI compute will not be on the traditional grid. The grid is plagued by Baumol’s cost disease—it is incredibly expensive and slow to build and maintain transmission lines due to regulatory hurdles, land acquisition, and labor costs. The solution is to build massive, self-contained, off-grid data centers on private land.

Here’s what a large-scale solar data center looks like:

• Scale: A 5-gigawatt data center requires approximately 50,000 acres of land for solar panels. While this sounds vast, it’s comparable in scale to historical government projects like the Hanford Site for the Manhattan Project.
• Location: The ideal location is in places with abundant, cheap land and permissive regulations, such as Texas or other parts of the American West. Land availability is not a real constraint in the US.
• Architecture: The facility is a self-contained computational world. It consists of fields of solar arrays, batteries for storage, and the data center racks themselves. The only connection needed to the outside world is an optical fiber cable.
• Cost: The cost of the GPUs (e.g., ~$250 billion for a 5 GW facility) dwarfs all other expenses. The land and energy infrastructure represent a tiny fraction of the total capital expenditure, reinforcing that hyperscalers are sensitive to power availability, not power cost.

The primary obstacle is not technology or land, but regulation. Environmental laws like NEPA, originally intended for protection, now create years-long delays for clean energy projects, making it easier to build a polluting chemical plant than a solar farm.

The Economic and Sci-Fi Endgame

The rise of AI and cheap solar energy will fundamentally reshape our economy and civilization.

• Batteries Replace the Grid: The grid performs spatial arbitrage (moving power over distance), while batteries perform temporal arbitrage (moving power through time). As batteries become cheaper, it will be more economical to store locally generated solar power than to transport it over expensive, underutilized transmission lines. The average distance an electron travels from generation to consumption will plummet.
• Redefining Economic Value: AI’s true impact may be masked by traditional metrics like GDP. Much like the internet created immense unmeasured consumer surplus, AI will cause massive deflation in the cost of cognitive work. In the long run, the most accurate measure of a civilization’s power may shift from GDP to its total energy consumption.
• The Energy Singularity: The ultimate trajectory is a collapse towards the simplest, most efficient stack for converting energy into thought. The future of intelligence may not be biological but silicon-based. The final form could be a square-meter silicon wafer floating in space—an integrated solar panel and computronium—capable of simulating a human mind, powered directly by the sun. This represents the most efficient method of turning the universe’s thermodynamic gradient into cognition.

• Tags:
• Feature
• Geopolitics
• AI Infrastructure
• Data Centers
• Energy
• Solar Power
• Industrial Policy
• Wright's Law