The AI Race the World Doesn’t Anticipate

The AI race will not be won with a better model. It may be won somewhere no one is watching, by a country whose only advantage is one nobody’s counting yet. Andy Weir wrote a fictional parallel years ago that is worth revisiting.

In Andy Weir’s novel, Artemis, he writes about a near future where countries with economic advantages compete in the space race. Weir doesn’t pick the countries you expect. Rather countries that are simply closest to the moon that can ferry thousands of space bound trips at a slightly less price. It is about economies of scale.

Weir is known for making the plausible believable, maybe even forecasting the future from his other novels, The Martian, and Project Hail Mary. What Artemis tells us about hyper scaling space travel is a good parallel to hyper scaling AI. The economic winners of the future will come from unexpected regions of the world, not those with the best AI models, or the fastest chips, but rather those with the best availability of electricity.

The Race Everyone Can See

At the moment, the focus of the AI race is building the most advanced models, navigating the uncertain regulation, creating faster NVIDIA GPUs, and seemingly boundless investments in building data centers. It is the right place to be for a (still) emerging technology. We need to keep reminding ourselves that Generative AI is only a few years old, and as such the environment it lives in needs upgrading.

This includes training data and data centers alike. Most of the latter are a holdover from the 1990’s era dot-com boom. The infrastructure of that built us faster internet and thousands of miles of fiber optic still serve as the backbone of GenAI. The prompt-to-inference-to-response loop works well as a result, but it also needs more compute power, notably in those places where all that fiber exists, like Northern Virginia.

Where The Power Is

While Virginia has the most data centers, places like Texas and California are also competing. California for its natural association with Silicon Valley, and Texas which offers an abundance of land and economic incentives for electricity. The Electric Reliability Council of Texas (ERCOT) manages the independent power grid of Texas and can set its own pricing/dispatch decisions without federal grid oversight. Proponents of these strategies point to the jobs and revenue that come from building data centers. But what happens after that is completed?

AI as a Commodity

AI capability is following its own version of Moore’s law. However, unlike computing power that relies on the complexity of a physical computer chip, AI will likely plateau in its capabilities. The AI race will end, or at least slow down, where there will not be one or two leaders, but rather a group of providers with similar capabilities. AI will mostly be a commodity. And when that happens how will each platform compete?

You might not need to look farther than the labor arbitrage of the past couple decades to answer the question. Economics always offers efficient solutions. The global workforce shifted when “knowledge” work could be moved from more expensive locations to less expensive ones. EU countries shifted jobs to eastern Europe, the United States shifted to India, Chile, and the Philippines. Each recipient region continued to invest in their data infrastructures and educational institutions to provide a value-added reason for hosting the work. Now imagine the same for AI and electricity.

The Real Competition Is Not the Model

AI platforms like Gemini (Google), Copilot (Microsoft), ChatGPT (OpenAI), and Claude (Anthropic) are all betting on revenue from corporate providers. Running the companies that run the world using AI is the sustainable objective of the AI race. They are currently operating at a large scale, OpenAI reports processing 2.5 billion prompts daily, and it is expected to grow over the next decade. When in the future these AI platforms and a host of others can essentially provide the same GenAI functionality, then the remaining frontier is pricing and the variable cost of electricity, and Texas may not be the only one offering incentives.

Kenya. This was the country that Weir wrote about that had the best advantage to launch high scale payloads into space. It was due to its location on the equator where it could take advantage of distance and the Earth’s rotational speed. It was an unlikely choice given the space race leaders in the U.S., Russia, and China that predated his 2080 story setting. In the novel Weir writes about how it changed Kenya’s geopolitical importance.

While Kenya may not be the electricity provider for the future of AI, it may come from other regions with better resources and regulations that are favorable. France is a leader in nuclear powered electricity, generating 70% of its needs from this source. A well-established industry that could scale with fewer regulatory or community barriers. China also has a robust nuclear electric capacity.

Venture into hydro-electricity and you will note that China, Brazil, Russia, Norway, and Canada are leading harnessing power from water. Canada alone has more lakes than the rest of the world combined and stores roughly 20% of the world’s freshwater.

---

The race for AI is consuming hundreds of millions of dollars to build the frontier models and the needed data centers. At the moment this is occurring in a handful of places in the U.S. and across the globe, hosted by nations with abundant capital and powered by the best intellectual talent. However, a decade from now AI has the potential, like Kenya in Artemis, to shift the geopolitical balance to other regions – those that can compete in a new way with electricity. Nations that recognize this opportunity now to be part of the AI supply chain will have a sustained advantage.