The US and China are at war to win the AI race. What is at stake? The chance to own and hold super-intelligence, which is the basis for the future economy. What does the future economy hold? The companies building the models right now think it holds a GPU for every human on the planet, a data center for every entrepreneur, and that we are heading into an age of abundance, that the limiting factors are compute and energy. 

The inflection point where an AI agent can go off and do complex tasks for hours without stopping has happened in the models that are now being released. This means the cost of a workforce has just gone down by 100x (assume US salaries).

“AI is overhyped”, said CIOs at a recent European roundtable. Is this true, or is it the case that Model owners are putting majority of their GPU horsepower towards training their models and be the first AI to reach superintelligence, at the short-term expense of the consumer. Is this why the disillusionment by enterprise CIOs? 

Models need 100.000s of GPUs and Terawatts of energy to reach superintelligence and build the next economy. Two handfuls of companies in China and America own these models and are building energy and data centers in those regions to secure their win.

For every other country, the game is about building national compute, so that governments and companies can use these great models to build products and intelligence on their own sovereign data, without training the model on proprietary data and releasing their local intelligence to the world for free. Laws are being written to ensure national data stays within borders, providing a base to use local models on local data, and maintain control over data access. Every piece of data lost to a public model is no longer proprietary, the competitive advantage of that data point is lost.

The costs of building software and thinking have gone down to virtually zero, so the new bottlenecks become ingenuity, energy and compute. Hyperscalers are not building in secondary markets now, as they deploy GW and GPUs at scale to get to super intelligence. Secondary markets have to build their own AI infrastructure. Now.