Dear Readers,

The global race to build AGI has quietly become a construction project. While the world debates which model will achieve superintelligence first, the real contest is being decided by concrete pourers in Abilene, Texas, grid engineers in Guangdong, and a lone exascale supercomputer humming in Jülich, Germany.

Today's deep dive unpacks the numbers behind the AI infrastructure explosion: why five American companies control 71% of global AI compute, how China is building an entire parallel chip ecosystem under sanctions pressure, and why the biggest bottleneck is no longer semiconductors but something far more mundane, electricity. We dig into OpenAI's surprisingly candid admission that AI wealth won't trickle down without deliberate intervention, and in Chubby's Opinion Corner, we ask the uncomfortable question European leaders keep dodging: can a continent that celebrates a 180 million euro cloud contract compete with companies spending 135 billion dollars in a single year, and if so, what would that actually look like?

Grab your coffee, this one is dense, but if you care about who will control the computational foundation of the global economy, it might be the most important issue we've published all year.

All the best,

Kim Isenberg

Somewhere in Abilene, Texas, construction crews are pouring concrete for a data center that will eventually draw 590 megawatts of electricity, enough to light up a small city. Across the Pacific, Chinese provinces are racing to build what Beijing calls "intelligent computing centers," fueled by domestically designed chips and state coordination. Meanwhile, in Jülich, Germany, Europe's first exascale supercomputer just came online, a potent but lonely signal from a continent still searching for its role in the AI infrastructure game. The race to build artificial general intelligence is no longer primarily a contest of algorithms and research papers. It has become, above all, a race for electricity, chips, cooling systems, and real estate.

This shift matters far beyond the tech industry. The ability to train and run the next generation of AI models depends on physical infrastructure that takes years to build, billions of dollars to finance, and enormous amounts of energy to operate. Whoever controls that infrastructure, not just the models running on it, will shape the economic and geopolitical order of the coming decade. The question this raises is straightforward but far from simple: In a world where AI wealth depends on data centers, who actually controls the compute, where are the bottlenecks, and can the economic gains this infrastructure generates ever be fairly distributed?