Data center capital is moving to the margin and the sky

Meta just signed its first data center deal in India, a 168-megawatt facility built with Reliance that can scale over time, according to TechCrunch. The move is less about Bangalore as a tech hub and more about finding power and land where the hyperscale giants have not yet locked up supply. India offers both, and Meta is betting that AI compute demand will outlast the next cycle.

Meanwhile, Affinius Capital closed $975 million in financing for a new Northern Virginia data center, GlobeSt reports. Northern Virginia remains the densest data center market in the world, and nearly a billion dollars for a single facility is a reminder that even in the established corridor, capital is chasing incremental capacity. The financing market for these projects has not frozen; it has professionalized around a thesis that AI workloads are durable.

Then there is Euwyn Poon, who built 250,000 e-scooters at Spin and now wants to put 10,000 data centers in space. His company, Orbital, has raised $5 million to start. The pitch is that space offers free cooling, abundant solar power, and no zoning fights. It sounds like science fiction until you remember that the cost curve for launches has collapsed and that latency-tolerant training workloads do not need to sit next to fiber.

These three deals span the plausible and the speculative, but they share a premise: the supply of power and land for AI computing is becoming the binding constraint. Investors are financing both the proven and the unproven because the alternative—waiting for clarity—means missing the build. The question is not whether demand will materialize. It is where the next megawatt comes from.