The artificial intelligence industry is running headfirst into physical limits that no amount of investment can quickly resolve. ASML CEO Christophe Fouquet — whose company holds a monopoly on the extreme ultraviolet lithography machines required to manufacture modern chips — stated plainly that despite a massive acceleration in chip production, the market will remain supply-constrained for the next two to five years. That means the world's largest hyperscalers — Google, Microsoft, Amazon, and Meta — will not receive all the chips they are paying for. Google Cloud COO Francis deSouza put the scale of unmet demand into perspective: Google Cloud's revenue crossed $20 billion last quarter with 63% growth, while its backlog of committed but undelivered revenue nearly doubled in a single quarter, jumping from $250 billion to $460 billion.But chips are only the first bottleneck. Energy is the one looming right behind it. deSouza revealed that Google is seriously exploring the idea of data centers in space as a response to terrestrial energy constraints, noting that orbit provides access to more abundant energy — though the absence of convection in a vacuum makes heat dissipation a major engineering challenge.

Google's internal answer to the energy problem is deep vertical integration: by co-engineering its TPU chips alongside its Gemini models, the company claims to run AI workloads far more efficiently than any configuration using off-the-shelf components. In a world where energy availability is becoming as critical as compute availability, that kind of end-to-end control over the stack could prove to be one of the most durable competitive advantages in the industry.