The huge financial obligations AI companies incur to build data centers could create a powerful incentive to continue scaling, even if significant safety risks emerge. This economic pressure represents a structural tension between commercial imperatives and safety concerns.

Companies wanting to keep sensitive research data on-site are discovering a major infrastructure challenge. Even a small, local data center can double a lab facility's total power consumption, a critical and costly factor that must be planned for well in advance of securing space.

The seemingly obvious solution of building a dedicated, off-grid power plant for a data center is highly risky. If the data center's technology becomes obsolete, the power plant, lacking a connection to the main grid, becomes a worthless "stranded asset" with no other customer to sell its energy to.

Despite staggering announcements for new AI data centers, a primary limiting factor will be the availability of electrical power. The current growth curve of the power infrastructure cannot support all the announced plans, creating a physical bottleneck that will likely lead to project failures and investment "carnage."

Contrary to the common focus on chip manufacturing, the immediate bottleneck for building new AI data centers is energy. Factors like power availability, grid interconnects, and high-voltage equipment are the true constraints, forcing companies to explore solutions like on-site power generation.

While GPUs dominated headlines, the most significant bottleneck in scaling AI data centers was 100-year-old power transformer technology. With lead times stretching over three years and costs surging 150%, connecting new data centers to the grid became the primary constraint on the AI buildout.

The AI boom has created such desperation for power that hyperscalers now prioritize immediate availability ('time to power') above all else. Cost has become a secondary concern, and sustainability, once a key objective, has fallen far lower on the priority list.

The report of XAI's low GPU utilization reveals a critical, non-obvious bottleneck in AI: it's not just about acquiring compute, but using it efficiently. This 'FLOPS utilization' problem, caused by architectural and load-balancing issues, means billions in hardware sits underused, creating an opportunity for companies that can optimize the compute stack.

The urgent need for AI compute capacity is outpacing grid upgrade timelines, which can take 3-5 years. In response, hyperscalers are installing "behind the meter" power solutions—often less-efficient, simple-cycle natural gas generators—as a pragmatic way to get data centers operational years faster than waiting for utility connections.