The artificial intelligence race is often framed as a competition between models and chips. In reality, it is increasingly determined by who can bring energized capacity online, and how quickly. In the most competitive markets, hyperscalers and neoclouds face multi-year waits for grid interconnection. They compete for a shrinking supply of powered land, placing as much value on time-to-power as on cost. Yet the global energy landscape is less about an absolute shortage than a lack of access. Significant generation remains stranded or underused, including surplus hydroelectric power, curtailed wind and solar energy, and flared gas. Geography, limited transmission infrastructure, or local demand that never materialized often prevent these resources from being fully utilized.
Deniss Katsoka, co-founder and CEO of Baselayer Energy, sees this disconnect as one of the biggest infrastructure opportunities of the current era. Companies that identify these pockets of energy early, secure the surrounding land, and convert them into operational computing capacity will gain an advantage that becomes difficult to replicate once the broader market recognizes their value.
“Compute is going to follow power, not the other way around,” Katsoka says. “For example, in parts of the U.S., the constraint is time-to-power. Large players are competing for a finite number of energized sites, and new interconnections can take years. Other regions, such as the Middle East, and the UAE in particular, present a different equation. Sovereign capital is flowing into AI and digital infrastructure, and there is real room to build compute capacity right next to power supply that already exists.” For developers, understanding these regional differences has become a central part of deployment strategy rather than something to consider later.
Recent industry research highlights the scale of the challenge. Deloitte’s 2026 Power and Utilities Industry Outlook estimates that U.S. peak electricity demand could increase by approximately 26% by 2035, while data centers alone could require up to 176 gigawatts of capacity by that time. Deloitte also notes that substantial amounts of new power generation are still waiting for grid connections. In other words, having energy available and being able to deliver it to a specific site are two distinct challenges. Much of the value being created in AI infrastructure today lies in bridging that gap.
PwC’s September 2026 Global Data Centre Outlook places the long-term opportunity on an even larger scale, projecting approximately $31.6 trillion in global AI infrastructure capital expenditure through 2050. Its analysis identifies power as one of the main factors shaping where that investment will ultimately go, alongside connectivity, security, policy conditions, community consent, and access to computing hardware. In simple terms, the locations where AI infrastructure is built over the coming decades will depend heavily on where reliable power is available.
This perspective also reflects the firsthand experience of Baselayer’s co-founders. Deniss Katsoka and Mattia Agosta met while working at Amazon Web Services, where they gained a close view of how a hyperscale plans, builds, and operates infrastructure on a global scale. They also witnessed the emergence of a critical constraint: advances in chips and software ultimately depended on energy, which was becoming the bottleneck. That realization led them to establish Baselayer.
“At AWS, we saw how much of the progress in AI rests on something most people never think about, which is the power behind it,” Katsoka says. “Chips get replaced every year or two. A grid connection or a long-term energy contract stays valuable for decades. We treat the powered site as the lasting asset, and the computing equipment on it as the part that changes.”
Agosta takes the same approach to the company’s long-term growth. “Hardware and software will look completely different in five or ten years. Land, grid access and power contracts stay useful through every one of those changes,” he says. Baselayer develops its sites to support AI training and inference today while preparing them for increasingly dense and heat-intensive equipment in the future. The latest AI server racks already consume more than 100 kilowatts each. Over time, these same sites could also accommodate other energy-intensive applications, including robotics and advanced manufacturing.
Speed is one area where Baselayer has already distinguished itself. The company progressed from concept to a live, operational 5-megawatt site in less than six months. This achievement came as infrastructure developers compete for the same limited supply of power equipment, with transformers and other essential grid components often taking 12 to 18 months or longer to arrive. With its first site now operational, Baselayer is accelerating its expansion toward a long-term target of multi-gigawatt capacity and in-house nuclear power generation.
However, the operational demands of running a data center remain significant as facilities expand. Heat, dust, cooling water, power quality, equipment availability, maintenance, and relationships with utilities all matter at 5 megawatts. These factors become even more consequential at several hundred megawatts, where the stakes are substantially higher. For this reason, Baselayer is establishing enterprise-grade operational standards from the beginning, ensuring that the systems and processes it builds today can support every stage of future growth.
“A computing facility is a physical operation before it is anything else,” Agosta says. “The economics depend on uptime, on supply chains that deliver, on power staying available, and on a team on the ground that can react when conditions change. We’d rather get that right at 5 megawatts than try to fix it at 500.”
Baselayer’s emphasis on execution reflects a broader transformation across the energy sector, as outlined by Deloitte. Companies are increasingly adopting digital tools, predictive maintenance, improved supply-chain monitoring, and automation to strengthen their physical operations. The report highlights reliability and efficiency as growing priorities as businesses navigate shifting demand, extended equipment lead times, tighter financing conditions, and evolving regulations. These same pressures affect companies seeking to develop computing infrastructure at scale.
Baselayer’s business model also extends beyond simply selling AI computing capacity, and this is intentional. The company combines energy supply relationships across several regions with modular construction, real-time technical monitoring, and cooling improvements. It also operates bitcoin mining at its sites as a flexible computing workload that can help balance the grid.
AI training and inference generally perform best with a consistent power supply. Bitcoin mining, by contrast, can increase when surplus electricity is available and shut down quickly when the grid needs that capacity back. This approach allows a site to generate revenue from its power as soon as it comes online, provide flexibility to local utilities, and gradually redirect capacity toward AI customers as demand increases. The broader market has already demonstrated the potential of this strategy, with several publicly listed bitcoin miners signing multi-year agreements to host AI workloads, largely because they already control sites with established power infrastructure.
Securing control over power and land can also create opportunities beyond computing. By converting underused energy into productive infrastructure, the same sites can accommodate manufacturing, robotics, connectivity, and other energy-intensive industries. They may also support improvements to local power networks that deliver benefits to surrounding communities.
Models, chips, and software will continue to advance at a pace that can make today’s hardware obsolete within just a few years. The physical infrastructure supporting them, however, will evolve much more slowly. This difference lies at the core of Baselayer’s strategy. Chips can be purchased in months, while new grid capacity can take years to develop. Between those timelines is a limited supply of electricity that can be put to work today, distributed across regions in pockets that much of the market has yet to fully value.
The lasting advantage will belong to those who identify these opportunities first, secure the land around them, and build infrastructure designed for the long term. “Chips will keep changing. Power and land won’t,” Katsoka says. “We want to own the part of the AI stack that lasts.”
